#Instalamos libreria unidecode para la conversión de textos con tilde a sin tilde

!pip install unidecode

Collecting unidecode
  Downloading Unidecode-1.4.0-py3-none-any.whl.metadata (13 kB)
Downloading Unidecode-1.4.0-py3-none-any.whl (235 kB)
   ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 0.0/235.8 kB ? eta -:--:--
   ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╸ 235.5/235.8 kB 13.9 MB/s eta 0:00:01
   ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 235.8/235.8 kB 5.3 MB/s eta 0:00:00
Installing collected packages: unidecode
Successfully installed unidecode-1.4.0

# Instalamos herramienta nbconvert de Jupyter que sirve para convertir notebooks
# en otros formatos como HTML, PDF o Markdown.

!pip install -q "nbconvert>=7.0.0"

# Importamos las librerías necesarias para el análisis exploratorio de datos (AED)

# pandas: manipulación y análisis de datos en estructuras tipo DataFrame
import pandas as pd

# numpy: operaciones matemáticas y funciones estadísticas de bajo nivel
import numpy as np

# matplotlib.pyplot: librería de visualización básica para gráficos
import matplotlib.pyplot as plt

# seaborn: librería de visualización basada en matplotlib, más orientada a análisis estadístico
import seaborn as sns

# unidecode: librería de conversión de texto a código ASCII (sin tildes ni ñ ni caracteres especiales)
from unidecode import unidecode

# os: es un módulo de Python que permite interactuar con el sistema operativo
# (crear carpetas, gestionar rutas, archivos, procesos)
import os

# Re: librería estándar de python que nos permite buscar, extraer y transformar cadenas de texto mediante expresiones regulares.
import re

# warnings: Libreria estándar de Python que permite gestionar y controlar los mensajes de advertencia que aparecen durante la ejecución del código
import warnings
warnings.filterwarnings("ignore")

# statsmodels: libreria de modelización estadística especializada en series temporales y econometría.
# Incluye implementaciones clásicas como ARIMA, SARIMA y Holt-Winters (ETS).
import statsmodels.api as sm

# datetime:
from datetime import datetime

# Json:
import json

import sys

import glob

# Intentamos importar scipy.sparse; si no está, seguimos con numpy
try:
    from scipy import sparse as sp
    SCIPY_AVAILABLE = True
except Exception:
    SCIPY_AVAILABLE = False

import matplotlib.image as mpimg

import requests

import hashlib, shutil, math, textwrap

# statsmodels para ExponentialSmoothing (Holt-Winters)
from statsmodels.tsa.holtwinters import ExponentialSmoothing
from sklearn.metrics import mean_squared_error

# Importamos STL para descomponer la serie en tendencia + estacionalidad + residuo
from statsmodels.tsa.seasonal import STL

from typing import Dict, Any, List, Tuple, Optional

# statsmodels para ETS/SARIMA ===

from statsmodels.tsa.holtwinters import ExponentialSmoothing

from statsmodels.tsa.stattools import acf

from statsmodels.tsa.holtwinters import SimpleExpSmoothing, ExponentialSmoothing
from statsmodels.tsa.arima.model import ARIMA

from statsmodels.tsa.statespace.sarimax import SARIMAX
from statsmodels.tools.sm_exceptions import ConvergenceWarning

# Silenciamos advertencias ruidosas de statsmodels
warnings.filterwarnings("ignore", category=ConvergenceWarning)
warnings.filterwarnings("ignore", message="Non-invertible.*")
warnings.filterwarnings("ignore", message="Non-stationary.*")


from sklearn.metrics import mean_absolute_error

from scipy.stats import median_abs_deviation, mannwhitneyu

# =========================
# Bloque 0) Imports
# =========================
# Cargamos librerías estándar y de modelización. Hacemos importes condicionales para ser robustos.


# Modelos statsmodels (ETS/Holt/Theta/ARIMA)
try:
    from statsmodels.tsa.holtwinters import ExponentialSmoothing
except Exception:
    ExponentialSmoothing = None

try:
    from statsmodels.tsa.statespace.sarimax import SARIMAX
except Exception:
    SARIMAX = None

try:
    from statsmodels.tsa.forecasting.theta import ThetaModel
except Exception:
    ThetaModel = None

    # Importamos statsmodels si está disponible; si no, caeremos a baselines
try:
    from statsmodels.tsa.holtwinters import ExponentialSmoothing
except Exception:
    ExponentialSmoothing = None

try:
    from statsmodels.tsa.statespace.sarimax import SARIMAX
except Exception:
    SARIMAX = None

try:
    from statsmodels.tsa.forecasting.theta import ThetaModel
except Exception:
    ThetaModel = None

from pandas.api.types import is_datetime64tz_dtype

from pathlib import Path


from __future__ import annotations



# Intentamos importar statsmodels; si no está, dejamos None y haremos fallbacks.
try:
    from statsmodels.tsa.holtwinters import ExponentialSmoothing
except Exception:
    ExponentialSmoothing = None

try:
    from statsmodels.tsa.statespace.sarimax import SARIMAX
except Exception:
    SARIMAX = None

try:
    from statsmodels.tsa.forecasting.theta import ThetaModel
except Exception:
    ThetaModel = None

# Montamos Google Drive para acceder al archivo desde Colab
from google.colab import drive
drive.mount('/content/drive')

Drive already mounted at /content/drive; to attempt to forcibly remount, call drive.mount("/content/drive", force_remount=True).

# Definimos la ruta al archivo Excel en tu Google Drive
ruta_excel = "/content/drive/MyDrive/PROYECTO_NEITH/FUENTE_DATOS/HECHOS/FAMA_Encriptado_21-25.xlsx"

# Nombre de la hoja que contiene los datos
hoja = "Desref_Deta_encri_fin_valo21-25"

# Leemos el archivo Excel en un DataFrame de pandas
df_fd1_full = pd.read_excel(ruta_excel, sheet_name=hoja, engine="openpyxl")

# Mostramos las primeras filas para comprobar la carga
df_fd1_full.head()

# Número de filas y columnas
print("Dimensiones del DataFrame:", df_fd1_full.shape)

# Tipos de datos por columna
print("\nTipos de datos:")
print(df_fd1_full.dtypes)

# Valores únicos en columnas clave para revisar codificación
for col in ["COST_TYPE", "SUPPLIER_TYPE", "COUNTRY", "FM_COST_TYPE"]:
    print(f"\nValores únicos en {col}:")
    print(df_fd1_full[col].unique())

Dimensiones del DataFrame: (324439, 15)

Tipos de datos:
ID_ORDER            object
COST_TYPE           object
COUNTRY             object
ID_BUILDING          int64
YEAR_MONTH           int64
FM_RESPONSIBLE      object
FM_COST_TYPE        object
COST                object
ID_CUSTOMER          int64
ID_SUPPLIER          int64
ID_BUSINESS_UNIT     int64
SUPPLIER_TYPE       object
MONTH                int64
YEAR                 int64
QUARTER              int64
dtype: object

Valores únicos en COST_TYPE:
['Gasto' 'Inversión']

Valores únicos en SUPPLIER_TYPE:
['EXTERNO' 'INTERNO' 0]

Valores únicos en COUNTRY:
['España' 'México' 'Costa Rica' 'Panamá' 'Perú' 'República Dominicana'
 'Colombia' 'Italia' 'Marruecos']

Valores únicos en FM_COST_TYPE:
['Mtto. Contratos' 'Servicios Ctto.' 'Suministros' 'Eficiencia Energética'
 'Servicios Extra' 'Obras' 'Mtto. Correctivo' 'Licencias'
 'Control y Reporting' 'Dirección FM' 'Oficina Técnica' 'Real Estate'
 'Vacío' 'Global' 'Operaciones']

# Creamos una nueva columna 'cost_float' a partir de COST
# Si algún valor no se puede convertir, quedará como NaN

df_fd1_full["cost_float"] = pd.to_numeric(df_fd1_full["COST"], errors="coerce")

# Comprobamos el tipo de dato de la nueva columna
print(df_fd1_full["cost_float"].dtype)

# Contamos cuántos valores quedaron como NaN
print("Valores NaN en cost_float:", df_fd1_full["cost_float"].isna().sum())

# Mostramos algunas filas para verificar
df_fd1_full[["COST", "cost_float"]].head()

float64
Valores NaN en cost_float: 47

# Filtramos los registros donde cost_float es NaN
df_cost_nan = df_fd1_full[df_fd1_full["cost_float"].isna()]

# Mostramos el número de registros y algunas columnas clave para revisar
print("Número de registros con cost_float NaN:", len(df_cost_nan))

# Seleccionamos columnas útiles para revisar qué ocurre
cols_revision = ["ID_ORDER", "COST", "COST_TYPE", "COUNTRY", "YEAR_MONTH", "FM_COST_TYPE"]
df_cost_nan[cols_revision].head(47)

Número de registros con cost_float NaN: 47

# Filtramos los registros donde SUPPLIER_TYPE = 0
df_supplier_zero = df_fd1_full[df_fd1_full["SUPPLIER_TYPE"] == 0]

# Mostramos el número de registros
print("Número de registros con SUPPLIER_TYPE = 0:", len(df_supplier_zero))

# Vemos algunas columnas relevantes para entender el contexto
cols_revision = ["ID_ORDER", "SUPPLIER_TYPE", "ID_SUPPLIER", "COST", "cost_float", "COST_TYPE", "COUNTRY", "YEAR", "MONTH"]
df_supplier_zero[cols_revision].head(20)

Número de registros con SUPPLIER_TYPE = 0: 8

# Definimos un diccionario con las correcciones manuales de SUPPLIER_TYPE
correcciones_supplier_type = {
    3006129: "EXTERNO",
    3007232: "EXTERNO"
}

# Creamos una nueva columna SUPPLIER_TYPE_MOD con la corrección aplicada
df_fd1_full["SUPPLIER_TYPE_MOD"] = df_fd1_full.apply(
    lambda row: correcciones_supplier_type.get(row["ID_SUPPLIER"], row["SUPPLIER_TYPE"]),
    axis=1
)

# Comprobamos los valores únicos después de la corrección
print("Valores únicos en SUPPLIER_TYPE_MOD:")
print(df_fd1_full["SUPPLIER_TYPE_MOD"].unique())

# Revisamos los registros corregidos
df_fd1_full[df_fd1_full["ID_SUPPLIER"].isin(correcciones_supplier_type.keys())][
    ["ID_ORDER", "ID_SUPPLIER", "SUPPLIER_TYPE", "SUPPLIER_TYPE_MOD"]
]

Valores únicos en SUPPLIER_TYPE_MOD:
['EXTERNO' 'INTERNO']

# Contamos cuántos registros tienen "Vacío" en FM_COST_TYPE
n_vacio = (df_fd1_full["FM_COST_TYPE"] == "Vacío").sum()
print("Número de registros con FM_COST_TYPE = 'Vacío':", n_vacio)

# Vemos algunas filas para revisar en qué contexto aparece
cols_revision = ["ID_ORDER", "COST_TYPE", "COUNTRY", "YEAR", "MONTH", "FM_RESPONSIBLE", "FM_COST_TYPE", "cost_float"]
df_fd1_full[df_fd1_full["FM_COST_TYPE"] == "Vacío"][cols_revision].head(20)

Número de registros con FM_COST_TYPE = 'Vacío': 17

# Creamos un nuevo dataset eliminando los registros con FM_COST_TYPE = "Vacío"
df_fd1_v1 = df_fd1_full[df_fd1_full["FM_COST_TYPE"] != "Vacío"].copy()

# Comprobamos el número de registros antes y después
print("Filas en df_fd1_full:", df_fd1_full.shape[0])
print("Filas en df_fd1_v1:", df_fd1_v1.shape[0])

# Validamos que ya no queden registros con FM_COST_TYPE = "Vacío"
print("Registros con FM_COST_TYPE = 'Vacío' en df_fd1_v1:",
      (df_fd1_v1["FM_COST_TYPE"] == "Vacío").sum())

Filas en df_fd1_full: 324439
Filas en df_fd1_v1: 324422
Registros con FM_COST_TYPE = 'Vacío' en df_fd1_v1: 0

# Valores únicos en FM_RESPONSIBLE
valores_unicos_resp = df_fd1_v1["FM_RESPONSIBLE"].unique()
print("Valores únicos en FM_RESPONSIBLE:", valores_unicos_resp)

# Conteo de registros por clase de FM_RESPONSIBLE
conteo_resp = df_fd1_v1["FM_RESPONSIBLE"].value_counts().sort_values(ascending=False)
print("\nDistribución de registros por clase:")
print(conteo_resp)

Valores únicos en FM_RESPONSIBLE: ['Mantenimiento Multipunto' 'Mantenimiento' 'Eficiencia Energética'
 'Licencias' 'Dirección FM' 'Obras Proyectos' 'Oficina Técnica'
 'Gestión Espacios' 'Control y Reporting' 'Real Estate' 'Global'
 'Operaciones']

Distribución de registros por clase:
FM_RESPONSIBLE
Mantenimiento               247288
Eficiencia Energética        48142
Mantenimiento Multipunto     12729
Obras Proyectos               7975
Licencias                     5951
Gestión Espacios              2091
Oficina Técnica                140
Dirección FM                    64
Global                          22
Real Estate                     10
Control y Reporting              6
Operaciones                      4
Name: count, dtype: int64

# Identificamos las clases minoritarias (<2000 registros)
conteo_resp = df_fd1_v1["FM_RESPONSIBLE"].value_counts()
clases_minoritarias = conteo_resp[conteo_resp < 2000].index.tolist()

print("Clases minoritarias (<2000 registros):", clases_minoritarias)

# Filtramos solo esas clases
df_minoritarias = df_fd1_v1[df_fd1_v1["FM_RESPONSIBLE"].isin(clases_minoritarias)].copy()

# Tabla pivoteada FM_RESPONSIBLE x YEAR
pivot_responsable_year = (
    df_minoritarias
    .groupby(["FM_RESPONSIBLE", "YEAR"])["cost_float"]
    .sum()
    .reset_index()
    .pivot(index="FM_RESPONSIBLE", columns="YEAR", values="cost_float")
    .fillna(0)
)

print("\nCoste total por FM_RESPONSIBLE (clases minoritarias) y Año:")
display(pivot_responsable_year)

# Detalle incluyendo ID_BUILDING
coste_minoritarias_building = (
    df_minoritarias
    .groupby(["FM_RESPONSIBLE", "YEAR", "ID_BUILDING"])["cost_float"]
    .sum()
    .reset_index()
    .sort_values(by=["FM_RESPONSIBLE", "YEAR", "cost_float"], ascending=[True, True, False])
)

print("\nDetalle coste por FM_RESPONSIBLE, Año e ID_BUILDING:")
coste_minoritarias_building

Clases minoritarias (<2000 registros): ['Oficina Técnica', 'Dirección FM', 'Global', 'Real Estate', 'Control y Reporting', 'Operaciones']

Coste total por FM_RESPONSIBLE (clases minoritarias) y Año:

Detalle coste por FM_RESPONSIBLE, Año e ID_BUILDING:

# Tipos de datos del dataset modificado
print("Tipos de datos en df_fd1_v1:")
print(df_fd1_v1.dtypes)

# Número de valores únicos por columna
print("\nNúmero de valores únicos por columna:")
valores_unicos = df_fd1_v1.nunique().sort_values(ascending=False)
print(valores_unicos)

Tipos de datos en df_fd1_v1:
ID_ORDER              object
COST_TYPE             object
COUNTRY               object
ID_BUILDING            int64
YEAR_MONTH             int64
FM_RESPONSIBLE        object
FM_COST_TYPE          object
COST                  object
ID_CUSTOMER            int64
ID_SUPPLIER            int64
ID_BUSINESS_UNIT       int64
SUPPLIER_TYPE         object
MONTH                  int64
YEAR                   int64
QUARTER                int64
cost_float           float64
SUPPLIER_TYPE_MOD     object
dtype: object

Número de valores únicos por columna:
ID_ORDER             324421
COST                  88837
cost_float            88836
ID_SUPPLIER            2594
ID_BUILDING             722
ID_CUSTOMER             181
YEAR_MONTH               60
FM_COST_TYPE             14
ID_BUSINESS_UNIT         14
FM_RESPONSIBLE           12
MONTH                    12
COUNTRY                   9
YEAR                      5
QUARTER                   4
SUPPLIER_TYPE             3
COST_TYPE                 2
SUPPLIER_TYPE_MOD         2
dtype: int64

# Tabla de clasificación de variables con tipo propuesto y justificación resumida

# Obtenemos tipos y número de únicos desde el dataset
tipo_variables = df_fd1_v1.dtypes.astype(str).rename("tipo_variables")
valores_unicos = df_fd1_v1.nunique().rename("valores_unicos")

# Definimos el rol de cada variable
rol_propuesto = {
    "ID_ORDER": "identificador",
    "COST_TYPE": "categorica",
    "COUNTRY": "categorica",
    "ID_BUILDING": "identificador",
    "YEAR_MONTH": "temporal",
    "FM_RESPONSIBLE": "categorica",
    "FM_COST_TYPE": "categorica",
    "COST": "texto_original",
    "ID_CUSTOMER": "identificador",
    "ID_SUPPLIER": "identificador",
    "ID_BUSINESS_UNIT": "identificador",
    "SUPPLIER_TYPE": "categorica_ref",
    "MONTH": "temporal",
    "YEAR": "temporal",
    "QUARTER": "temporal",
    "cost_float": "numerica_continua",
    "SUPPLIER_TYPE_MOD": "categorica"
}

# Justificación breve por variable
justificacion = {
    "ID_ORDER": "Clave técnica única por fila; no aporta patrones generalizables.",
    "COST_TYPE": "Etiqueta OPEX/CAPEX (Gasto/Inversión); categórica de alto valor explicativo.",
    "COUNTRY": "Diferencias regulatorias/mercado; categórica geográfica.",
    "ID_BUILDING": "Identificador del espacio; el código en sí no es informativo.",
    "YEAR_MONTH": "Periodo YYYYMM; útil para agregaciones temporales si se requiere.",
    "FM_RESPONSIBLE": "Vertical responsable; categórica potencialmente correlacionada con la familia de gasto.",
    "FM_COST_TYPE": "Familia del gasto (contrato/bajo demanda, etc.); categórica clave.",
    "COST": "Valor original de coste en texto; se mantiene para trazabilidad.",
    "ID_CUSTOMER": "Clave de sociedad; mejor usar rasgos derivados que el código.",
    "ID_SUPPLIER": "Clave de proveedor; preferible usar atributos derivados (tipo, histórico).",
    "ID_BUSINESS_UNIT": "Clave de división; el número no tiene significado intrínseco.",
    "SUPPLIER_TYPE": "Categoría original con codificación inconsistente; queda como referencia.",
    "MONTH": "Mes numérico; temporal para agregaciones.",
    "YEAR": "Año; temporal para cortes y tendencias.",
    "QUARTER": "Trimestre; temporal para agregaciones trimestrales.",
    "cost_float": "Coste en euros listo para análisis; variable numérica continua.",
    "SUPPLIER_TYPE_MOD": "Tipo de proveedor corregido; usar esta en lugar de la original."
}

# Construimos el dataframe de esquema
cols_presentes = [c for c in df_fd1_v1.columns]  # preserva el orden actual del dataset
esquema_variables = (
    pd.DataFrame(index=cols_presentes)
      .join(tipo_variables, how="left")
      .join(valores_unicos, how="left")
      .reset_index()
      .rename(columns={"index": "variable"})
)

# Añadimos rol y justificación
esquema_variables["tipo_variable"] = esquema_variables["variable"].map(rol_propuesto).fillna("por_definir")
esquema_variables["justificacion"] = esquema_variables["variable"].map(justificacion).fillna("Pendiente de definición.")

# Mostramos la tabla
esquema_variables

# Ordenamos la tabla por tipo_propuesto y después por nombre de variable
esquema_variables_ordenado = esquema_variables.sort_values(
    by=["tipo_variable", "variable"]
).reset_index(drop=True)

# Mostramos la tabla ordenada
esquema_variables_ordenado

# Resumen estadístico de las variables numéricas del dataset
df_fd1_v1.describe().T

# Filtramos los registros con ID_BUSINESS_UNIT = 0
df_bu_cero = df_fd1_v1[df_fd1_v1["ID_BUSINESS_UNIT"] == 0]

# Número de registros con valor 0
print("Número de registros con ID_BUSINESS_UNIT = 0:", len(df_bu_cero))

# Proporción sobre el total
print("Proporción sobre el total:", round(len(df_bu_cero) / len(df_fd1_v1) * 100, 4), "%")

# Revisamos algunas columnas clave de esos registros (incluyendo ID_BUILDING)
cols_revision = [
    "ID_ORDER", "COST_TYPE", "COUNTRY", "YEAR", "MONTH",
    "FM_RESPONSIBLE", "FM_COST_TYPE", "ID_BUILDING", "cost_float"
]
df_bu_cero[cols_revision].head(23)

Número de registros con ID_BUSINESS_UNIT = 0: 23
Proporción sobre el total: 0.0071 %

# Diccionario de corrección: ID_BUILDING -> ID_BUSINESS_UNIT correcto
correccion_bu_por_building = {
    1001191: 43,
    1000496: 1000021,
    1065: 50,
    1000389: 43,
    1270: 43,
    1201: 43,
}

# Nueva columna con corrección aplicada solo si ID_BUSINESS_UNIT == 0
df_fd1_v1["ID_BUSINESS_UNIT_MOD"] = np.where(
    (df_fd1_v1["ID_BUSINESS_UNIT"] == 0) & (df_fd1_v1["ID_BUILDING"].isin(correccion_bu_por_building.keys())),
    df_fd1_v1["ID_BUILDING"].map(correccion_bu_por_building),
    df_fd1_v1["ID_BUSINESS_UNIT"]
).astype("int64")   # usamos int64 estándar

# Flag para auditar qué filas fueron corregidas
df_fd1_v1["bu_corr_flag"] = (
    (df_fd1_v1["ID_BUSINESS_UNIT"] == 0)
    & (df_fd1_v1["ID_BUILDING"].isin(correccion_bu_por_building.keys()))
)

# Resumen
print("Filas corregidas:", df_fd1_v1["bu_corr_flag"].sum())
print("Filas con ID_BUSINESS_UNIT_MOD = 0 (después):", (df_fd1_v1["ID_BUSINESS_UNIT_MOD"] == 0).sum())

# Vista de verificación
cols_view = ["ID_ORDER", "ID_BUILDING", "ID_BUSINESS_UNIT", "ID_BUSINESS_UNIT_MOD", "COUNTRY", "YEAR", "MONTH"]
df_fd1_v1.loc[df_fd1_v1["bu_corr_flag"], cols_view].head(20)

Filas corregidas: 23
Filas con ID_BUSINESS_UNIT_MOD = 0 (después): 0

# Valores únicos de ID_BUSINESS_UNIT y su frecuencia
conteo_bu = df_fd1_v1["ID_BUSINESS_UNIT_MOD"].value_counts().sort_index()
print("Valores únicos de ID_BUSINESS_UNIT_MOD en el dataset:\n")
conteo_bu

Valores únicos de ID_BUSINESS_UNIT_MOD en el dataset:

# Lista de unidades de negocio obsoletas
unidades_obsoletas = [42, 1000000, 1000043, 1000109]

# Filtramos los registros con esas unidades de negocio
df_obsoletas = df_fd1_v1[df_fd1_v1["ID_BUSINESS_UNIT_MOD"].isin(unidades_obsoletas)]

# Identificamos los ID_BUILDING únicos afectados
buildings_obsoletos = df_obsoletas["ID_BUILDING"].unique()

print("Número de registros con unidades obsoletas:", len(df_obsoletas))
print("Número de ID_BUILDING distintos afectados:", len(buildings_obsoletos))
print("\nListado de ID_BUILDING con unidades obsoletas:")
print(buildings_obsoletos)

# Resumen por unidad de negocio obsoleta y building
resumen_bu_building = (
    df_obsoletas.groupby(["ID_BUSINESS_UNIT_MOD", "ID_BUILDING"])["ID_ORDER"]
    .count()
    .reset_index()
    .rename(columns={"ID_ORDER": "n_registros"})
    .sort_values(by=["ID_BUSINESS_UNIT_MOD", "n_registros"], ascending=[True, False])
)

print("\nResumen por ID_BUSINESS_UNIT_MOD e ID_BUILDING:")
print(resumen_bu_building)

Número de registros con unidades obsoletas: 182
Número de ID_BUILDING distintos afectados: 5

Listado de ID_BUILDING con unidades obsoletas:
[1001203      18       9    1065     788]

Resumen por ID_BUSINESS_UNIT_MOD e ID_BUILDING:
   ID_BUSINESS_UNIT_MOD  ID_BUILDING  n_registros
1                    42           18          134
0                    42            9           23
2                    42         1065           13
3               1000000           18            3
5               1000043          788            4
4               1000043            9            1
6               1000109      1001203            4

# Definimos lista de ID_BUSINESS_UNIT obsoletas
unidades_obsoletas = [42, 1000000, 1000043, 1000109]

# Buildings afectados
buildings_obsoletos = (
    df_fd1_v1.loc[df_fd1_v1["ID_BUSINESS_UNIT_MOD"].isin(unidades_obsoletas), "ID_BUILDING"]
    .unique()
)

print("Buildings afectados:", buildings_obsoletos)

Buildings afectados: [1001203      18       9    1065     788]

# Para cada building afectado, contamos las unidades de negocio no obsoletas y elegimos la mayoritaria

# Filtramos solo filas de esos buildings y excluimos unidades obsoletas para el cómputo de mayoría
df_para_mayoria = df_fd1_v1[
    (df_fd1_v1["ID_BUILDING"].isin(buildings_obsoletos)) &
    (~df_fd1_v1["ID_BUSINESS_UNIT_MOD"].isin(unidades_obsoletas))
].copy()

# Contamos por building y unidad
conteo_bu_por_building = (
    df_para_mayoria.groupby(["ID_BUILDING", "ID_BUSINESS_UNIT_MOD"])["ID_ORDER"]
    .count()
    .reset_index(name="n")
)

# Elegimos la unidad mayoritaria por building
mayoria_por_building = (
    conteo_bu_por_building.sort_values(["ID_BUILDING", "n"], ascending=[True, False])
    .drop_duplicates(subset=["ID_BUILDING"])
    .rename(columns={"ID_BUSINESS_UNIT_MOD": "ID_BUSINESS_UNIT_MAYORITARIA"})
    .loc[:, ["ID_BUILDING", "ID_BUSINESS_UNIT_MAYORITARIA", "n"]]
)

print("Mayoría por building (excluyendo obsoletas):")
print(mayoria_por_building)

Mayoría por building (excluyendo obsoletas):
    ID_BUILDING  ID_BUSINESS_UNIT_MAYORITARIA     n
1             9                            50  2812
6            18                            50  4094
9           788                            50    25
11         1065                            50  1990
14      1001203                            51   298

# Partimos del diccionario manual ya definido anteriormente
# correccion_bu_por_building = { ... }  # ya creado para el caso de los 0.

# Definimos diccionario de mayorías por building
# mayoria_por_building contiene columnas: ID_BUILDING, ID_BUSINESS_UNIT_MAYORITARIA
map_mayoria = dict(zip(
    mayoria_por_building["ID_BUILDING"],
    mayoria_por_building["ID_BUSINESS_UNIT_MAYORITARIA"]
))

# Combinamos ambos en un único diccionario
#    Priorizamos las correcciones manuales: si un building está en ambos,
#    prevalece 'correccion_bu_por_building'.
mapa_unico_building_bu = {**map_mayoria, **correccion_bu_por_building}

# Aplicamos solo a filas con unidad obsoleta y con building presente en el mapa
unidades_obsoletas = [42, 1000000, 1000043, 1000109]

mascara_corregibles = (
    df_fd1_v1["ID_BUSINESS_UNIT_MOD"].isin(unidades_obsoletas) &
    df_fd1_v1["ID_BUILDING"].isin(mapa_unico_building_bu.keys())
)

print("Filas a corregir con mapa unificado:", mascara_corregibles.sum())

# Sustitución en ID_BUSINESS_UNIT_MOD usando el mapa unificado
df_fd1_v1.loc[mascara_corregibles, "ID_BUSINESS_UNIT_MOD"] = (
    df_fd1_v1.loc[mascara_corregibles, "ID_BUILDING"].map(mapa_unico_building_bu).astype("int64")
)

# Auditamos si quedan unidades obsoletas en los buildings afectados
resto_obsoletas = df_fd1_v1[
    df_fd1_v1["ID_BUSINESS_UNIT_MOD"].isin(unidades_obsoletas) &
    df_fd1_v1["ID_BUILDING"].isin(mapa_unico_building_bu.keys())
]
print("Filas que siguen obsoletas tras la corrección:", len(resto_obsoletas))

# Hacemos un resumen final en los buildings afectados
resumen_final = (
    df_fd1_v1[df_fd1_v1["ID_BUILDING"].isin(mapa_unico_building_bu.keys())]
      .groupby(["ID_BUILDING", "ID_BUSINESS_UNIT_MOD"])["ID_ORDER"]
      .count()
      .reset_index(name="n")
      .sort_values(["ID_BUILDING", "n"], ascending=[True, False])
)
resumen_final.head(50)

Filas a corregir con mapa unificado: 182
Filas que siguen obsoletas tras la corrección: 0

# Revisión de valores únicos y contraste entre ID_BUSINESS_UNIT y ID_BUSINESS_UNIT_MOD

# Contamos los únicos y los ordenamos
orig_unicos = sorted(df_fd1_v1["ID_BUSINESS_UNIT"].unique().tolist())
mod_unicos  = sorted(df_fd1_v1["ID_BUSINESS_UNIT_MOD"].unique().tolist())

print("Número de únicos (ORIGINAL):", len(orig_unicos))
print("Número de únicos (MOD):", len(mod_unicos))
print("\nPrimeros 30 únicos ORIGINAL:", orig_unicos[:30])
print("\nPrimeros 30 únicos MOD:", mod_unicos[:30])

# Diferenciamos los conjuntos
nuevos_en_mod = sorted(list(set(mod_unicos) - set(orig_unicos)))
desaparecen_en_mod = sorted(list(set(orig_unicos) - set(mod_unicos)))

print("\nValores nuevos que APARECEN solo en MOD:", nuevos_en_mod)
print("Valores que DESAPARECEN en MOD (estaban en ORIGINAL):", desaparecen_en_mod)

# Añadimos frecuencias por valor
vc_orig = df_fd1_v1["ID_BUSINESS_UNIT"].value_counts().sort_index()
vc_mod  = df_fd1_v1["ID_BUSINESS_UNIT_MOD"].value_counts().sort_index()

print("\nFrecuencias ORIGINAL (primeros 20 por código):")
print(vc_orig.head(20))
print("\nFrecuencias MOD (primeros 20 por código):")
print(vc_mod.head(20))

# Identificamos las filas que cambiaron de valor
mask_changed = df_fd1_v1["ID_BUSINESS_UNIT"] != df_fd1_v1["ID_BUSINESS_UNIT_MOD"]
print("\nFilas que cambiaron de unidad de negocio:", mask_changed.sum())

# Realizamos una tabla de mapeo de contingencia ORIGINAL vs MOD SOLO para filas que cambiaron
crosstab_cambios = (
    pd.crosstab(
        df_fd1_v1.loc[mask_changed, "ID_BUSINESS_UNIT"],
        df_fd1_v1.loc[mask_changed, "ID_BUSINESS_UNIT_MOD"]
    )
)

print("\nMatriz ORIGINAL x MOD para filas cambiadas (muestra 20x20 si es grande):")
display(crosstab_cambios.iloc[:20, :20])

# Mostramos los registros cambiados con contexto mínimo
cols_ver = ["ID_ORDER", "ID_BUILDING", "ID_BUSINESS_UNIT", "ID_BUSINESS_UNIT_MOD", "COUNTRY", "YEAR", "MONTH"]
df_fd1_v1.loc[mask_changed, cols_ver].head(20)

Número de únicos (ORIGINAL): 14
Número de únicos (MOD): 9

Primeros 30 únicos ORIGINAL: [0, 42, 43, 49, 50, 51, 2606, 1000000, 1000020, 1000021, 1000043, 1000070, 1000097, 1000109]

Primeros 30 únicos MOD: [43, 49, 50, 51, 2606, 1000020, 1000021, 1000070, 1000097]

Valores nuevos que APARECEN solo en MOD: []
Valores que DESAPARECEN en MOD (estaban en ORIGINAL): [0, 42, 1000000, 1000043, 1000109]

Frecuencias ORIGINAL (primeros 20 por código):
ID_BUSINESS_UNIT
0              23
42            170
43         226525
49           2409
50          10027
51           8261
2606         5569
1000000         3
1000020     26955
1000021     35432
1000043         5
1000070      8044
1000097       995
1000109         4
Name: count, dtype: int64

Frecuencias MOD (primeros 20 por código):
ID_BUSINESS_UNIT_MOD
43         226536
49           2409
50          10216
51           8265
2606         5569
1000020     26955
1000021     35433
1000070      8044
1000097       995
Name: count, dtype: int64

Filas que cambiaron de unidad de negocio: 205

Matriz ORIGINAL x MOD para filas cambiadas (muestra 20x20 si es grande):

# Diccionario inicial manual (cuando ID_BUSINESS_UNIT era 0 en ciertos edificios)
correccion_bu_por_building = {
    1001191: 43,
    1000496: 1000021,
    1065: 50,
    1000389: 43,
    1270: 43,
    1201: 43,
}

# Diccionario adicional por mayoría en los edificios afectados con unidades obsoletas
# (estos valores ya los vimos en la matriz de cambios: 42 → 50, 1000109 → 51, etc.)
correccion_obsoletas = {
    # Building : Unidad mayoritaria asignada
    18: 43,        # ID_BUSINESS_UNIT 42 y 1000000 reclasificados a 43
    9: 43,         # ID_BUSINESS_UNIT 42 y 1000043 reclasificados a 43
    1065: 50,      # ID_BUSINESS_UNIT 42 reclasificado a 50
    788: 43,       # ID_BUSINESS_UNIT 1000043 reclasificado a 43
    1001203: 51,   # ID_BUSINESS_UNIT 1000109 reclasificado a 51
}

# Diccionario final consolidado
diccionario_final_bu = {**correccion_obsoletas, **correccion_bu_por_building}

# Lo mostramos para dejarlo documentado
print("Diccionario final consolidado de correcciones ID_BUILDING → ID_BUSINESS_UNIT:")
for k, v in diccionario_final_bu.items():
    print(f"ID_BUILDING {k} → ID_BUSINESS_UNIT {v}")

Diccionario final consolidado de correcciones ID_BUILDING → ID_BUSINESS_UNIT:
ID_BUILDING 18 → ID_BUSINESS_UNIT 43
ID_BUILDING 9 → ID_BUSINESS_UNIT 43
ID_BUILDING 1065 → ID_BUSINESS_UNIT 50
ID_BUILDING 788 → ID_BUSINESS_UNIT 43
ID_BUILDING 1001203 → ID_BUSINESS_UNIT 51
ID_BUILDING 1001191 → ID_BUSINESS_UNIT 43
ID_BUILDING 1000496 → ID_BUSINESS_UNIT 1000021
ID_BUILDING 1000389 → ID_BUSINESS_UNIT 43
ID_BUILDING 1270 → ID_BUSINESS_UNIT 43
ID_BUILDING 1201 → ID_BUSINESS_UNIT 43

# Filtramos los registros con costes negativos
df_costes_negativos = df_fd1_v1[df_fd1_v1["cost_float"] < 0]

# Visualizamos el número de registros con coste negativo
print("Número de registros con coste negativo:", len(df_costes_negativos))

# Visualizamos el valor mínimo y resumen estadístico de estos casos
print("\nResumen estadístico de los costes negativos:")
print(df_costes_negativos["cost_float"].describe())

# Revisamos algunas filas con contexto
cols_revision = [
    "ID_ORDER", "COUNTRY", "YEAR", "MONTH",
    "FM_RESPONSIBLE", "FM_COST_TYPE", "ID_BUILDING",
    "ID_BUSINESS_UNIT_MOD", "cost_float"
]
df_costes_negativos[cols_revision].head(20)

Número de registros con coste negativo: 686

Resumen estadístico de los costes negativos:
count      686.000000
mean     -1409.865955
std       2960.970573
min     -40880.070000
25%      -1319.882500
50%       -628.655000
75%       -153.458500
max         -0.000600
Name: cost_float, dtype: float64

# Revisamos los valores únicos en los registros con coste negativo
for col in ["FM_RESPONSIBLE", "FM_COST_TYPE", "COST_TYPE", "COUNTRY", "SUPPLIER_TYPE_MOD"]:
    print(f"\nValores únicos en {col}:")
    print(df_costes_negativos[col].unique())

Valores únicos en FM_RESPONSIBLE:
['Eficiencia Energética']

Valores únicos en FM_COST_TYPE:
['Suministros']

Valores únicos en COST_TYPE:
['Gasto']

Valores únicos en COUNTRY:
['Colombia' 'España' 'Costa Rica' 'Italia' 'México' 'Perú' 'Panamá']

Valores únicos en SUPPLIER_TYPE_MOD:
['EXTERNO' 'INTERNO']

# Filtramos registros de proveedores internos con coste distinto de 0
df_interno_con_coste = df_fd1_v1[
    (df_fd1_v1["SUPPLIER_TYPE_MOD"] == "INTERNO") &
    (df_fd1_v1["cost_float"] != 0)
]

# Visualizamso el número de registros que obtenemos
print("Número de registros con SUPPLIER_TYPE_MOD = INTERNO y coste ≠ 0:", len(df_interno_con_coste))

# Calculamos la proporción respecto al total de registros con proveedores internos
total_internos = (df_fd1_v1["SUPPLIER_TYPE_MOD"] == "INTERNO").sum()
print("Proporción sobre total de internos:", round(len(df_interno_con_coste)/total_internos*100, 4), "%")

# Listamos los ID_SUPPLIER únicos implicados
proveedores_internos_afectados = df_interno_con_coste["ID_SUPPLIER"].unique()
print("\nID_SUPPLIER con SUPPLIER_TYPE_MOD = INTERNO y coste ≠ 0:")
print(proveedores_internos_afectados)

# Mostramos una muestra de registros con contexto para revisión
cols_revision = ["ID_ORDER", "ID_SUPPLIER", "SUPPLIER_TYPE_MOD", "COUNTRY", "YEAR", "MONTH", "FM_RESPONSIBLE", "FM_COST_TYPE", "cost_float"]
df_interno_con_coste[cols_revision].head(20)

Número de registros con SUPPLIER_TYPE_MOD = INTERNO y coste ≠ 0: 1027
Proporción sobre total de internos: 0.9459 %

ID_SUPPLIER con SUPPLIER_TYPE_MOD = INTERNO y coste ≠ 0:
[     12 3005192 3007021 3004568     158 3007026       4    1662    1811
 3004715 3001385    2548 3001092 3004995 3002884 3004389 3003582 3004890
 3001934 3004607 3003294 3004303 3003255]

# Usamos directamente la variable con los proveedores internos afectados
print("Número de proveedores internos afectados:", len(proveedores_internos_afectados))
print("Listado de proveedores internos afectados:")
print(proveedores_internos_afectados)

# Filtramos solo estos proveedores internos con coste distinto de 0
df_interno_coste_proveedores = df_fd1_v1[
    (df_fd1_v1["SUPPLIER_TYPE_MOD"] == "INTERNO") &
    (df_fd1_v1["cost_float"] != 0) &
    (df_fd1_v1["ID_SUPPLIER"].isin(proveedores_internos_afectados))
]

# Calculamos el coste total por proveedor
coste_por_proveedor = (
    df_interno_coste_proveedores
    .groupby("ID_SUPPLIER")["cost_float"]
    .sum()
    .reset_index()
    .sort_values(by="cost_float", ascending=False)
)

# Realizamos un conteo de registros por proveedor
conteo_por_proveedor = (
    df_interno_coste_proveedores
    .groupby("ID_SUPPLIER")["ID_ORDER"]
    .count()
    .reset_index()
    .rename(columns={"ID_ORDER": "n_registros"})
)

# Visualizamos un resumen combinado: coste total y número de registros
resumen_proveedores = coste_por_proveedor.merge(conteo_por_proveedor, on="ID_SUPPLIER")

print("\nResumen de proveedores internos con coste distinto de 0:")
resumen_proveedores

Número de proveedores internos afectados: 23
Listado de proveedores internos afectados:
[     12 3005192 3007021 3004568     158 3007026       4    1662    1811
 3004715 3001385    2548 3001092 3004995 3002884 3004389 3003582 3004890
 3001934 3004607 3003294 3004303 3003255]

Resumen de proveedores internos con coste distinto de 0:

# Filtramos los registros de proveedores internos con coste distinto de 0
df_interno_coste_proveedores = df_fd1_v1[
    (df_fd1_v1["SUPPLIER_TYPE_MOD"] == "INTERNO") &
    (df_fd1_v1["cost_float"] != 0) &
    (df_fd1_v1["ID_SUPPLIER"].isin(proveedores_internos_afectados))
]

# Agrupamos por proveedor y FM_RESPONSIBLE
resumen_proveedores_resp = (
    df_interno_coste_proveedores
    .groupby(["ID_SUPPLIER", "FM_RESPONSIBLE"])
    .agg(
        coste_total=("cost_float", "sum"),
        n_registros=("ID_ORDER", "count"),
        coste_medio=("cost_float", "mean")
    )
    .reset_index()
    .sort_values(by=["ID_SUPPLIER", "coste_total"], ascending=[True, False])
)

print("Resumen segmentado por proveedor interno y FM_RESPONSIBLE (coste ≠ 0):")
resumen_proveedores_resp

Resumen segmentado por proveedor interno y FM_RESPONSIBLE (coste ≠ 0):

# Caso 1: El SUPPLIER_ID concatenado con el FM_RESPONSIBLE y concatenado con SUPPLIER_TYPE_MOD deben tener cost_float igual a 0.

# Creamos la clave de concatenación en el dataset
df_fd1_v1["CONCAT_key"] = (
    df_fd1_v1["ID_SUPPLIER"].astype(str) +
    df_fd1_v1["FM_RESPONSIBLE"].astype(str) +
    df_fd1_v1["SUPPLIER_TYPE_MOD"].astype(str)
)

# Definimos el listado de claves donde el coste debe ser 0 según el equipo de FM
claves_a_cero = [
    "4Obras ProyectosINTERNO",
    "4Gestión EspaciosINTERNO",
    "4Oficina TécnicaINTERNO",
    "1662LicenciasINTERNO",
    "1811MantenimientoINTERNO",
    "2548MantenimientoINTERNO",
    "3001092MantenimientoINTERNO",
    "3001385MantenimientoINTERNO",
    "3001934MantenimientoINTERNO",
    "3002884MantenimientoINTERNO",
    "3003255MantenimientoINTERNO",
    "3003294MantenimientoINTERNO",
    "3003582MantenimientoINTERNO",
    "3004303MantenimientoINTERNO",
    "3004389MantenimientoINTERNO",
    "3004568Eficiencia EnergéticaINTERNO",
    "3004607MantenimientoINTERNO",
    "3004715Mantenimiento MultipuntoINTERNO",
    "3004995Mantenimiento MultipuntoINTERNO",
    "3004890MantenimientoINTERNO"
]

# Creamos cost_float_mod aplicando la regla
df_fd1_v1["cost_float_mod"] = np.where(
    df_fd1_v1["CONCAT_key"].isin(claves_a_cero),
    0,
    df_fd1_v1["cost_float"]
)

# Auditamos cuántos cambios se hicieron
cambios = (df_fd1_v1["cost_float"] != df_fd1_v1["cost_float_mod"]).sum()
print("Número de registros ajustados a 0 en cost_float_mod:", cambios)

# Visualizamos los registros modificados
cols_check = ["ID_ORDER", "ID_SUPPLIER", "FM_RESPONSIBLE", "SUPPLIER_TYPE_MOD", "cost_float", "cost_float_mod"]
df_fd1_v1.loc[df_fd1_v1["CONCAT_key"].isin(claves_a_cero), cols_check].head(20)

Número de registros ajustados a 0 en cost_float_mod: 81

# Contamos cuántos registros cambiaron de cost_float a cost_float_mod
cambios_total = (df_fd1_v1["cost_float"] != df_fd1_v1["cost_float_mod"]).sum()
print("Número de registros modificados en cost_float_mod:", cambios_total)

# Validamos si coincide con los 34 que nos ha pasado el equipo de FM.
if cambios_total == 34:
    print("Coincide exactamente con los 34 registros esperados.")
else:
    print(f"No coincide, se modificaron {cambios_total} registros en lugar de 34.")

Número de registros modificados en cost_float_mod: 81
No coincide, se modificaron 81 registros en lugar de 34.

# Filtramos solo los registros que están en claves_a_cero
df_cambios_segmento = df_fd1_v1[
    (df_fd1_v1["CONCAT_key"].isin(claves_a_cero)) &
    (df_fd1_v1["cost_float"] != df_fd1_v1["cost_float_mod"])
]

# Número de registros modificados en el segmento
print("Número de registros modificados en claves_a_cero:", len(df_cambios_segmento))

# Mostramos una muestra para revisión
cols_check = ["ID_ORDER", "ID_SUPPLIER", "FM_RESPONSIBLE", "SUPPLIER_TYPE_MOD", "cost_float", "cost_float_mod"]
df_cambios_segmento[cols_check].head(34)

Número de registros modificados en claves_a_cero: 34

# Filtramos registros donde cost_float y cost_float_mod son distintos
df_cambios_totales = df_fd1_v1[df_fd1_v1["cost_float"] != df_fd1_v1["cost_float_mod"]]

# Número de registros modificados en todo el dataset
print("Número total de registros modificados:", len(df_cambios_totales))

# Proporción respecto al total
print("Proporción sobre el total del dataset:", round(len(df_cambios_totales) / len(df_fd1_v1) * 100, 4), "%")

# Mostramos algunos registros de ejemplo
cols_check = ["ID_ORDER", "ID_SUPPLIER", "FM_RESPONSIBLE", "SUPPLIER_TYPE_MOD", "cost_float", "cost_float_mod"]
df_cambios_totales[cols_check].head(20)

Número total de registros modificados: 81
Proporción sobre el total del dataset: 0.025 %

# Caso 2: proveedores que deben pasar a EXTERNO según la concatenación SUPPLIER_ID + FM_RESPONSIBLE + SUPPLIER_TYPE_MOD

# Definimos el segmento objetivo (claves que deben ser EXTERNO)
claves_externo = [
    "12Eficiencia EnergéticaINTERNO",
    "158Eficiencia EnergéticaINTERNO",
    "3005192Eficiencia EnergéticaINTERNO",
    "3007021Eficiencia EnergéticaINTERNO",
    "3007026Eficiencia EnergéticaINTERNO",
]

# Generamos la nueva variable SUPPLIER_TYPE_MOD_2
#    - Si la concatenación está en claves_externo -> "EXTERNO"
#    - En caso contrario, mantiene el valor de SUPPLIER_TYPE_MOD
df_fd1_v1["SUPPLIER_TYPE_MOD_2"] = np.where(
    df_fd1_v1["CONCAT_key"].isin(claves_externo),
    "EXTERNO",
    df_fd1_v1["SUPPLIER_TYPE_MOD"]
)

# Verificamos el número de cambios respecto a SUPPLIER_TYPE_MOD
cambios_sup_type = (df_fd1_v1["SUPPLIER_TYPE_MOD_2"] != df_fd1_v1["SUPPLIER_TYPE_MOD"]).sum()
print("Número de cambios en SUPPLIER_TYPE_MOD_2 respecto a SUPPLIER_TYPE_MOD:", cambios_sup_type)

# Comprobamos contra el objetivo (993) que nos especifican desde FM
if cambios_sup_type == 993:
    print("Verificación OK: hay exactamente 993 cambios.")
else:
    print(f"Atención: se han detectado {cambios_sup_type} cambios (no coincide con 993).")

# Visualizamos una muestra de las filas cambiadas para auditoría
cols_check = ["ID_ORDER", "ID_SUPPLIER", "FM_RESPONSIBLE", "SUPPLIER_TYPE_MOD", "SUPPLIER_TYPE_MOD_2"]
df_fd1_v1.loc[
    df_fd1_v1["SUPPLIER_TYPE_MOD_2"] != df_fd1_v1["SUPPLIER_TYPE_MOD"],
    cols_check
].head(20)

Número de cambios en SUPPLIER_TYPE_MOD_2 respecto a SUPPLIER_TYPE_MOD: 993
Verificación OK: hay exactamente 993 cambios.

# Creamos el dataset definitivo excluyendo Marruecos e Italia
df_fd1_v2 = df_fd1_v1[~df_fd1_v1["COUNTRY"].isin(["Marruecos", "Italia"])].copy()

# Comprobamos las dimensiones antes y después
print("Filas en df_fd1_v1:", df_fd1_v1.shape[0])
print("Filas en df_fd1_v2:", df_fd1_v2.shape[0])

# Validamos que ya no haya registros de Marruecos o Italia
print("Países presentes en df_fd1_v2:", df_fd1_v2["COUNTRY"].unique())

Filas en df_fd1_v1: 324422
Filas en df_fd1_v2: 323320
Países presentes en df_fd1_v2: ['España' 'México' 'Costa Rica' 'Panamá' 'Perú' 'República Dominicana'
 'Colombia']

# Creamos una nueva columna FM_RESPONSIBLE_MOD que unifica Mantenimiento y Mantenimiento Multipunto
df_fd1_v2["FM_RESPONSIBLE_MOD"] = df_fd1_v2["FM_RESPONSIBLE"].replace(
    {"Mantenimiento Multipunto": "Mantenimiento"}
)

# Auditamos valores únicos antes y después
print("Valores únicos en FM_RESPONSIBLE original:", df_fd1_v2["FM_RESPONSIBLE"].unique())
print("Valores únicos en FM_RESPONSIBLE_MOD unificada:", df_fd1_v2["FM_RESPONSIBLE_MOD"].unique())

# Comprobamos cuántos registros fueron reclasificados
cambiados = (df_fd1_v2["FM_RESPONSIBLE"] != df_fd1_v2["FM_RESPONSIBLE_MOD"]).sum()
print("Número de registros reclasificados de 'Mantenimiento Multipunto' a 'Mantenimiento':", cambiados)

Valores únicos en FM_RESPONSIBLE original: ['Mantenimiento Multipunto' 'Mantenimiento' 'Eficiencia Energética'
 'Licencias' 'Dirección FM' 'Obras Proyectos' 'Oficina Técnica'
 'Gestión Espacios' 'Control y Reporting' 'Real Estate' 'Global'
 'Operaciones']
Valores únicos en FM_RESPONSIBLE_MOD unificada: ['Mantenimiento' 'Eficiencia Energética' 'Licencias' 'Dirección FM'
 'Obras Proyectos' 'Oficina Técnica' 'Gestión Espacios'
 'Control y Reporting' 'Real Estate' 'Global' 'Operaciones']
Número de registros reclasificados de 'Mantenimiento Multipunto' a 'Mantenimiento': 12729

# Definimos las clases minoritarias a eliminar
clases_minor_eliminar = [
    "Oficina Técnica", "Dirección FM", "Global",
    "Real Estate", "Control y Reporting", "Operaciones"
]

# Número de registros antes
print("Filas antes de eliminar:", df_fd1_v2.shape[0])

# Coste total antes de eliminar
coste_total_antes = df_fd1_v2["cost_float_mod"].sum()

# Coste total de las clases eliminadas
coste_eliminado = (
    df_fd1_v2.loc[df_fd1_v2["FM_RESPONSIBLE"].isin(clases_minor_eliminar), "cost_float_mod"]
    .sum()
)

# Creamos un nuevo dataset sin estas clases
df_fd1_v3 = df_fd1_v2[~df_fd1_v2["FM_RESPONSIBLE"].isin(clases_minor_eliminar)].copy()
print(df_fd1_v3)

# Número de registros después
print("Filas después de eliminar:", df_fd1_v3.shape[0])

# Coste total después de eliminar
coste_total_despues = df_fd1_v3["cost_float_mod"].sum()

# Verificamos que ya no existan esas categorías
print("\nValores únicos de FM_RESPONSIBLE en df_fd1_v3:")
print(df_fd1_v3["FM_RESPONSIBLE_MOD"].unique())

# Proporción eliminada del coste
proporcion_eliminada = round(coste_eliminado / coste_total_antes * 100, 4)

print("Coste total antes de eliminar:", round(coste_total_antes, 2))
print("Coste total eliminado:", round(coste_eliminado, 2))
print("Coste total después de eliminar:", round(coste_total_despues, 2))
print("Proporción eliminada sobre el total del coste:", proporcion_eliminada, "%")

Filas antes de eliminar: 323320
          ID_ORDER COST_TYPE COUNTRY  ID_BUILDING  YEAR_MONTH  \
0       F001108989     Gasto  España      1001131      202111
1       F001119565     Gasto  España      1000026      202107
2       F001119567     Gasto  España      1000515      202107
3       F001120426     Gasto  España            9      202101
4       F001120655     Gasto  España          116      202102
...            ...       ...     ...          ...         ...
324434   701948945     Gasto  España          594      202509
324435   701948953     Gasto  España      1001601      202509
324436   701948955     Gasto  España      1001598      202509
324437   701948956     Gasto  España         1572      202509
324438   701948957     Gasto  España         1572      202509

                  FM_RESPONSIBLE      FM_COST_TYPE    COST  ID_CUSTOMER  \
0       Mantenimiento Multipunto   Mtto. Contratos      77      3004581
1                  Mantenimiento   Mtto. Contratos   36.95           13
2                  Mantenimiento   Mtto. Contratos   36.95          222
3                  Mantenimiento   Mtto. Contratos  605.45           12
4       Mantenimiento Multipunto   Mtto. Contratos  277.19          140
...                          ...               ...     ...          ...
324434             Mantenimiento  Mtto. Correctivo     128          623
324435           Obras Proyectos             Obras     326      3003189
324436           Obras Proyectos             Obras  348.75      3005142
324437             Mantenimiento  Mtto. Correctivo       0          157
324438             Mantenimiento  Mtto. Correctivo       0          157

        ID_SUPPLIER  ...  YEAR QUARTER  cost_float  SUPPLIER_TYPE_MOD  \
0           3004832  ...  2021       4       77.00            EXTERNO
1              1532  ...  2021       3       36.95            EXTERNO
2              1532  ...  2021       3       36.95            EXTERNO
3           3002542  ...  2021       1      605.45            EXTERNO
4              1532  ...  2021       1      277.19            EXTERNO
...             ...  ...   ...     ...         ...                ...
324434      3005407  ...  2025       3      128.00            EXTERNO
324435      3005230  ...  2025       3      326.00            EXTERNO
324436      3006714  ...  2025       3      348.75            EXTERNO
324437         1811  ...  2025       3        0.00            INTERNO
324438         1811  ...  2025       3        0.00            INTERNO

        ID_BUSINESS_UNIT_MOD  bu_corr_flag  \
0                    1000021         False
1                    1000021         False
2                    1000021         False
3                         50         False
4                    1000020         False
...                      ...           ...
324434               1000020         False
324435                  2606         False
324436                  2606         False
324437                    43         False
324438                    43         False

                                    CONCAT_key  cost_float_mod  \
0       3004832Mantenimiento MultipuntoEXTERNO           77.00
1                     1532MantenimientoEXTERNO           36.95
2                     1532MantenimientoEXTERNO           36.95
3                  3002542MantenimientoEXTERNO          605.45
4          1532Mantenimiento MultipuntoEXTERNO          277.19
...                                        ...             ...
324434             3005407MantenimientoEXTERNO          128.00
324435           3005230Obras ProyectosEXTERNO          326.00
324436           3006714Obras ProyectosEXTERNO          348.75
324437                1811MantenimientoINTERNO            0.00
324438                1811MantenimientoINTERNO            0.00

        SUPPLIER_TYPE_MOD_2 FM_RESPONSIBLE_MOD
0                   EXTERNO      Mantenimiento
1                   EXTERNO      Mantenimiento
2                   EXTERNO      Mantenimiento
3                   EXTERNO      Mantenimiento
4                   EXTERNO      Mantenimiento
...                     ...                ...
324434              EXTERNO      Mantenimiento
324435              EXTERNO    Obras Proyectos
324436              EXTERNO    Obras Proyectos
324437              INTERNO      Mantenimiento
324438              INTERNO      Mantenimiento

[323090 rows x 23 columns]
Filas después de eliminar: 323090

Valores únicos de FM_RESPONSIBLE en df_fd1_v3:
['Mantenimiento' 'Eficiencia Energética' 'Licencias' 'Obras Proyectos'
 'Gestión Espacios']
Coste total antes de eliminar: 272276546.67
Coste total eliminado: 136814.78
Coste total después de eliminar: 272139731.89
Proporción eliminada sobre el total del coste: 0.0502 %

!pip install unidecode

Collecting unidecode
  Downloading Unidecode-1.4.0-py3-none-any.whl.metadata (13 kB)
Downloading Unidecode-1.4.0-py3-none-any.whl (235 kB)
   ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 0.0/235.8 kB ? eta -:--:--
   ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╸ 235.5/235.8 kB 12.0 MB/s eta 0:00:01
   ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 235.8/235.8 kB 6.4 MB/s eta 0:00:00
Installing collected packages: unidecode
Successfully installed unidecode-1.4.0

# Montamos Drive si no está montado
from google.colab import drive
drive.mount('/content/drive')

# Definimos las Rutas de salida
ruta_csv_tildes = "/content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/df_fd1_v3.csv"
ruta_csv_sin_tildes = "/content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/df_fd1_v3_sin_tildes.csv"
ruta_xlsx_tildes = "/content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/df_fd1_v3.xlsx"
ruta_xlsx_sin_tildes = "/content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/df_fd1_v3_sin_tildes.xlsx"

# Guardamos la versión con tildes
df_fd1_v3.to_csv(ruta_csv_tildes, sep=",", index=False, encoding="utf-8")
df_fd1_v3.to_excel(ruta_xlsx_tildes, index=False, engine="openpyxl")

# Creamos copia sin tildes
from unidecode import unidecode

df_fd1_v3_sin_tildes = df_fd1_v3.copy()
for col in df_fd1_v3_sin_tildes.select_dtypes(include=["object"]).columns:
    df_fd1_v3_sin_tildes[col] = df_fd1_v3_sin_tildes[col].apply(lambda x: unidecode(str(x)) if pd.notnull(x) else x)

# Guardamos la versión sin tildes
df_fd1_v3_sin_tildes.to_csv(ruta_csv_sin_tildes, sep=",", index=False, encoding="utf-8")
df_fd1_v3_sin_tildes.to_excel(ruta_xlsx_sin_tildes, index=False, engine="openpyxl")

print("Versión con tildes guardada en CSV y XLSX.")
print("Versión sin tildes guardada en CSV y XLSX.")

Drive already mounted at /content/drive; to attempt to forcibly remount, call drive.mount("/content/drive", force_remount=True).
Versión con tildes guardada en CSV y XLSX.
Versión sin tildes guardada en CSV y XLSX.

# Identificamos las columnas categóricas (tipo object)
cols_categoricas = df_fd1_v3_sin_tildes.select_dtypes(include=["object"]).columns

print("Columnas categóricas:", list(cols_categoricas))

# Revisamos valores únicos por cada columna categórica
for col in cols_categoricas:
    print(f"\nValores únicos en {col}:")
    print(df_fd1_v3_sin_tildes[col].unique())

Columnas categóricas: ['ID_ORDER', 'COST_TYPE', 'COUNTRY', 'FM_RESPONSIBLE', 'FM_COST_TYPE', 'COST', 'SUPPLIER_TYPE', 'SUPPLIER_TYPE_MOD', 'CONCAT_key', 'SUPPLIER_TYPE_MOD_2', 'FM_RESPONSIBLE_MOD']

Valores únicos en ID_ORDER:
['F001108989' 'F001119565' 'F001119567' ... '701948955' '701948956'
 '701948957']

Valores únicos en COST_TYPE:
['Gasto' 'Inversion']

Valores únicos en COUNTRY:
['Espana' 'Mexico' 'Costa Rica' 'Panama' 'Peru' 'Republica Dominicana'
 'Colombia']

Valores únicos en FM_RESPONSIBLE:
['Mantenimiento Multipunto' 'Mantenimiento' 'Eficiencia Energetica'
 'Licencias' 'Obras Proyectos' 'Gestion Espacios']

Valores únicos en FM_COST_TYPE:
['Mtto. Contratos' 'Servicios Ctto.' 'Suministros' 'Eficiencia Energetica'
 'Servicios Extra' 'Obras' 'Mtto. Correctivo' 'Licencias']

Valores únicos en COST:
['77' '36.95' '605.45' ... '89.412' '151.7735' '298.75']

Valores únicos en SUPPLIER_TYPE:
['EXTERNO' 'INTERNO' '0']

Valores únicos en SUPPLIER_TYPE_MOD:
['EXTERNO' 'INTERNO']

Valores únicos en CONCAT_key:
['3004832Mantenimiento MultipuntoEXTERNO' '1532MantenimientoEXTERNO'
 '3002542MantenimientoEXTERNO' ... '3007635MantenimientoEXTERNO'
 '3007643Obras ProyectosEXTERNO' '3006925MantenimientoEXTERNO']

Valores únicos en SUPPLIER_TYPE_MOD_2:
['EXTERNO' 'INTERNO']

Valores únicos en FM_RESPONSIBLE_MOD:
['Mantenimiento' 'Eficiencia Energetica' 'Licencias' 'Obras Proyectos'
 'Gestion Espacios']

# Montar Drive (si no está montado)
from google.colab import drive
drive.mount('/content/drive')

Mounted at /content/drive

# Ruta del notebook .ipynb que vamos a convertir

RUTA_NOTEBOOK = "/content/drive/MyDrive/PROYECTO_NEITH/DOCUMENTOS_PROYECTOS/Neith.ipynb"

# Carpeta de salida y nombre del HTML
OUTPUT_DIR = "/content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO"
OUTPUT_HTML = "Neith_notebook_v1-v3.html"

# Aseguramos carpeta de salida (ya en cabecera)
# import os
os.makedirs(OUTPUT_DIR, exist_ok=True)

# Instalamos nbconvert (ya en cabecera)
# !pip install -q "nbconvert>=7.0.0"

# Convertimos notebook a HTML y lo guardamos con el nombre indicado en la carpeta de salida
!jupyter nbconvert --to html "$RUTA_NOTEBOOK" --output "$OUTPUT_HTML" --output-dir "$OUTPUT_DIR"

# Verificamos
salida = os.path.join(OUTPUT_DIR, OUTPUT_HTML)
print("Archivo HTML generado en:", salida, "\nExiste:", os.path.exists(salida))

Mounted at /content/drive
[NbConvertApp] Converting notebook /content/drive/MyDrive/PROYECTO_NEITH/DOCUMENTOS_PROYECTOS/Neith.ipynb to html
[NbConvertApp] Writing 727099 bytes to /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/Neith_notebook_v1-v3.html
Archivo HTML generado en: /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/Neith_notebook_v1-v3.html | Existe: True

# Montar Drive (si no está montado)
from google.colab import drive
drive.mount('/content/drive')

# cargamos df_fd1_v3.xlsx y Catalogo_inmuebles_Actualizado_a_092025.xlsx en dos dataframes.

# Definimos rutas (ajustar si fuese necesario)
ruta_fd1_v3 = "/content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/df_fd1_v3.xlsx"
ruta_catalogo = "/content/drive/MyDrive/PROYECTO_NEITH/FUENTE_DATOS/CATALOGOS/Catalogo_inmuebles_Actualizado_a_092025.xlsx"

# Validamos que las rutas existen para evitar errores silenciosos
for ruta in [ruta_fd1_v3, ruta_catalogo]:
    if not os.path.exists(ruta):
        raise FileNotFoundError(f"No encontramos el archivo en la ruta: {ruta}. Verifiquemos el nombre o la carpeta.")

# Cargamos los Excel
#    Si los ficheros tienen múltiples hojas y debemos elegir una, usamos sheet_name="NombreDeHoja".
#    Por defecto, pandas leerá la primera hoja.
df_fd1_v3 = pd.read_excel(ruta_fd1_v3)  # sheet_name=0 por defecto
df_catalogo_inmuebles = pd.read_excel(ruta_catalogo)  # sheet_name=0 por defecto

# Mostramos un resumen rápido para verificar que todo está correcto
def resumen_df(nombre, df):
    print(f"\n=== {nombre} ===")
    print(f"Filas: {df.shape[0]:,} | Columnas: {df.shape[1]:,}")
    print("Columnas:", list(df.columns)[:15], ("..." if df.shape[1] > 15 else ""))
    display(df.head(3))

resumen_df("df_fd1_v3", df_fd1_v3)
resumen_df("df_catalogo_inmuebles", df_catalogo_inmuebles)

=== df_fd1_v3 ===
Filas: 323,090 | Columnas: 23
Columnas: ['ID_ORDER', 'COST_TYPE', 'COUNTRY', 'ID_BUILDING', 'YEAR_MONTH', 'FM_RESPONSIBLE', 'FM_COST_TYPE', 'COST', 'ID_CUSTOMER', 'ID_SUPPLIER', 'ID_BUSINESS_UNIT', 'SUPPLIER_TYPE', 'MONTH', 'YEAR', 'QUARTER'] ...

=== df_catalogo_inmuebles ===
Filas: 1,650 | Columnas: 10
Columnas: ['ID_BUILDING', 'ID_REGION', 'COUNTRY', 'STATUS', 'FM_PERIMETER', 'SERVICE_LEVEL', 'TIPO_USO', 'M2_AVAIL_PERIMETER', 'M2_AVAIL', 'M2_PROM_USO']

# Realizamos sun análisis descriptivo del catálogo de inmuebles
# para obtener estadísticas generales de df_catalogo_inmuebles

# Analisis estadístico estándar de pandas para las variables numéricas
print("=== DESCRIBE (numéricas) ===")
display(df_catalogo_inmuebles.describe().T)

# Añadimos también categóricas (object / string)
print("\n=== DESCRIBE (incluyendo categóricas) ===")
display(df_catalogo_inmuebles.describe(include="all").T)

# Analizamos el tipos de datos y vemos si hay nulos
print("\n=== INFO ===")
print(df_catalogo_inmuebles.info())

# Contamos los valores nulos por columna
print("\n=== NULOS POR COLUMNA ===")
print(df_catalogo_inmuebles.isna().sum())

# Identificamos variables categóricas
cat_cols = df_catalogo_inmuebles.select_dtypes(include=["object"]).columns

# Visualizamos sus valores únicos
print("=== VARIABLES CATEGÓRICAS ===")
for col in cat_cols:
    print(f"\nColumna: {col}")
    print(df_catalogo_inmuebles[col].unique())

# Identificamos variables numéricas
num_cols = df_catalogo_inmuebles.select_dtypes(include=["int64", "float64"]).columns

# Visualizamos los valores únicos de las variables numéricas con menos de 5 clases
print("\n=== VARIABLES NUMÉRICAS CON < 5 CLASES ===")
for col in num_cols:
    valores_unicos = df_catalogo_inmuebles[col].unique()
    n_unicos = len(valores_unicos)
    if n_unicos < 5:
        print(f"\nColumna: {col} | Nº clases: {n_unicos}")
        print(valores_unicos)

=== DESCRIBE (numéricas) ===

=== DESCRIBE (incluyendo categóricas) ===

=== INFO ===
<class 'pandas.core.frame.DataFrame'>
RangeIndex: 1650 entries, 0 to 1649
Data columns (total 10 columns):
 #   Column              Non-Null Count  Dtype
---  ------              --------------  -----
 0   ID_BUILDING         1650 non-null   int64
 1   ID_REGION           1650 non-null   int64
 2   COUNTRY             1650 non-null   object
 3   STATUS              1650 non-null   int64
 4   FM_PERIMETER        1650 non-null   int64
 5   SERVICE_LEVEL       1650 non-null   object
 6   TIPO_USO            1650 non-null   object
 7   M2_AVAIL_PERIMETER  1650 non-null   float64
 8   M2_AVAIL            1650 non-null   float64
 9   M2_PROM_USO         1650 non-null   float64
dtypes: float64(3), int64(4), object(3)
memory usage: 129.0+ KB
None

=== NULOS POR COLUMNA ===
ID_BUILDING           0
ID_REGION             0
COUNTRY               0
STATUS                0
FM_PERIMETER          0
SERVICE_LEVEL         0
TIPO_USO              0
M2_AVAIL_PERIMETER    0
M2_AVAIL              0
M2_PROM_USO           0
dtype: int64
=== VARIABLES CATEGÓRICAS ===

Columna: COUNTRY
['España' 'Perú' 'Panamá' 'México' 'Costa Rica' 'Colombia' 'Italia'
 'República Dominicana' 'desconocido' 'Malta' 'Puerto Rico' 'Marruecos'
 'Venezuela']

Columna: SERVICE_LEVEL
['Compact' 0 'Solo Suministros' 'Sin Servicio' 'Básico' 'Completo'
 'Esencial' 'Económico' 'Fuera Perímetro' 'eCorner' 'Solo Licencias']

Columna: TIPO_USO
['Bingo' 'Sin Uso Actual' 'Salón de Juego' 'Bar/Restaurante' 'Delegación'
 'Sin actividad' 'Parking' 'Oficinas' 'Casino Electrónico' 'Subarrendado'
 'Almacén/Bodega' 'Casino Tradicional' 'Industrial' 'Local Apuestas'
 'Hotel']

=== VARIABLES NUMÉRICAS CON < 5 CLASES ===

Columna: STATUS | Nº clases: 2
[1 0]

Columna: FM_PERIMETER | Nº clases: 2
[1 0]

# Filtramos las filas con SERVICE_LEVEL = 0
df_service_level_cero = df_catalogo_inmuebles[df_catalogo_inmuebles["SERVICE_LEVEL"] == 0]

# Mostramos cantidad de registros encontrados
print(f"Registros con SERVICE_LEVEL = 0: {len(df_service_level_cero)}")

# Obtenemos valores únicos de FM_PERIMETER y STATUS para los casos con SERVICE_LEVEL igual a 0.
print("\nValores únicos de FM_PERIMETER en esos casos:")
print(df_service_level_cero["FM_PERIMETER"].unique())

print("\nValores únicos de STATUS en esos casos:")
print(df_service_level_cero["STATUS"].unique())

# Mostramos las primeras filas para inspección manual
display(df_service_level_cero.head())

Registros con SERVICE_LEVEL = 0: 824

Valores únicos de FM_PERIMETER en esos casos:
[0]

Valores únicos de STATUS en esos casos:
[0]

# Creamos un nuevo dataframe filtrado
df_catalogo_inmuebles_filtrado = df_catalogo_inmuebles[
    (df_catalogo_inmuebles["STATUS"] == 1) &
    (df_catalogo_inmuebles["FM_PERIMETER"] == 1)
].copy()

print("=== RESUMEN FILTRADO ===")
print(f"Filas originales: {len(df_catalogo_inmuebles):,}")
print(f"Filas después del filtrado: {len(df_catalogo_inmuebles_filtrado):,}")

# Verificamos que ya no exista el valor 0 en SERVICE_LEVEL
valores_unicos_service = df_catalogo_inmuebles_filtrado["SERVICE_LEVEL"].unique()
print("\nValores únicos de SERVICE_LEVEL después del filtrado:")
print(valores_unicos_service)

# Verificamos también que STATUS y FM_PERIMETER sean constantes en 1
print("\nValores únicos de STATUS en el catálogo filtrado:", df_catalogo_inmuebles_filtrado["STATUS"].unique())
print("Valores únicos de FM_PERIMETER en el catálogo filtrado:", df_catalogo_inmuebles_filtrado["FM_PERIMETER"].unique())

# Realizamos una vista rápida de los primeros registros filtrados
display(df_catalogo_inmuebles_filtrado.head())

=== RESUMEN FILTRADO ===
Filas originales: 1,650
Filas después del filtrado: 617

Valores únicos de SERVICE_LEVEL después del filtrado:
['Compact' 'Solo Suministros' 'Básico' 'Completo' 'Esencial' 'Económico'
 'Solo Licencias']

Valores únicos de STATUS en el catálogo filtrado: [1]
Valores únicos de FM_PERIMETER en el catálogo filtrado: [1]

# Alineamos tipos de ID_BUILDING para evitar desajustes en la comparación
#    (lo hacemos de forma segura por si alguna fuente viene como string)
df_fd1_v3["ID_BUILDING"] = pd.to_numeric(df_fd1_v3["ID_BUILDING"], errors="coerce")
df_catalogo_inmuebles_filtrado["ID_BUILDING"] = pd.to_numeric(df_catalogo_inmuebles_filtrado["ID_BUILDING"], errors="coerce")

# Obtenemos conjuntos de IDs
ids_v3 = set(df_fd1_v3["ID_BUILDING"].dropna().unique())
ids_catalogo = set(df_catalogo_inmuebles_filtrado["ID_BUILDING"].dropna().unique())

# Calculamos los ID_BUILDING de v3 que no existen en el catálogo filtrado
ids_fuera_catalogo = ids_v3 - ids_catalogo
print(f"ID_BUILDING en v3: {len(ids_v3):,} | ID_BUILDING en catálogo filtrado: {len(ids_catalogo):,}")
print(f"ID_BUILDING de v3 fuera del catálogo filtrado: {len(ids_fuera_catalogo):,}")

# Creamos el diccionario con nº de registros por ID_BUILDING excluido
dicc_ids_fuera = (
    df_fd1_v3[df_fd1_v3["ID_BUILDING"].isin(ids_fuera_catalogo)]
    .groupby("ID_BUILDING")
    .size()
    .to_dict()
)

print("\n=== Diccionario (primeros 10) de inmuebles excluidos ===")
for i, (k, v) in enumerate(dicc_ids_fuera.items()):
    if i < 10:
        print(f"ID_BUILDING: {k} -> Registros: {v}")
    else:
        break
print(f"Total inmuebles excluidos: {len(dicc_ids_fuera):,}")

# Separamos datasets:
#    - df_fd1_v3_inmuebles_no_vigentes: registros a eliminar del histórico
#    - df_fd1_v3_limpio: registros válidos que conservaremos
df_fd1_v3_inmuebles_no_vigentes = df_fd1_v3[df_fd1_v3["ID_BUILDING"].isin(ids_fuera_catalogo)].copy()
df_fd1_v3_limpio = df_fd1_v3[~df_fd1_v3["ID_BUILDING"].isin(ids_fuera_catalogo)].copy()

print("\n=== RESUMEN SEPARACIÓN ===")
print(f"Filas originales en v3: {len(df_fd1_v3):,}")
print(f"Filas excluidas (no vigentes / fuera de perímetro): {len(df_fd1_v3_inmuebles_no_vigentes):,}")
print(f"Filas restantes (v3 limpio): {len(df_fd1_v3_limpio):,}")

# Reemplazamos df_fd1_v3 por su versión limpia para continuar el pipeline
df_fd1_v3 = df_fd1_v3_limpio
del df_fd1_v3_limpio  # liberamos memoria

# Verificamos que todos los ID_BUILDING de v3 existan en el catálogo filtrado
resto_ids_fuera = set(df_fd1_v3["ID_BUILDING"].dropna().unique()) - ids_catalogo
print(f"\nIDs fuera de catálogo tras limpieza (debería ser 0): {len(resto_ids_fuera)}")

ID_BUILDING en v3: 693 | ID_BUILDING en catálogo filtrado: 617
ID_BUILDING de v3 fuera del catálogo filtrado: 127

=== Diccionario (primeros 10) de inmuebles excluidos ===
ID_BUILDING: 169 -> Registros: 86
ID_BUILDING: 180 -> Registros: 31
ID_BUILDING: 182 -> Registros: 110
ID_BUILDING: 196 -> Registros: 2
ID_BUILDING: 198 -> Registros: 61
ID_BUILDING: 342 -> Registros: 1
ID_BUILDING: 344 -> Registros: 14
ID_BUILDING: 357 -> Registros: 421
ID_BUILDING: 566 -> Registros: 1
ID_BUILDING: 618 -> Registros: 149
Total inmuebles excluidos: 127

=== RESUMEN SEPARACIÓN ===
Filas originales en v3: 323,090
Filas excluidas (no vigentes / fuera de perímetro): 6,219
Filas restantes (v3 limpio): 316,871

IDs fuera de catálogo tras limpieza (debería ser 0): 0

# Visualizamos el dataset del historico de contratación asociado a espacios no vigentes o fuera de perimetro FM.

df_fd1_v3_inmuebles_no_vigentes

# Guardamos el diccionario de excluidos como dataframe para trazabilidad
df_inmuebles_excluidos = (
    pd.DataFrame(list(dicc_ids_fuera.items()), columns=["ID_BUILDING", "num_registros_excluidos"])
    .sort_values("num_registros_excluidos", ascending=False)
    .reset_index(drop=True)
)
print("=== TRAZABILIDAD EXCLUIDOS (top 10) ===")
display(df_inmuebles_excluidos.head(10))

# Seleccionamos columnas del catálogo a incorporar como contexto
cols_ctx = ["ID_BUILDING", "ID_REGION", "SERVICE_LEVEL", "TIPO_USO", "COUNTRY"]
catalogo_ctx = df_catalogo_inmuebles_filtrado[cols_ctx].copy()

# Verificamos unicidad de ID_BUILDING en el catálogo filtrado
dups = catalogo_ctx["ID_BUILDING"].duplicated(keep="first").sum()
if dups > 0:
    print(f"Aviso: encontramos {dups} ID_BUILDING duplicados en el catálogo filtrado. "
          f"Nos quedamos con la primera aparición.")
    catalogo_ctx = catalogo_ctx.drop_duplicates(subset=["ID_BUILDING"], keep="first")

# Alineamos tipos (por seguridad)
df_fd1_v3["ID_BUILDING"] = pd.to_numeric(df_fd1_v3["ID_BUILDING"], errors="coerce")
catalogo_ctx["ID_BUILDING"] = pd.to_numeric(catalogo_ctx["ID_BUILDING"], errors="coerce")

# Calculamos intersección de ID_BUILDING como control previo
ids_inter = pd.Index(df_fd1_v3["ID_BUILDING"].unique()).intersection(catalogo_ctx["ID_BUILDING"].unique())
print(f"Edificios únicos en v3 limpio: {df_fd1_v3['ID_BUILDING'].nunique():,}")
print(f"Edificios únicos en catálogo filtrado: {catalogo_ctx['ID_BUILDING'].nunique():,}")
print(f"Edificios en intersección: {len(ids_inter):,}")

# Realizamos la unión (left join) v3 limpio + catálogo filtrado para crear v4
df_fd1_v4 = df_fd1_v3.merge(
    catalogo_ctx,
    on="ID_BUILDING",
    how="left",
    validate="m:1"  # muchas órdenes por edificio, un registro por edificio en el catálogo filtrado
)

# === Normalizamos COUNTRY tras el merge y calculamos cobertura sin errores ===

# Si existen ambas, nos quedamos con COUNTRY (v3) y renombramos la del catálogo
cols = set(df_fd1_v4.columns)
if "COUNTRY_x" in cols and "COUNTRY_y" in cols:
    df_fd1_v4 = df_fd1_v4.rename(columns={
        "COUNTRY_x": "COUNTRY",          # la de v3
        "COUNTRY_y": "COUNTRY_CATALOGO"  # la del catálogo
    })
elif "COUNTRY_y" in cols and "COUNTRY" not in cols:
    # Caso raro: solo llegó la del catálogo
    df_fd1_v4 = df_fd1_v4.rename(columns={"COUNTRY_y": "COUNTRY_CATALOGO"})
# Si solo existe COUNTRY, no hacemos nada.

# Definimos las columnas nuevas de contexto de forma segura
candidatas_nuevas = ["ID_REGION", "SERVICE_LEVEL", "TIPO_USO", "COUNTRY_CATALOGO", "COUNTRY"]
cols_nuevas_presentes = [c for c in candidatas_nuevas if c in df_fd1_v4.columns]

# Cobertura: filas con todas las nuevas columnas a NA
if cols_nuevas_presentes:
    mask_na = df_fd1_v4[cols_nuevas_presentes].isna().all(axis=1)
    num_na = mask_na.sum()
    porc_na = 100.0 * num_na / len(df_fd1_v4)
else:
    num_na, porc_na = 0, 0.0  # si por algún motivo no está ninguna, evitamos fallo

print("\n=== RESUMEN UNIÓN v3 limpio + catálogo inmuebles filtrado ===")
print(f"Filas v3 limpio: {len(df_fd1_v3):,}  ->  Filas v4: {len(df_fd1_v4):,}")
print(f"Columnas añadidas detectadas: {cols_nuevas_presentes}")
print(f"Filas sin match (todas las nuevas columnas NA): {num_na:,} ({porc_na:.2f}%)")

# Visualizamos si tenemos valores NAs por columna nueva para diagnóstico
if cols_nuevas_presentes:
    print("\nNAs por columna nueva:")
    print(df_fd1_v4[cols_nuevas_presentes].isna().sum().sort_values(ascending=False))

=== TRAZABILIDAD EXCLUIDOS (top 10) ===

Edificios únicos en v3 limpio: 566
Edificios únicos en catálogo filtrado: 617
Edificios en intersección: 566

=== RESUMEN UNIÓN v3 limpio + catálogo inmuebles filtrado ===
Filas v3 limpio: 316,871  ->  Filas v4: 316,871
Columnas añadidas detectadas: ['ID_REGION', 'SERVICE_LEVEL', 'TIPO_USO', 'COUNTRY_CATALOGO', 'COUNTRY']
Filas sin match (todas las nuevas columnas NA): 0 (0.00%)

NAs por columna nueva:
ID_REGION           0
SERVICE_LEVEL       0
TIPO_USO            0
COUNTRY_CATALOGO    0
COUNTRY             0
dtype: int64

# Identificamos los valores únicos de la variable COST_TYPE

valores_cost_type = df_fd1_v3["COST_TYPE"].unique()
print("=== Valores únicos de COST_TYPE ===")
print(valores_cost_type)

# (Opcional) Conteo de frecuencia de cada valor
print("\n=== Frecuencia de COST_TYPE ===")
print(df_fd1_v3["COST_TYPE"].value_counts())

=== Valores únicos de COST_TYPE ===
['Gasto' 'Inversión']

=== Frecuencia de COST_TYPE ===
COST_TYPE
Gasto        306775
Inversión     10096
Name: count, dtype: int64

# Separaramos registros de inversión y dejamos solo gastos en df_fd1_v4

# Creamos dataset con registros de inversión (CAPEX)
df_fd1_v4_capex = df_fd1_v4[df_fd1_v4["COST_TYPE"] == "Inversión"].copy()

# Creamos dataset con registros de gasto (OPEX)
df_fd1_v4_opex = df_fd1_v4[df_fd1_v4["COST_TYPE"] == "Gasto"].copy()

# Visualizamos un resumen de la operación
print("=== RESUMEN CAPEX/OPEX EN v4 ===")
print(f"Filas totales en v4: {len(df_fd1_v4):,}")
print(f"Filas detectadas como CAPEX (Inversión): {len(df_fd1_v4_capex):,}")
print(f"Filas restantes (OPEX - Gasto): {len(df_fd1_v4_opex):,}")

# Visualizamos una vista rápida de control de cada dataset
print("\nEjemplo registros CAPEX:")
display(df_fd1_v4_capex.head(3)[["ID_ORDER", "ID_BUILDING", "COST_TYPE", "FM_COST_TYPE", "COST"]])

print("\nEjemplo registros OPEX:")
display(df_fd1_v4_opex.head(3)[["ID_ORDER", "ID_BUILDING", "COST_TYPE", "FM_COST_TYPE", "COST"]])

=== RESUMEN CAPEX/OPEX EN v4 ===
Filas totales en v4: 316,871
Filas detectadas como CAPEX (Inversión): 10,096
Filas restantes (OPEX - Gasto): 306,775

Ejemplo registros CAPEX:

Ejemplo registros OPEX:

# Exportamos al Drive df_fd1_v4 (total), df_fd1_v4_capex (Inversión) y df_fd1_v4_opex (Gasto)
# en formato Excel (.xlsx) y CSV con separador ';'

# Definimos ruta de salida
ruta_out = "/content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO"

# Guardamos la versión completa (antes de separar CAPEX/OPEX)
df_fd1_v4.to_excel(f"{ruta_out}/df_fd1_v4_total.xlsx", index=False)
df_fd1_v4.to_csv(f"{ruta_out}/df_fd1_v4_total.csv", sep=";", index=False)

# Guardamos la versión CAPEX (Inversión)
df_fd1_v4_capex.to_excel(f"{ruta_out}/df_fd1_v4_capex.xlsx", index=False)
df_fd1_v4_capex.to_csv(f"{ruta_out}/df_fd1_v4_capex.csv", sep=";", index=False)

# Guardamos la versión OPEX (Gasto) → la que será la oficial para modelado
df_fd1_v4_opex.to_excel(f"{ruta_out}/df_fd1_v4_opex.xlsx", index=False)
df_fd1_v4_opex.to_csv(f"{ruta_out}/df_fd1_v4_opex.csv", sep=";", index=False)

print("=== ARCHIVOS GUARDADOS ===")
print("1) df_fd1_v4_total.xlsx / .csv")
print("2) df_fd1_v4_capex.xlsx / .csv")
print("3) df_fd1_v4_opex.xlsx / .csv")
print(f"Ubicación: {ruta_out}")

=== ARCHIVOS GUARDADOS ===
1) df_fd1_v4_total.xlsx / .csv
2) df_fd1_v4_capex.xlsx / .csv
3) df_fd1_v4_opex.xlsx / .csv
Ubicación: /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO

# Cargamos el dataset v4_opex desde los archivos guardados

ruta_out = "/content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO"

# cargamos desde Excel
df_fd1_v4_opex = pd.read_excel(f"{ruta_out}/df_fd1_v4_opex.xlsx")

# cargamos desde CSV (usando ; como separador)
# df_fd1_v4_opex = pd.read_csv(f"{ruta_out}/df_fd1_v4_opex.csv", sep=";")

print("=== Dataset recargado ===")
print(f"Filas: {len(df_fd1_v4_opex):,} | Columnas: {len(df_fd1_v4_opex.columns):,}")
print("Columnas:", df_fd1_v4_opex.columns.tolist()[:15], "...")

=== Dataset recargado ===
Filas: 306,775 | Columnas: 27
Columnas: ['ID_ORDER', 'COST_TYPE', 'COUNTRY', 'ID_BUILDING', 'YEAR_MONTH', 'FM_RESPONSIBLE', 'FM_COST_TYPE', 'COST', 'ID_CUSTOMER', 'ID_SUPPLIER', 'ID_BUSINESS_UNIT', 'SUPPLIER_TYPE', 'MONTH', 'YEAR', 'QUARTER'] ...

# Categorizamos todas las columnas según su naturaleza y el valor aportado


# Listamos todas las columnas
todas_cols = df_fd1_v4_opex.columns.tolist()
print("=== LISTA DE TODAS LAS VARIABLES ===")
print(todas_cols)

# Definimos la dependiente (target = cost_float_mod)
var_dependiente = "cost_float_mod" if "cost_float_mod" in df_fd1_v4_opex.columns else None

# Definimos las variables identificadoras (claves y códigos que no deben usarse como predictores directos)
vars_id = [
    "ID_ORDER", "ID_BUILDING", "ID_CUSTOMER", "ID_SUPPLIER",
    "ID_BUSINESS_UNIT", "ID_BUSINESS_UNIT_MOD", "CONCAT_key"
]
vars_id = [v for v in vars_id if v in df_fd1_v4_opex.columns]

# Definimos las variables temporales
vars_tiempo = [c for c in ["YEAR_MONTH", "YEAR", "MONTH", "QUARTER"] if c in df_fd1_v4_opex.columns]

# Definimos las variables de coste
vars_coste = [c for c in ["COST", "cost_float", "cost_float_mod"] if c in df_fd1_v4_opex.columns]

# Definimos las variables categóricas basadas en dtype (object)
vars_categoricas = df_fd1_v4_opex.select_dtypes(include=["object"]).columns.tolist()

# Forzamos ID_REGION como categórica
if "ID_REGION" in df_fd1_v4_opex.columns and "ID_REGION" not in vars_categoricas:
    vars_categoricas.append("ID_REGION")
    # Convertimos el dtype a 'category' para evitar que entre en análisis numérico
    df_fd1_v4_opex["ID_REGION"] = df_fd1_v4_opex["ID_REGION"].astype("category")

# Variables numéricas restantes (posibles features cuantitativos)
vars_numericas = [
    c for c in df_fd1_v4_opex.select_dtypes(include=["int64", "float64"]).columns.tolist()
    if c not in vars_coste + vars_tiempo + vars_id
]


# Visualizamos un resumen de clasificación
print("\n=== CLASIFICACIÓN DE VARIABLES ===")
print(f"Dependiente (target): {var_dependiente}")
print(f"Identificadores: {vars_id}")
print(f"Temporales: {vars_tiempo}")
print(f"Variables de coste: {vars_coste}")
print(f"Categóricas: {vars_categoricas}")
print(f"Numéricas (otras): {vars_numericas}")

=== LISTA DE TODAS LAS VARIABLES ===
['ID_ORDER', 'COST_TYPE', 'COUNTRY', 'ID_BUILDING', 'YEAR_MONTH', 'FM_RESPONSIBLE', 'FM_COST_TYPE', 'COST', 'ID_CUSTOMER', 'ID_SUPPLIER', 'ID_BUSINESS_UNIT', 'SUPPLIER_TYPE', 'MONTH', 'YEAR', 'QUARTER', 'cost_float', 'SUPPLIER_TYPE_MOD', 'ID_BUSINESS_UNIT_MOD', 'bu_corr_flag', 'CONCAT_key', 'cost_float_mod', 'SUPPLIER_TYPE_MOD_2', 'FM_RESPONSIBLE_MOD', 'ID_REGION', 'SERVICE_LEVEL', 'TIPO_USO', 'COUNTRY_CATALOGO']

=== CLASIFICACIÓN DE VARIABLES ===
Dependiente (target): cost_float_mod
Identificadores: ['ID_ORDER', 'ID_BUILDING', 'ID_CUSTOMER', 'ID_SUPPLIER', 'ID_BUSINESS_UNIT', 'ID_BUSINESS_UNIT_MOD', 'CONCAT_key']
Temporales: ['YEAR_MONTH', 'YEAR', 'MONTH', 'QUARTER']
Variables de coste: ['COST', 'cost_float', 'cost_float_mod']
Categóricas: ['ID_ORDER', 'COST_TYPE', 'COUNTRY', 'FM_RESPONSIBLE', 'FM_COST_TYPE', 'SUPPLIER_TYPE', 'SUPPLIER_TYPE_MOD', 'CONCAT_key', 'SUPPLIER_TYPE_MOD_2', 'FM_RESPONSIBLE_MOD', 'SERVICE_LEVEL', 'TIPO_USO', 'COUNTRY_CATALOGO', 'ID_REGION']
Numéricas (otras): []

# === DICCIONARIO DE VARIABLES v5 ===

vars_v5 = {
    "keep": [
        "ID_ORDER"          # despues de la AED la eliminaremos
        "COUNTRY",
        "ID_BUILDING",
        "FM_COST_TYPE",
        "MONTH",
        "YEAR",
        "cost_float_mod",
        "SUPPLIER_TYPE_MOD_2",
        "FM_RESPONSIBLE_MOD",
        "ID_REGION",
        "TIPO_USO",
        "COUNTRY_CATALOGO"  # pendiente revisar si coincide
    ],
    "drop": [
        "COST_TYPE",
        "YEAR_MONTH",
        "FM_RESPONSIBLE",
        "COST",
        "SUPPLIER_TYPE",
        "SUPPLIER_TYPE_MOD",
        "ID_BUSINESS_UNIT",
        "bu_corr_flag",
        "CONCAT_key",
        "cost_float",
        "QUARTER",
        "ID_CUSTOMER",
        "ID_SUPPLIER",
        "ID_BUSINESS_UNIT_MOD",
        "SERVICE_LEVEL",
    ]
}

print("=== Diccionario de variables v5 ===")
print(vars_v5)

=== Diccionario de variables v5 ===
{'keep': ['ID_ORDERCOUNTRY', 'ID_BUILDING', 'FM_COST_TYPE', 'MONTH', 'YEAR', 'cost_float_mod', 'SUPPLIER_TYPE_MOD_2', 'FM_RESPONSIBLE_MOD', 'ID_REGION', 'TIPO_USO', 'COUNTRY_CATALOGO'], 'drop': ['COST_TYPE', 'YEAR_MONTH', 'FM_RESPONSIBLE', 'COST', 'SUPPLIER_TYPE', 'SUPPLIER_TYPE_MOD', 'ID_BUSINESS_UNIT', 'bu_corr_flag', 'CONCAT_key', 'cost_float', 'QUARTER', 'ID_CUSTOMER', 'ID_SUPPLIER', 'ID_BUSINESS_UNIT_MOD', 'SERVICE_LEVEL']}

# Detectamos las discrepancias entre COUNTRY y COUNTRY_CATALOGO, las listamos
# y las exportamos a Drive en formato Excel

# Verificamos que existan las columnas requeridas
cols_req = ["COUNTRY", "COUNTRY_CATALOGO", "ID_BUILDING"]
faltantes = [c for c in cols_req if c not in df_fd1_v4_opex.columns]
if faltantes:
    raise KeyError(f"Faltan columnas en df_fd1_v4_opex: {faltantes}")

# Creamos columnas normalizadas para comparar (sin espacios y en mayúsculas)
country_norm = df_fd1_v4_opex["COUNTRY"].astype(str).str.strip().str.upper()
country_cat_norm = df_fd1_v4_opex["COUNTRY_CATALOGO"].astype(str).str.strip().str.upper()

# Generamos la máscara de diferencias
mask_diff = country_norm != country_cat_norm

# Construimos el dataset de diferencias con columnas clave
cols_salida = ["ID_BUILDING", "COUNTRY", "COUNTRY_CATALOGO"]
if "ID_ORDER" in df_fd1_v4_opex.columns:
    cols_salida = ["ID_BUILDING", "ID_ORDER", "COUNTRY", "COUNTRY_CATALOGO", "FM_RESPONSIBLE_MOD"]

df_diff_country = df_fd1_v4_opex.loc[mask_diff, cols_salida].copy()

# Visualizamos un resumen por pantalla
total = len(df_fd1_v4_opex)
num_diff = len(df_diff_country)
print("=== COMPARACIÓN COUNTRY vs COUNTRY_CATALOGO ===")
print(f"Total registros: {total:,}")
print(f"Diferencias detectadas: {num_diff:,} ({(100*num_diff/total):.2f}%)")

=== COMPARACIÓN COUNTRY vs COUNTRY_CATALOGO ===
Total registros: 306,775
Diferencias detectadas: 2,248 (0.73%)

# Exportamos a Excel
ruta_out = "/content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO"
ruta_excel = f"{ruta_out}/df_v4_opex_diferencias_COUNTRY_vs_CATALOGO.xlsx"

# Exportamos usando openpyxl
with pd.ExcelWriter(ruta_excel, engine="openpyxl") as writer:
    df_diff_country.to_excel(writer, index=False, sheet_name="diferencias")

print(f"Archivo exportado correctamente a: {ruta_excel}")

# Eliminamos las variables redundantes y nos quedarnos con la estructura
# final definida

# Creamos df_fd1_v5 eliminando las columnas en vars_v5["drop"]
df_fd1_v5 = df_fd1_v4_opex.drop(columns=vars_v5["drop"], errors="ignore").copy()

# Verificamos que mantenemos todas las columnas clave de vars_v5["keep"]
cols_presentes = [c for c in vars_v5["keep"] if c in df_fd1_v5.columns]
cols_faltantes = [c for c in vars_v5["keep"] if c not in df_fd1_v5.columns]

print("=== RESUMEN CREACIÓN v5 ===")
print(f"Filas: {len(df_fd1_v5):,}")
print(f"Columnas: {len(df_fd1_v5.columns):,}")
print("\nColumnas presentes (keep):", cols_presentes)
print("Columnas faltantes (keep):", cols_faltantes)

# Visualizamos el dataset resultante
display(df_fd1_v5.head(5))

=== RESUMEN CREACIÓN v5 ===
Filas: 306,775
Columnas: 12

Columnas presentes (keep): ['ID_BUILDING', 'FM_COST_TYPE', 'MONTH', 'YEAR', 'cost_float_mod', 'SUPPLIER_TYPE_MOD_2', 'FM_RESPONSIBLE_MOD', 'ID_REGION', 'TIPO_USO', 'COUNTRY_CATALOGO']
Columnas faltantes (keep): ['ID_ORDERCOUNTRY']

# Analizar cost_float_mod y revisar categorías de las variables categóricas

# Definimos la variable dependiente (numérica)
var_target = "cost_float_mod"

print("=== Estadísticos de cost_float_mod ===")
display(df_fd1_v5[var_target].describe(percentiles=[0.05, 0.25, 0.5, 0.75, 0.95]).to_frame())

# Boxplot para cost_float_mod
plt.figure(figsize=(8, 3))
plt.boxplot(df_fd1_v5[var_target].dropna(), vert=False)
plt.title("Boxplot de cost_float_mod")
plt.xlabel(var_target)
plt.show()

# Variables categóricas a analizar
vars_categoricas = [
    "COUNTRY", "FM_COST_TYPE", "SUPPLIER_TYPE_MOD_2", "FM_RESPONSIBLE_MOD",
    "ID_REGION", "TIPO_USO", "COUNTRY_CATALOGO"
]
vars_categoricas = [c for c in vars_categoricas if c in df_fd1_v5.columns]

print("\n=== ANÁLISIS DE VARIABLES CATEGÓRICAS ===")
for col in vars_categoricas:
    print(f"\n--- {col} ---")
    print(f"Nº de categorías: {df_fd1_v5[col].nunique()}")
    print(df_fd1_v5[col].value_counts(dropna=False).head(10))  # top 10 categorías más frecuentes

=== Estadísticos de cost_float_mod ===

=== ANÁLISIS DE VARIABLES CATEGÓRICAS ===

--- COUNTRY ---
Nº de categorías: 7
COUNTRY
España                  115099
Colombia                 74909
Perú                     34859
Panamá                   29055
México                   28200
Costa Rica               14172
República Dominicana     10481
Name: count, dtype: int64

--- FM_COST_TYPE ---
Nº de categorías: 8
FM_COST_TYPE
Mtto. Correctivo         176085
Suministros               42040
Servicios Extra           35263
Servicios Ctto.           27345
Mtto. Contratos           15766
Licencias                  4510
Obras                      4031
Eficiencia Energética      1735
Name: count, dtype: int64

--- SUPPLIER_TYPE_MOD_2 ---
Nº de categorías: 2
SUPPLIER_TYPE_MOD_2
EXTERNO    199957
INTERNO    106818
Name: count, dtype: int64

--- FM_RESPONSIBLE_MOD ---
Nº de categorías: 5
FM_RESPONSIBLE_MOD
Mantenimiento            252512
Eficiencia Energética     44441
Licencias                  5596
Obras Proyectos            2380
Gestión Espacios           1846
Name: count, dtype: int64

--- ID_REGION ---
Nº de categorías: 104
ID_REGION
2          26578
1000022    25459
103        23785
94         23041
105        19484
8          16130
117        12449
17         11958
32          8727
1000039     8725
Name: count, dtype: int64

--- TIPO_USO ---
Nº de categorías: 14
TIPO_USO
Casino Electrónico    106781
Casino Tradicional    104602
Salón de Juego         35625
Bingo                  25517
Oficinas               14727
Delegación             10037
Industrial              3385
Almacén/Bodega          2332
Hotel                   1967
Sin actividad            655
Name: count, dtype: int64

--- COUNTRY_CATALOGO ---
Nº de categorías: 7
COUNTRY_CATALOGO
España                  114607
Colombia                 75141
Perú                     34859
Panamá                   30811
México                   28460
Costa Rica               14172
República Dominicana      8725
Name: count, dtype: int64

# Calculamos frecuencia y media de cost_float_mod por región

# Agrupamos por ID_REGION
df_region_stats = (
    df_fd1_v5.groupby("ID_REGION")
    .agg(
        num_registros=("cost_float_mod", "size"),
        media_coste=("cost_float_mod", "mean")
    )
    .reset_index()
    .sort_values("num_registros", ascending=False)
)

# Visualizamos el resumen general
print("=== RESUMEN ID_REGION ===")
print(f"Nº de regiones distintas: {df_region_stats.shape[0]}")
print(f"Registros totales: {df_region_stats['num_registros'].sum():,}")

# Visualizamos el dataset
display(df_region_stats.head(15))

# Calculamos los estadísticos de frecuencia de regiones
print("\n=== DISTRIBUCIÓN DE FRECUENCIAS (nº de registros por región) ===")
print(df_region_stats["num_registros"].describe(percentiles=[0.05, 0.25, 0.5, 0.75, 0.95]))

=== RESUMEN ID_REGION ===
Nº de regiones distintas: 104
Registros totales: 306,775

/tmp/ipython-input-1167182373.py:5: FutureWarning: The default of observed=False is deprecated and will be changed to True in a future version of pandas. Pass observed=False to retain current behavior or observed=True to adopt the future default and silence this warning.
  df_fd1_v5.groupby("ID_REGION")

=== DISTRIBUCIÓN DE FRECUENCIAS (nº de registros por región) ===
count      104.000000
mean      2949.759615
std       5406.722998
min         11.000000
5%         177.150000
25%        538.000000
50%        958.000000
75%       2352.000000
95%      15577.850000
max      26578.000000
Name: num_registros, dtype: float64

# Creamos ID_REGION_GRUPO, donde las regiones con pocos registros de un mismo pais (< umbral) se agrupan como "Otros_{COUNTRY}"

# Definimos el umbral de frecuencia mínima por país
umbral = 1000  # ajustamos de 500 a 1000 para conseguir una cardinalidad aceptable (< 55 para OHE)

# Calculamos la frecuencia por (COUNTRY, ID_REGION)
freq_por_pais_region = (
    df_fd1_v5
    .groupby(["COUNTRY", "ID_REGION"], dropna=False)
    .size()
    .reset_index(name="n")
)

# Creamos un diccionario de frecuencias para consulta rápida
# Clave: (COUNTRY, ID_REGION) -> Valor: n
freq_map = {(row["COUNTRY"], row["ID_REGION"]): row["n"] for _, row in freq_por_pais_region.iterrows()}

# Función para asignar el grupo
def asignar_grupo(country, id_region):
    # Si falta el país o la región, devolvemos NA de forma segura
    if pd.isna(country) or pd.isna(id_region):
        return np.nan

    # Buscamos en el diccionario 'freq_map' cuántos registros hay para esa combinación (COUNTRY, ID_REGION).
    # Si no está, devolvemos 0 por defecto.
    n = freq_map.get((country, id_region), 0)

    # Si el número de registros de esa región está por debajo del umbral definido,
    # agrupamos esa región dentro de una categoría "Otros_{COUNTRY}".
    if n < umbral:
        return f"Otros_{country}"
    else:
        # Si está por encima del umbral, dejamos la región como está.
        return str(id_region)


# Creamos la nueva columna ID_REGION_GRUPO
df_fd1_v5["ID_REGION_GRUPO"] = df_fd1_v5.apply(
    lambda r: asignar_grupo(r["COUNTRY"], r["ID_REGION"]),
    axis=1
).astype("category")

# Visualizamos el resumen de la operación
num_regiones = freq_por_pais_region.shape[0]
num_regiones_poco_repr = (freq_por_pais_region["n"] < umbral).sum()
num_labels_finales = df_fd1_v5["ID_REGION_GRUPO"].nunique()

print("=== AGRUPACIÓN DE REGIONES POR PAÍS ===")
print(f"Umbral: {umbral} registros por (COUNTRY, ID_REGION)")
print(f"Total de pares (COUNTRY, ID_REGION): {num_regiones}")
print(f"Pares por debajo del umbral que agrupamos: {num_regiones_poco_repr}")
print(f"Nº de etiquetas finales en ID_REGION_GRUPO: {num_labels_finales}")

# Visualizamos las etiquetas más frecuentes
print("\n=== FRECUENCIAS DE ID_REGION_GRUPO (top 15) ===")
print(df_fd1_v5["ID_REGION_GRUPO"].value_counts().head(15))

# Definimos una tabla de referencia de cómo hemos agrupado
tabla_grupos = (
    df_fd1_v5
    .groupby(["COUNTRY", "ID_REGION", "ID_REGION_GRUPO"], dropna=False)
    .size()
    .reset_index(name="n_registros")
    .sort_values(["COUNTRY", "ID_REGION_GRUPO", "n_registros"], ascending=[True, True, False])
)
display(tabla_grupos.head(20))

/tmp/ipython-input-626660888.py:9: FutureWarning: The default of observed=False is deprecated and will be changed to True in a future version of pandas. Pass observed=False to retain current behavior or observed=True to adopt the future default and silence this warning.
  .groupby(["COUNTRY", "ID_REGION"], dropna=False)

=== AGRUPACIÓN DE REGIONES POR PAÍS ===
Umbral: 1000 registros por (COUNTRY, ID_REGION)
Total de pares (COUNTRY, ID_REGION): 728
Pares por debajo del umbral que agrupamos: 678
Nº de etiquetas finales en ID_REGION_GRUPO: 55

=== FRECUENCIAS DE ID_REGION_GRUPO (top 15) ===
ID_REGION_GRUPO
2               26578
1000022         25459
103             23785
94              23041
105             19484
8               16130
Otros_España    13505
117             12449
17              11958
Otros_México     8832
32               8727
1000039          8725
3                7571
1000005          6830
1000023          6027
Name: count, dtype: int64

/tmp/ipython-input-626660888.py:61: FutureWarning: The default of observed=False is deprecated and will be changed to True in a future version of pandas. Pass observed=False to retain current behavior or observed=True to adopt the future default and silence this warning.
  .groupby(["COUNTRY", "ID_REGION", "ID_REGION_GRUPO"], dropna=False)

# Guardamos la tabla de referencia con correspondencias por país
# y un resumen con el nº de registros por grupo

# Definimos ruta de salida
ruta_out = "/content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO"
ruta_excel_tabla = f"{ruta_out}/v5_id_region_grupos.xlsx"

# Tabla de correspondencias ya calculada previamente -> tabla_grupos
# Contiene: COUNTRY, ID_REGION, ID_REGION_GRUPO, n_registros

# Generamos el resumen de frecuencias por grupo
resumen_grupo = (
    df_fd1_v5["ID_REGION_GRUPO"]
    .value_counts(dropna=False)
    .rename_axis("ID_REGION_GRUPO")
    .reset_index(name="n_registros")
)

# Exportamos ambos a un archivo Excel con dos hojas
with pd.ExcelWriter(ruta_excel_tabla, engine="openpyxl") as writer:
    tabla_grupos.to_excel(writer, index=False, sheet_name="agrupacion")
    resumen_grupo.to_excel(writer, index=False, sheet_name="resumen_grupos")

print("=== EXPORTACIÓN COMPLETA ===")
print(f"Archivo generado: {ruta_excel_tabla}")
print(f"Filas en tabla de correspondencias: {len(tabla_grupos):,}")
print(f"Filas en resumen de grupos: {len(resumen_grupo):,}")

=== EXPORTACIÓN COMPLETA ===
Archivo generado: /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/v5_id_region_grupos.xlsx
Filas en tabla de correspondencias: 40,040
Filas en resumen de grupos: 55

# Añadimos ID_REGION_GRUPO y eliminamos la ID_REGION original


# Verificamos que ambas columnas existan
if "ID_REGION_GRUPO" not in df_fd1_v5.columns:
    raise KeyError("No existe la columna 'ID_REGION_GRUPO'. Debemos recalcularla antes.")
if "ID_REGION" not in df_fd1_v5.columns:
    raise KeyError("No existe la columna original 'ID_REGION'. Ya podría haberse eliminado.")

# Eliminamos la columna original ID_REGION
df_fd1_v5 = df_fd1_v5.drop(columns=["ID_REGION"])

# Nos aseguramos de que ID_REGION_GRUPO sea categórica
df_fd1_v5["ID_REGION_GRUPO"] = df_fd1_v5["ID_REGION_GRUPO"].astype("category")

# Visualizamos el resumen de verificación
print("=== RESUMEN DE ID_REGION_GRUPO ===")
print(f"Columnas actuales: {df_fd1_v5.columns.tolist()}")
print(f"Nº de clases en ID_REGION_GRUPO: {df_fd1_v5['ID_REGION_GRUPO'].nunique()}")
print("\nTop 15 categorías por frecuencia:")
print(df_fd1_v5['ID_REGION_GRUPO'].value_counts().head(15))

=== RESUMEN DE ID_REGION_GRUPO ===
Columnas actuales: ['ID_ORDER', 'COUNTRY', 'ID_BUILDING', 'FM_COST_TYPE', 'MONTH', 'YEAR', 'cost_float_mod', 'SUPPLIER_TYPE_MOD_2', 'FM_RESPONSIBLE_MOD', 'TIPO_USO', 'COUNTRY_CATALOGO', 'ID_REGION_GRUPO']
Nº de clases en ID_REGION_GRUPO: 55

Top 15 categorías por frecuencia:
ID_REGION_GRUPO
2               26578
1000022         25459
103             23785
94              23041
105             19484
8               16130
Otros_España    13505
117             12449
17              11958
Otros_México     8832
32               8727
1000039          8725
3                7571
1000005          6830
1000023          6027
Name: count, dtype: int64

# === DETECCIÓN DE OUTLIERS EN cost_float_mod ===

# Calculamos cuartiles e IQR
Q1 = df_fd1_v5["cost_float_mod"].quantile(0.25)
Q3 = df_fd1_v5["cost_float_mod"].quantile(0.75)
IQR = Q3 - Q1

limite_inferior = Q1 - 1.5 * IQR
limite_superior = Q3 + 1.5 * IQR

print("=== UMBRALES DE OUTLIERS (IQR) ===")
print(f"Q1 (25%): {Q1:.2f}")
print(f"Q3 (75%): {Q3:.2f}")
print(f"IQR: {IQR:.2f}")
print(f"Límite inferior: {limite_inferior:.2f}")
print(f"Límite superior: {limite_superior:.2f}")

# Creamos máscara de outliers
mask_outliers = (df_fd1_v5["cost_float_mod"] < limite_inferior) | (df_fd1_v5["cost_float_mod"] > limite_superior)

# Visualizamos el resumen
num_outliers = mask_outliers.sum()
total = len(df_fd1_v5)
print(f"\nOutliers detectados: {num_outliers:,} ({100*num_outliers/total:.2f}%) de {total:,} registros")

# Generamos el dataset de outliers
df_outliers = df_fd1_v5.loc[mask_outliers, ["ID_ORDER", "ID_BUILDING",
                                            "FM_RESPONSIBLE_MOD", "FM_COST_TYPE",
                                            "COUNTRY", "ID_REGION_GRUPO", "TIPO_USO",
                                            "cost_float_mod"]].copy()

# Vista rápida de los valores extremos
print("\nEjemplos de outliers:")
display(df_outliers.sort_values("cost_float_mod", ascending=False).head(10))

=== UMBRALES DE OUTLIERS (IQR) ===
Q1 (25%): 0.00
Q3 (75%): 233.31
IQR: 233.31
Límite inferior: -349.96
Límite superior: 583.26

Outliers detectados: 42,174 (13.75%) de 306,775 registros

Ejemplos de outliers:

# Definimos ruta de salida
ruta_out = "/content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO"
ruta_excel_outliers = f"{ruta_out}/df_v5_outliers_cost_float_mod.xlsx"

# Exportamos a Excel
with pd.ExcelWriter(ruta_excel_outliers, engine="openpyxl") as writer:
    df_outliers.to_excel(writer, index=False, sheet_name="outliers")

print("=== EXPORTACIÓN COMPLETA ===")
print(f"Archivo generado: {ruta_excel_outliers}")
print(f"Total de outliers exportados: {len(df_outliers):,}")

=== EXPORTACIÓN COMPLETA ===
Archivo generado: /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/df_v5_outliers_cost_float_mod.xlsx
Total de outliers exportados: 42,174

# Comprobamos si cada ID_ORDER aparece una sola vez en df_fd1_v5

# Calculamos el número de veces que aparece cada ID_ORDER
conteo_id_order = df_fd1_v5["ID_ORDER"].value_counts()

# Visualizamos un resumen
total_orders = len(df_fd1_v5["ID_ORDER"].unique())
duplicados = (conteo_id_order > 1).sum()

print("=== ANÁLISIS DE ID_ORDER EN v5 ===")
print(f"Total de ID_ORDER distintos: {total_orders:,}")
print(f"ID_ORDER con más de una aparición: {duplicados:,}")

# Visualizamos los los más repetidos
print("\n=== Ejemplos de ID_ORDER repetidos ===")
display(conteo_id_order[conteo_id_order > 1].head(20))

=== ANÁLISIS DE ID_ORDER EN v5 ===
Total de ID_ORDER distintos: 306,775
ID_ORDER con más de una aparición: 0

=== Ejemplos de ID_ORDER repetidos ===

# Obtenemos estadísticas generales de todas las variables en df_fd1_v5

print("=== DESCRIBE DE VARIABLES NUMÉRICAS ===")
display(df_fd1_v5.describe().T)

print("\n=== DESCRIBE DE TODAS LAS VARIABLES (incluye categóricas) ===")
display(df_fd1_v5.describe(include="all").T)

print("\n=== INFO DEL DATAFRAME ===")
print(df_fd1_v5.info())

print("\n=== NULOS POR COLUMNA ===")
print(df_fd1_v5.isna().sum())

=== DESCRIBE DE VARIABLES NUMÉRICAS ===

=== DESCRIBE DE TODAS LAS VARIABLES (incluye categóricas) ===

=== INFO DEL DATAFRAME ===
<class 'pandas.core.frame.DataFrame'>
RangeIndex: 306775 entries, 0 to 306774
Data columns (total 12 columns):
 #   Column               Non-Null Count   Dtype
---  ------               --------------   -----
 0   ID_ORDER             306775 non-null  object
 1   COUNTRY              306775 non-null  object
 2   ID_BUILDING          306775 non-null  int64
 3   FM_COST_TYPE         306775 non-null  object
 4   MONTH                306775 non-null  int64
 5   YEAR                 306775 non-null  int64
 6   cost_float_mod       306775 non-null  float64
 7   SUPPLIER_TYPE_MOD_2  306775 non-null  object
 8   FM_RESPONSIBLE_MOD   306775 non-null  object
 9   TIPO_USO             306775 non-null  object
 10  COUNTRY_CATALOGO     306775 non-null  object
 11  ID_REGION_GRUPO      306775 non-null  category
dtypes: category(1), float64(1), int64(3), object(7)
memory usage: 26.0+ MB
None

=== NULOS POR COLUMNA ===
ID_ORDER               0
COUNTRY                0
ID_BUILDING            0
FM_COST_TYPE           0
MONTH                  0
YEAR                   0
cost_float_mod         0
SUPPLIER_TYPE_MOD_2    0
FM_RESPONSIBLE_MOD     0
TIPO_USO               0
COUNTRY_CATALOGO       0
ID_REGION_GRUPO        0
dtype: int64

# Guardamos la versión 5 del dataset preparada para la segregación de iteraciones

# Definimos ruta de salida
ruta_out = "/content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO"

# Exportamos en Excel
ruta_excel = f"{ruta_out}/df_fd1_v5.xlsx"
df_fd1_v5.to_excel(ruta_excel, index=False)

# Exportamos en CSV con separador ';'
ruta_csv = f"{ruta_out}/df_fd1_v5.csv"
df_fd1_v5.to_csv(ruta_csv, sep=";", index=False)

print("=== EXPORTACIÓN COMPLETA ===")
print(f"Archivo Excel: {ruta_excel}")
print(f"Archivo CSV  : {ruta_csv}")
print(f"Filas: {len(df_fd1_v5):,} | Columnas: {len(df_fd1_v5.columns):,}")

=== EXPORTACIÓN COMPLETA ===
Archivo Excel: /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/df_fd1_v5.xlsx
Archivo CSV  : /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/df_fd1_v5.csv
Filas: 306,775 | Columnas: 12

# Recuperamos el dataset v5 guardado previamente

ruta_out = "/content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO"

# Cargamos desde Excel
df_fd1_v5 = pd.read_excel(f"{ruta_out}/df_fd1_v5.xlsx")

# Alternativa: cargar desde CSV con separador ';'
# df_fd1_v5 = pd.read_csv(f"{ruta_out}/df_fd1_v5.csv", sep=";")

print("=== DATASET CARGADO ===")
print(f"Filas: {len(df_fd1_v5):,} | Columnas: {len(df_fd1_v5.columns):,}")
print("Columnas:", df_fd1_v5.columns.tolist())

=== DATASET CARGADO ===
Filas: 306,775 | Columnas: 12
Columnas: ['ID_ORDER', 'COUNTRY', 'ID_BUILDING', 'FM_COST_TYPE', 'MONTH', 'YEAR', 'cost_float_mod', 'SUPPLIER_TYPE_MOD_2', 'FM_RESPONSIBLE_MOD', 'TIPO_USO', 'COUNTRY_CATALOGO', 'ID_REGION_GRUPO']

# Quedarnos con COUNTRY o COUNTRY_CATALOGO (según decisión) y
# renombrarla a COUNTRY_DEF.

# Definimos los parámetros de decisión
#    - Establecemos cuál preferimos conservar: "COUNTRY" o "COUNTRY_CATALOGO"
prefer_pais = "COUNTRY_CATALOGO"  # cambiaremos a "COUNTRY" si decidimos usar la de la v3.
#    - Indicamos si queremos eliminar la columna alternativa una vez creada COUNTRY_DEF
drop_columna_alternativa = True

# Verificamos qué columnas están disponibles
cols_pais = [c for c in ["COUNTRY", "COUNTRY_CATALOGO"] if c in df_fd1_v5.columns]
if not cols_pais:
    raise KeyError("No encontramos ni 'COUNTRY' ni 'COUNTRY_CATALOGO' en df_fd1_v5.")

# Determinamos la columna a conservar como COUNTRY_DEF
if prefer_pais in cols_pais:
    col_keep = prefer_pais
else:
    # Si la preferida no existe, usamos la disponible
    col_keep = cols_pais[0]

# Creamos/renombramos COUNTRY_DEF a partir de la columna elegida
df_fd1_v5["COUNTRY_DEF"] = df_fd1_v5[col_keep]

# Eliminamos la columna alternativa si procede
col_alternativa = None
for c in ["COUNTRY", "COUNTRY_CATALOGO"]:
    if c in df_fd1_v5.columns and c != col_keep:
        col_alternativa = c
        break

if drop_columna_alternativa and col_alternativa is not None:
    df_fd1_v5.drop(columns=[col_alternativa], inplace=True)

# Convertimos a categórica para dejar clara su naturaleza
df_fd1_v5["COUNTRY_DEF"] = df_fd1_v5["COUNTRY_DEF"].astype("category")

# Visualización de resumen
print("=== NORMALIZACIÓN DE PAÍS ===")
print(f"Preferencia solicitada: {prefer_pais}")
print(f"Columna utilizada para COUNTRY_DEF: {col_keep}")
if col_alternativa:
    print(f"Columna alternativa detectada: {col_alternativa} | Eliminada: {drop_columna_alternativa}")
print("Valores únicos en COUNTRY_DEF:", df_fd1_v5["COUNTRY_DEF"].cat.categories.tolist())

=== NORMALIZACIÓN DE PAÍS ===
Preferencia solicitada: COUNTRY_CATALOGO
Columna utilizada para COUNTRY_DEF: COUNTRY_CATALOGO
Columna alternativa detectada: COUNTRY | Eliminada: True
Valores únicos en COUNTRY_DEF: ['Colombia', 'Costa Rica', 'España', 'México', 'Panamá', 'Perú', 'República Dominicana']

# Creamod df_fd1_v5_ITE1 (2021-2023) y df_fd1_v5_ITE2 (2021-2024)

# Revisamos estructura correcta
if "YEAR" not in df_fd1_v5.columns:
    raise KeyError("No encontramos la columna 'YEAR' en df_fd1_v5.")
# Forzamos YEAR a numérico por si viniese como string
df_fd1_v5["YEAR"] = pd.to_numeric(df_fd1_v5["YEAR"], errors="coerce")

# Definimos rangos de cada iteración
rango_ite1 = [2021, 2022, 2023]
rango_ite2 = [2021, 2022, 2023, 2024]

# Creamos los datasets de entrenamiento por iteración
df_fd1_v5_ITE1 = df_fd1_v5[df_fd1_v5["YEAR"].isin(rango_ite1)].copy()
df_fd1_v5_ITE2 = df_fd1_v5[df_fd1_v5["YEAR"].isin(rango_ite2)].copy()

# Visualizamos resumen de cada iteración
print("=== RESUMEN ITERACIÓN 1 (2021-2023) ===")
print(f"Filas: {len(df_fd1_v5_ITE1):,} | Columnas: {len(df_fd1_v5_ITE1.columns):,}")
print("Años incluidos:", sorted(df_fd1_v5_ITE1['YEAR'].dropna().unique().tolist()))
print(df_fd1_v5_ITE1['YEAR'].value_counts().sort_index())

print("\n=== RESUMEN ITERACIÓN 2 (2021-2024) ===")
print(f"Filas: {len(df_fd1_v5_ITE2):,} | Columnas: {len(df_fd1_v5_ITE2.columns):,}")
print("Años incluidos:", sorted(df_fd1_v5_ITE2['YEAR'].dropna().unique().tolist()))
print(df_fd1_v5_ITE2['YEAR'].value_counts().sort_index())

=== RESUMEN ITERACIÓN 1 (2021-2023) ===
Filas: 174,626 | Columnas: 12
Años incluidos: [2021, 2022, 2023]
YEAR
2021    48106
2022    61034
2023    65486
Name: count, dtype: int64

=== RESUMEN ITERACIÓN 2 (2021-2024) ===
Filas: 243,916 | Columnas: 12
Años incluidos: [2021, 2022, 2023, 2024]
YEAR
2021    48106
2022    61034
2023    65486
2024    69290
Name: count, dtype: int64

# Exportación a Drive
ruta_out = "/content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO"

# ITERACIÓN 1
df_fd1_v5_ITE1.to_excel(f"{ruta_out}/df_fd1_v5_ITE1.xlsx", index=False)
df_fd1_v5_ITE1.to_csv(f"{ruta_out}/df_fd1_v5_ITE1.csv", sep=";", index=False)

# ITERACIÓN 2
df_fd1_v5_ITE2.to_excel(f"{ruta_out}/df_fd1_v5_ITE2.xlsx", index=False)
df_fd1_v5_ITE2.to_csv(f"{ruta_out}/df_fd1_v5_ITE2.csv", sep=";", index=False)

print("\n=== EXPORTACIÓN COMPLETA ===")
print(f"Guardado en: {ruta_out}")
print("- df_fd1_v5_ITE1.xlsx / .csv")
print("- df_fd1_v5_ITE2.xlsx / .csv")

=== EXPORTACIÓN COMPLETA ===
Guardado en: /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO
- df_fd1_v5_ITE1.xlsx / .csv
- df_fd1_v5_ITE2.xlsx / .csv

# Creamos df_fd1_v5_Real_ITE1 (2024) y df_fd1_v5_Real_ITE2 (2025)


# Revisamos estructura correcta
if "YEAR" not in df_fd1_v5.columns:
    raise KeyError("No encontramos la columna 'YEAR' en df_fd1_v5.")
df_fd1_v5["YEAR"] = pd.to_numeric(df_fd1_v5["YEAR"], errors="coerce")

# Real ITE1 → año 2024
df_fd1_v5_Real_ITE1 = df_fd1_v5[df_fd1_v5["YEAR"] == 2024].copy()

# Real ITE2 → año 2025
df_fd1_v5_Real_ITE2 = df_fd1_v5[df_fd1_v5["YEAR"] == 2025].copy()

# Acotamos opcionalmente el año 2025 hasta agosto
acotar_2025_hasta_agosto = True  # ponemos False cuando no queramos acotar
if acotar_2025_hasta_agosto and "MONTH" in df_fd1_v5_Real_ITE2.columns:
    df_fd1_v5_Real_ITE2 = df_fd1_v5_Real_ITE2[pd.to_numeric(df_fd1_v5_Real_ITE2["MONTH"], errors="coerce") <= 8].copy()

# Visualizamos los resúmenes
print("=== RESUMEN Reales ITE1 (2024) ===")
print(f"Filas: {len(df_fd1_v5_Real_ITE1):,} | Columnas: {len(df_fd1_v5_Real_ITE1.columns):,}")
if not df_fd1_v5_Real_ITE1.empty and "MONTH" in df_fd1_v5_Real_ITE1.columns:
    print("Meses incluidos:", sorted(pd.to_numeric(df_fd1_v5_Real_ITE1["MONTH"], errors="coerce").dropna().unique().tolist()))

print("\n=== RESUMEN Reales ITE2 (2025) ===")
print(f"Filas: {len(df_fd1_v5_Real_ITE2):,} | Columnas: {len(df_fd1_v5_Real_ITE2.columns):,}")
if not df_fd1_v5_Real_ITE2.empty and "MONTH" in df_fd1_v5_Real_ITE2.columns:
    print("Meses incluidos:", sorted(pd.to_numeric(df_fd1_v5_Real_ITE2["MONTH"], errors="coerce").dropna().unique().tolist()))

=== RESUMEN Reales ITE1 (2024) ===
Filas: 69,290 | Columnas: 12
Meses incluidos: [1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12]

=== RESUMEN Reales ITE2 (2025) ===
Filas: 57,932 | Columnas: 12
Meses incluidos: [1, 2, 3, 4, 5, 6, 7, 8]

# Exportamos a Drive
ruta_out = "/content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO"

# Conjunto de Reales ITE1 (2024)
df_fd1_v5_Real_ITE1.to_excel(f"{ruta_out}/df_fd1_v5_Real_ITE1_2024.xlsx", index=False)
df_fd1_v5_Real_ITE1.to_csv(f"{ruta_out}/df_fd1_v5_Real_ITE1_2024.csv", sep=";", index=False)

# Conjunto de Reales ITE2 (2025; hasta agosto)
sufijo_ite2 = "_2025_hasta_agosto" if acotar_2025_hasta_agosto else "_2025"
df_fd1_v5_Real_ITE2.to_excel(f"{ruta_out}/df_fd1_v5_Real_ITE2{sufijo_ite2}.xlsx", index=False)
df_fd1_v5_Real_ITE2.to_csv(f"{ruta_out}/df_fd1_v5_Real_ITE2{sufijo_ite2}.csv", sep=";", index=False)

print("\n=== EXPORTACIÓN COMPLETA ===")
print(f"- df_fd1_v5_Real_ITE1_2024.xlsx / .csv")
print(f"- df_fd1_v5_Real_ITE2{sufijo_ite2}.xlsx / .csv")
print(f"Ubicación: {ruta_out}")

=== EXPORTACIÓN COMPLETA ===
- df_fd1_v5_Real_ITE1_2024.xlsx / .csv
- df_fd1_v5_Real_ITE2_2025_hasta_agosto.xlsx / .csv
Ubicación: /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO

# Ruta del notebook .ipynb que vamos a convertir

RUTA_NOTEBOOK = "/content/drive/MyDrive/PROYECTO_NEITH/DOCUMENTOS_PROYECTOS/Neith.ipynb"

# Carpeta de salida y nombre del HTML
OUTPUT_DIR = "/content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO"
OUTPUT_HTML = "Neith_notebook_v5_con_Iteraciones_vs_Reales.html"

# Aseguramos carpeta de salida (ya en cabecera)
# import os
os.makedirs(OUTPUT_DIR, exist_ok=True)

# Instalamos nbconvert (ya en cabecera)
# !pip install -q "nbconvert>=7.0.0"

# Convertimos notebook a HTML y lo guardamos con el nombre indicado en la carpeta de salida
!jupyter nbconvert --to html "$RUTA_NOTEBOOK" --output "$OUTPUT_HTML" --output-dir "$OUTPUT_DIR"

# Verificamos
salida = os.path.join(OUTPUT_DIR, OUTPUT_HTML)
print("Archivo HTML generado en:", salida, "\nExiste:", os.path.exists(salida))

[NbConvertApp] Converting notebook /content/drive/MyDrive/PROYECTO_NEITH/DOCUMENTOS_PROYECTOS/Neith.ipynb to html
[NbConvertApp] WARNING | Alternative text is missing on 1 image(s).
[NbConvertApp] Writing 1133173 bytes to /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/Neith_notebook_v5_con_Iteraciones_vs_Reales.html
Archivo HTML generado en: /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/Neith_notebook_v5_con_Iteraciones_vs_Reales.html
Existe: True

# Cargamos datasets df_fd1_v5_ITE1 y df_fd1_v5_Real_ITE1_2024

# Ruta base donde guardaste los outputs
ruta_base = "/content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/"

# Dataset de entrenamiento (Iteración 1: 2021-2023)
df_fd1_v5_ITE1 = pd.read_csv(ruta_base + "df_fd1_v5_ITE1.csv", sep=";")

# Dataset de reales (Iteración 1: datos del 2024)
df_fd1_v5_Real_ITE1_2024 = pd.read_csv(ruta_base + "df_fd1_v5_Real_ITE1_2024.csv", sep=";")

# Confirmamos tamaños y columnas
print("ITE1 shape:", df_fd1_v5_ITE1.shape)
print("Real_ITE1_2024 shape:", df_fd1_v5_Real_ITE1_2024.shape)
print("\nColumnas ITE1:", df_fd1_v5_ITE1.columns.tolist())
print("Columnas Real_ITE1_2024:", df_fd1_v5_Real_ITE1_2024.columns.tolist())

ITE1 shape: (174626, 12)
Real_ITE1_2024 shape: (69290, 12)

Columnas ITE1: ['ID_ORDER', 'ID_BUILDING', 'FM_COST_TYPE', 'MONTH', 'YEAR', 'cost_float_mod', 'SUPPLIER_TYPE_MOD_2', 'FM_RESPONSIBLE_MOD', 'TIPO_USO', 'COUNTRY_CATALOGO', 'ID_REGION_GRUPO', 'COUNTRY_DEF']
Columnas Real_ITE1_2024: ['ID_ORDER', 'ID_BUILDING', 'FM_COST_TYPE', 'MONTH', 'YEAR', 'cost_float_mod', 'SUPPLIER_TYPE_MOD_2', 'FM_RESPONSIBLE_MOD', 'TIPO_USO', 'COUNTRY_CATALOGO', 'ID_REGION_GRUPO', 'COUNTRY_DEF']

# Exploración de ambos datasets - verificación que son correctos

# Exploración rápida de df_fd1_v5_ITE1
print("=== Dataset de Entrenamiento (ITE1) ===")
print(df_fd1_v5_ITE1.info())
print("\nPrimeras filas:")
display(df_fd1_v5_ITE1.head())
print("\nDescripción estadística:")
display(df_fd1_v5_ITE1.describe(include='all'))

# Exploración rápida de df_fd1_v5_Real_ITE1_2024
print("\n=== Dataset de Reales (ITE1 - 2024) ===")
print(df_fd1_v5_Real_ITE1_2024.info())
print("\nPrimeras filas:")
display(df_fd1_v5_Real_ITE1_2024.head())
print("\nDescripción estadística:")
display(df_fd1_v5_Real_ITE1_2024.describe(include='all'))

# Comparamos tamaños y columnas
print("\nComparación de columnas:")
print("ITE1:", df_fd1_v5_ITE1.shape, "columnas:", df_fd1_v5_ITE1.columns.tolist())
print("Real_ITE1_2024:", df_fd1_v5_Real_ITE1_2024.shape, "columnas:", df_fd1_v5_Real_ITE1_2024.columns.tolist())

=== Dataset de Entrenamiento (ITE1) ===
<class 'pandas.core.frame.DataFrame'>
RangeIndex: 174626 entries, 0 to 174625
Data columns (total 12 columns):
 #   Column               Non-Null Count   Dtype
---  ------               --------------   -----
 0   ID_ORDER             174626 non-null  object
 1   ID_BUILDING          174626 non-null  int64
 2   FM_COST_TYPE         174626 non-null  object
 3   MONTH                174626 non-null  int64
 4   YEAR                 174626 non-null  int64
 5   cost_float_mod       174626 non-null  float64
 6   SUPPLIER_TYPE_MOD_2  174626 non-null  object
 7   FM_RESPONSIBLE_MOD   174626 non-null  object
 8   TIPO_USO             174626 non-null  object
 9   COUNTRY_CATALOGO     174626 non-null  object
 10  ID_REGION_GRUPO      174626 non-null  object
 11  COUNTRY_DEF          174626 non-null  object
dtypes: float64(1), int64(3), object(8)
memory usage: 16.0+ MB
None

Primeras filas:

Descripción estadística:

=== Dataset de Reales (ITE1 - 2024) ===
<class 'pandas.core.frame.DataFrame'>
RangeIndex: 69290 entries, 0 to 69289
Data columns (total 12 columns):
 #   Column               Non-Null Count  Dtype
---  ------               --------------  -----
 0   ID_ORDER             69290 non-null  object
 1   ID_BUILDING          69290 non-null  int64
 2   FM_COST_TYPE         69290 non-null  object
 3   MONTH                69290 non-null  int64
 4   YEAR                 69290 non-null  int64
 5   cost_float_mod       69290 non-null  float64
 6   SUPPLIER_TYPE_MOD_2  69290 non-null  object
 7   FM_RESPONSIBLE_MOD   69290 non-null  object
 8   TIPO_USO             69290 non-null  object
 9   COUNTRY_CATALOGO     69290 non-null  object
 10  ID_REGION_GRUPO      69290 non-null  object
 11  COUNTRY_DEF          69290 non-null  object
dtypes: float64(1), int64(3), object(8)
memory usage: 6.3+ MB
None

Primeras filas:

Descripción estadística:

Comparación de columnas:
ITE1: (174626, 12) columnas: ['ID_ORDER', 'ID_BUILDING', 'FM_COST_TYPE', 'MONTH', 'YEAR', 'cost_float_mod', 'SUPPLIER_TYPE_MOD_2', 'FM_RESPONSIBLE_MOD', 'TIPO_USO', 'COUNTRY_CATALOGO', 'ID_REGION_GRUPO', 'COUNTRY_DEF']
Real_ITE1_2024: (69290, 12) columnas: ['ID_ORDER', 'ID_BUILDING', 'FM_COST_TYPE', 'MONTH', 'YEAR', 'cost_float_mod', 'SUPPLIER_TYPE_MOD_2', 'FM_RESPONSIBLE_MOD', 'TIPO_USO', 'COUNTRY_CATALOGO', 'ID_REGION_GRUPO', 'COUNTRY_DEF']

# === Definimos una linea base con Holt-Winters por (ID_BUILDING, FM_COST_TYPE)
# con evaluación de las previsiones contra reales del 2024 ===

# Definimos una primera combinación que deseamos modelar
ID_BUILDING_OBJETIVO = 1001131      # <-- ejemplo
FM_COST_TYPE_OBJETIVO = "Mtto. Contratos"  # <-- ejemplo

# Preparamos función auxiliar para MAPE seguro (ignoramos reales=0 para evitar divisiones por cero)
def mape_safe(y_true, y_pred):
    y_true = np.array(y_true, dtype=float)
    y_pred = np.array(y_pred, dtype=float)
    mask = y_true != 0
    if mask.sum() == 0:
        return np.nan, 0  # no podemos calcular MAPE si todos son 0
    return np.mean(np.abs((y_true[mask] - y_pred[mask]) / y_true[mask])) * 100.0, int((~mask).sum())

# Creamos columna de fecha (usamos el día 1 de cada mes) y agregamos por mes
def preparar_serie(df, id_building, cost_type):
    # Filtramos por la combinación deseada
    df_f = df[(df["ID_BUILDING"] == id_building) & (df["FM_COST_TYPE"] == cost_type)].copy()
    # Creamos columna fecha y agregamos coste mensual (por si hay múltiples órdenes en el mismo mes)
    df_f["FECHA"] = pd.to_datetime(dict(year=df_f["YEAR"], month=df_f["MONTH"], day=1))
    serie_mensual = (
        df_f.groupby("FECHA", as_index=True)["cost_float_mod"]
           .sum()
           .sort_index()
    )
    # Reindexamos a frecuencia mensual continua para asegurar meses sin datos (los ponemos a 0)
    idx_completo = pd.period_range(serie_mensual.index.min(), serie_mensual.index.max(), freq="M").to_timestamp()
    serie_mensual = serie_mensual.reindex(idx_completo, fill_value=0.0)
    return serie_mensual

# Construimos serie de entrenamiento (2021-2023) y de test (2024)
serie_train = preparar_serie(df_fd1_v5_ITE1, ID_BUILDING_OBJETIVO, FM_COST_TYPE_OBJETIVO)
serie_test = preparar_serie(df_fd1_v5_Real_ITE1_2024, ID_BUILDING_OBJETIVO, FM_COST_TYPE_OBJETIVO)

# Nos aseguramos de que las fechas de train estén entre 2021-01 y 2023-12 y test en 2024
serie_train = serie_train[(serie_train.index.year >= 2021) & (serie_train.index.year <= 2023)]
serie_test = serie_test[(serie_test.index.year == 2024)]

# Comprobaciones rápidas
print("=== Combinación objetivo ===")
print(f"ID_BUILDING={ID_BUILDING_OBJETIVO} | FM_COST_TYPE='{FM_COST_TYPE_OBJETIVO}'")
print(f"Meses en TRAIN: {len(serie_train)} | Rango: {serie_train.index.min().date()} -> {serie_train.index.max().date()}")
print(f"Meses en TEST : {len(serie_test)} | Rango: {serie_test.index.min().date() if len(serie_test)>0 else None} -> {serie_test.index.max().date() if len(serie_test)>0 else None}")

if len(serie_train) < 18:
    raise ValueError("Tenemos muy pocos meses en entrenamiento para ajustar una estacionalidad anual. Cambiemos la combinación o usemos un modelo más simple.")

# Ajustamos Holt-Winters con estacionalidad mensual (12)
# Elegimos aditiva como punto de partida; si vemos multiplicatividad, podremos cambiar a 'multiplicative'.
model = ExponentialSmoothing(
    serie_train,
    trend="add",
    seasonal="add",
    seasonal_periods=12,
    initialization_method="estimated"  # dejamos que estime los estados iniciales
)

fit = model.fit(optimized=True)

# Predecimos exactamente los meses del test (2024)
steps = len(serie_test)
if steps == 0:
    raise ValueError("No hay meses en el conjunto de test (2024) para esta combinación. Prueba con otra combinación o verifica los datos.")

pred_test = fit.forecast(steps=steps)
pred_test.index = serie_test.index  # alineamos índices por claridad

# Métricas: RMSE y MAPE (seguro)
rmse = np.sqrt(mean_squared_error(serie_test.values, pred_test.values))
mape, excluidos = mape_safe(serie_test.values, pred_test.values)

print("\n=== Métricas baseline Holt-Winters (2024) ===")
print(f"RMSE: {rmse:,.2f}")
if np.isnan(mape):
    print("MAPE: no calculable (todas las observaciones reales son 0).")
else:
    print(f"MAPE: {mape:,.2f}% (se excluyeron {excluidos} meses con real=0 del cómputo)")

# Mostramos una tabla corta comparando real vs predicho para inspección
df_eval = pd.DataFrame({
    "real_2024": serie_test,
    "pred_hw": pred_test
})
print("\n=== Muestra real vs predicho (primeros 6 meses) ===")
print(df_eval.head(6))

# Guardamos resultados
# ruta_salida = "/content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/"
# df_eval.to_csv(ruta_salida + f"eval_hw_{ID_BUILDING_OBJETIVO}_{FM_COST_TYPE_OBJETIVO.replace(' ', '_')}_2024.csv", sep=";", index=True)

=== Combinación objetivo ===
ID_BUILDING=1001131 | FM_COST_TYPE='Mtto. Contratos'
Meses en TRAIN: 35 | Rango: 2021-01-01 -> 2023-11-01
Meses en TEST : 9 | Rango: 2024-01-01 -> 2024-09-01

=== Métricas baseline Holt-Winters (2024) ===
RMSE: 139.50
MAPE: 92.83% (se excluyeron 6 meses con real=0 del cómputo)

=== Muestra real vs predicho (primeros 6 meses) ===
            real_2024     pred_hw
2024-01-01      235.0   11.197043
2024-02-01        0.0  149.929063
2024-03-01        0.0   14.928970
2024-04-01        0.0   14.928911
2024-05-01      135.0   14.928805
2024-06-01        0.0  149.928728

# ===============================
# PASO 1. Filtramos entrenamiento
# ===============================
# Partimos de df_fd1_v5_ITE1 ya cargado (2021–2023).
# Creamos FECHA como el día 1 de cada mes y nos aseguramos del rango.
df_train_base = df_fd1_v5_ITE1.copy()
df_train_base["FECHA"] = pd.to_datetime(dict(year=df_train_base["YEAR"],
                                            month=df_train_base["MONTH"],
                                            day=1), errors="coerce")

df_train_base = df_train_base[(df_train_base["YEAR"] >= 2021) & (df_train_base["YEAR"] <= 2023)]

# Eliminamos cualquier posible fila con FECHA nula (por seguridad)
df_train_base = df_train_base[df_train_base["FECHA"].notna()]

# =========================================
# PASO 2. Agregamos a nivel mensual/pareja
# =========================================
# Sumamos el coste mensual por (ID_BUILDING, FM_COST_TYPE, FECHA)
monthly_train = (df_train_base
                 .groupby(["ID_BUILDING", "FM_COST_TYPE", "FECHA"], as_index=False)["cost_float_mod"]
                 .sum())

# ===================================================
# PASO 3. Rellenamos meses faltantes por cada pareja
# ===================================================
# Nos definimos una función para rellenar los meses faltantes donde la entrada
# g es un DataFrame para una pareja con columnas ['ID_BUILDING','FM_COST_TYPE'
# ,'FECHA','cost_float_mod']

def reindex_group(g):
    g = g.sort_values("FECHA")
    fecha_min = g["FECHA"].min()
    fecha_max = g["FECHA"].max()
    # Creamos el índice mensual continuo entre min y max
    idx = pd.period_range(fecha_min, fecha_max, freq="M").to_timestamp()
    # Reindexamos
    g2 = g.set_index("FECHA").reindex(idx)
    g2.index.name = "FECHA"
    # Rellenamos coste a 0 donde no haya registro
    g2["cost_float_mod"] = g2["cost_float_mod"].fillna(0.0)
    # Recuperamos las claves
    g2["ID_BUILDING"] = g["ID_BUILDING"].iloc[0]
    g2["FM_COST_TYPE"] = g["FM_COST_TYPE"].iloc[0]
    return g2.reset_index()

df_train_fd1_v5_ITE1 = (monthly_train
                        .groupby(["ID_BUILDING","FM_COST_TYPE"], group_keys=False)
                        .apply(reindex_group)
                        .reset_index(drop=True))

# ===========================================
# PASO 4. Contexto por cada pareja
# ===========================================
# Nos definimos una función para añadir contexto de cada pareja
# ID_BUILDING-FM_COST_TYPE

def mode_or_nan(s: pd.Series):
    m = s.mode()
    return m.iloc[0] if len(m) else np.nan

ctx_cols = ["COUNTRY_DEF","ID_REGION_GRUPO","TIPO_USO","SUPPLIER_TYPE_MOD_2","FM_RESPONSIBLE_MOD"]

ctx_train = (df_train_base
             .groupby(["ID_BUILDING","FM_COST_TYPE"])
             .agg({c: mode_or_nan for c in ctx_cols})
             .reset_index())

# Unimos contexto al panel mensual
df_train_fd1_v5_ITE1 = df_train_fd1_v5_ITE1.merge(ctx_train,
                                                  on=["ID_BUILDING","FM_COST_TYPE"],
                                                  how="left")

# =============================
# PASO 5. Orden y columnas base
# =============================

# Añadimos YEAR y MONTH a partir de FECHA
df_train_fd1_v5_ITE1["YEAR"] = df_train_fd1_v5_ITE1["FECHA"].dt.year
df_train_fd1_v5_ITE1["MONTH"] = df_train_fd1_v5_ITE1["FECHA"].dt.month

# Ordenamos las columnas para modelizar series temporales
cols_orden = [
    "ID_BUILDING","FM_COST_TYPE","FECHA","YEAR","MONTH",
    "cost_float_mod",  # coste mensual agregado
    "COUNTRY_DEF","ID_REGION_GRUPO","TIPO_USO","SUPPLIER_TYPE_MOD_2","FM_RESPONSIBLE_MOD"
]
df_train_fd1_v5_ITE1 = df_train_fd1_v5_ITE1[cols_orden]

# ==========================
# PASO 6. Comprobaciones
# ==========================
print("Forma df_train_fd1_v5_ITE1:", df_train_fd1_v5_ITE1.shape)
print("Rango fechas:", df_train_fd1_v5_ITE1["FECHA"].min().date(), "->", df_train_fd1_v5_ITE1["FECHA"].max().date())

# Verificamos duplicados por clave que no deberían haber
dups = df_train_fd1_v5_ITE1.duplicated(subset=["ID_BUILDING","FM_COST_TYPE","FECHA"]).sum()
print("Duplicados (ID_BUILDING, FM_COST_TYPE, FECHA):", dups)

print("\nMuestra:")
print(df_train_fd1_v5_ITE1.head(10))

# Guardamos en Drive
# ruta_salida = "/content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/"
# df_train_fd1_v5_ITE1.to_csv(ruta_salida + "df_train_fd1_v5_ITE1.csv", sep=";", index=False)

Forma df_train_fd1_v5_ITE1: (65466, 11)
Rango fechas: 2021-01-01 -> 2023-12-01
Duplicados (ID_BUILDING, FM_COST_TYPE, FECHA): 0

Muestra:
   ID_BUILDING           FM_COST_TYPE      FECHA  YEAR  MONTH  cost_float_mod  \
0            2  Eficiencia Energética 2021-12-01  2021     12            0.00
1            2              Licencias 2021-01-01  2021      1         1145.46
2            2              Licencias 2021-02-01  2021      2           55.95
3            2              Licencias 2021-03-01  2021      3            0.00
4            2              Licencias 2021-04-01  2021      4            0.00
5            2              Licencias 2021-05-01  2021      5          305.95
6            2              Licencias 2021-06-01  2021      6           75.00
7            2              Licencias 2021-07-01  2021      7          850.95
8            2              Licencias 2021-08-01  2021      8            0.00
9            2              Licencias 2021-09-01  2021      9            0.00

  COUNTRY_DEF ID_REGION_GRUPO  TIPO_USO SUPPLIER_TYPE_MOD_2  \
0      España               2  Oficinas             INTERNO
1      España               2  Oficinas             INTERNO
2      España               2  Oficinas             INTERNO
3      España               2  Oficinas             INTERNO
4      España               2  Oficinas             INTERNO
5      España               2  Oficinas             INTERNO
6      España               2  Oficinas             INTERNO
7      España               2  Oficinas             INTERNO
8      España               2  Oficinas             INTERNO
9      España               2  Oficinas             INTERNO

      FM_RESPONSIBLE_MOD
0  Eficiencia Energética
1              Licencias
2              Licencias
3              Licencias
4              Licencias
5              Licencias
6              Licencias
7              Licencias
8              Licencias
9              Licencias

def safe_autocorr(x: pd.Series, lag: int) -> float:
    """
      Calcula la autocorrelación de una serie en un retardo (lag) concreto
      Nos sirve para tomar decisiones en función del valor que devuelve y el lag de entrada.
      Devuelve un float con la autocorrelación de Pearson de la serie en el retardo lag (valor entre -1 y 1).
      Si no hay datos suficientes (len(x) <= lag) o algo falla durante el cálculo, devuelve np.nan
    """
    # Calculamos la autocorrelación a un retardo (lag) dado.
    # Si la serie es demasiado corta para ese lag, devolvemos NaN para evitar errores.
    if len(x) <= lag:
        return np.nan
    try:
        # Usamos el método autocorr de pandas; convertimos a float por consistencia.
        return float(pd.Series(x).autocorr(lag=lag))
    except Exception:
        # Si algo falla (valores raros, etc.), devolvemos NaN de forma segura.
        return np.nan

def seasonal_strength_stl(x: pd.Series, period: int = 12) -> float:
    """
      Estima cuánta “fuerza” tiene la
      estacionalidad en una serie mensual (o con el periodo que indiquemos) usando
      una descomposición STL.
      Devuelve un número entre 0 y 1.
    """
    # Estimamos la "fuerza estacional" según Hyndman:
    # Fs = max(0, 1 - Var(remainder) / Var(remainder + seasonal))
    # Intuición: si la estacionalidad explica mucha varianza, Fs se acerca a 1.
    # Requisito: al menos ~2 periodos para que STL sea mínimamente fiable.
    if len(x) < 2*period:
        return np.nan
    try:
        # Descomposición STL robusta para minimizar el impacto de outliers
        res = STL(x, period=period, robust=True).fit()
        rem = res.resid     # componente residual
        seas = res.seasonal # componente estacional
        # Var(remainder + seasonal) = varianza de todo lo que no es tendencia
        denom = np.var(rem + seas)
        if denom <= 0:
            # Si el denominador es 0 (serie constante), forzamos fuerza 0
            return 0.0
        fs = 1 - (np.var(rem) / denom)
        # Acotamos en [0,1] por seguridad numérica
        return float(np.clip(fs, 0, 1))
    except Exception:
        # Si STL falla por cualquier motivo, devolvemos NaN
        return np.nan

def robust_outlier_rate(x: pd.Series, thr: float = 3.5) -> float:
    """
      Identificar % de outliers de cada pareja
      ID_BUILDING - FM_COST_TYPE para cada mes y año.
      Calculamos el % de meses que consideramos outliers usando un z-score robusto
      basado en MAD.
      Devuelve un valor binario en función si es outlier o no.
    """
    # Calculamos:
    # La Mediana y MAD (desviación absoluta mediana).
    # El z_robusto = 0.6745 * (x - mediana) / MAD  (0.6745 normaliza a equivalente-z)
    if len(x) == 0:
        return np.nan
    med = np.median(x)
    mad = np.median(np.abs(x - med))
    if mad == 0:
        # Si MAD=0 (serie casi constante), no hay outliers robustos
        return 0.0
    z = 0.6745 * (x - med) / mad
    return float((np.abs(z) > thr).mean())
    # Marcamos outlier según la salida|z_robusto| > thr (por defecto 3.5, criterio habitual)

def trend_slope(x: pd.Series) -> float:
    """
      Estimamos la pendiente de tendencia mediante
      una regresión lineal simple sobre el índice temporal (0..n-1).
      Si n<3 devolvemos 0 por estabilidad.
    """
    n = len(x)
    if n < 3:
        return 0.0
    t = np.arange(n)
    try:
        # np.polyfit devuelve [pendiente, intercepto]; nos quedamos con la pendiente
        b1, b0 = np.polyfit(t, x.values, 1)
        return float(b1)
    except Exception:
        # Ante cualquier problema numérico, devolvemos 0
        return 0.0

# Función adi_metrics en la que se

def adi_metrics(x: pd.Series) -> dict:
    """
      Mide cuántos meses pasan entre dos meses con gasto positivo.
      Métricas ADI (intermitencia):
      - ADI_n_over_d = n / (#meses con >0)  [clásico y simple]
      - ADI_mean_gap = media de separaciones entre meses positivos consecutivos (en nº de meses)
      Devuelve las 2 métricas ADI y el n. meses con demanda.
    """
    n = len(x)  # nº total de meses en el tramo (p.ej., 36 si 2021–2023 completo)
    pos_idx = np.where(x.values > 0)[0] # posiciones (meses) con gasto > 0
    dcount = len(pos_idx)  # nº de meses con demanda (positivos)

    # Usamos la definición de ADI: “periodos totales” / “nº de demandas”.
    # Si no hay ningún mes positivo (dcount=0) nos devuelve infinito
    # (intermitencia extrema).
    # Si hay 12 meses positivos en 36 meses entonces ADI = 36/12 = 3.0.
    adi_n_over_d = (n / dcount) if dcount > 0 else np.inf

    # Calculamos la media de las separaciones (en meses) entre posiciones
    # de demanda (>0). Ignoramos los “ceros iniciales” antes del primer positivo y
    # los “ceros finales” después del último positivo
    # (a diferencia de n/dcount que implícitamente los incluye).
    # Si solo hay 1 mes positivo, devuelve n (no hay un “gap” entre positivos).
    # Si no hay ninguno, devuelve infinito.
    if dcount >= 2:
        gaps = np.diff(pos_idx)  # separaciones en meses
        adi_mean_gap = float(np.mean(gaps))
    elif dcount == 1:
        adi_mean_gap = float(n)  # una sola demanda en todo el tramo
    else:
        adi_mean_gap = np.inf
    return {"ADI_n_over_d": float(adi_n_over_d), "ADI_mean_gap": adi_mean_gap, "demand_months": int(dcount)}

# Función cv2_positivos

def cv2_positivos(x: pd.Series) -> float:

    """
      Mide la variabilidad relativa de los importes solo en los meses con gasto > 0.
      Devuelve el coeficiente de variación al cuadrado (CV2) de esos valores positivos.
    """
    # CV² sobre meses con >0 (si hay suficientes y media>0)
    xp = x[x > 0]
    if len(xp) <= 1:
        return np.nan
    m = float(xp.mean())
    if m == 0:
        return np.nan
    s = float(xp.std(ddof=1))
    return float((s / m) ** 2)

def max_zero_run(x: pd.Series) -> int:
    """
    Mide la racha (longitud) más larga de meses consecutivos con valor 0.
    Es un indicador muy claro de “sequías” de gasto.
    """
    # Longitud máxima de racha consecutiva de ceros
    if x is None or len(x) == 0:
        return 0
    arr = (x.values == 0).astype(int)
    max_run = run = 0
    for v in arr:
        run = run + 1 if v == 1 else 0
        if run > max_run:
            max_run = run
    return int(max_run)

# Función occ_prob_and_drift

def occ_prob_and_drift(x: pd.Series) -> dict:
    """
      Resume, para una serie mensual x (costes), con qué frecuencia “ocurre gasto”
      y si esa probabilidad está cambiando en el tiempo.
      Devuelve el valor de la probabilidad media de la ocurrencia p_occurrence
      (que es un valor entre 0 y 1) y el valor de occurrence_slope = \β1, que es
      el cambio promedio por mes en la probabilidad de que haya gasto.
    """
    # Probabilidad de ocurrencia (>0) y pendiente de su serie indicadora (posible obsolescencia)
    y = (x.values > 0).astype(float)
    p = float(y.mean())
    if len(y) < 3:
        slope = 0.0
    else:
        try:
            t = np.arange(len(y))
            slope, _ = np.polyfit(t, y, 1)  # OLS sobre indicador
            slope = float(slope)
        except Exception:
            slope = 0.0
    return {"p_occurrence": p, "occurrence_slope": slope}

def zero_ratio(x):
    """
    Calcula el porcentaje de meses con gasto = 0 en la serie mensual x.
    Devuelve un float entre 0 y 1.
    """
    if len(x) == 0:
        return float('nan')
    return (x == 0).sum() / len(x)

# ---------------------------------------------------------------------
# VAR_CTX DE ELEGIBILIDAD 1: has_min_history
# ---------------------------------------------------------------------
def has_min_history(x: pd.Series,
                    min_months: int = 12,
                    require_consecutive: bool = False) -> int:
    """
    Devuelve 1 si la serie x tiene al menos `min_months` puntos temporales tras
    la preparación (reindex mensual); en caso contrario devuelve 0.

    Parámetros
    ----------
    x : pd.Series
        Serie mensual de costes ya reindexada. El índice debe ser DatetimeIndex
        con frecuencia mensual. Los valores deben ser numéricos (floats).
    min_months : int, por defecto 12
        Número mínimo de observaciones para considerar la serie modelable.
    require_consecutive : bool, por defecto False
        Si True, exige que exista al menos un tramo con `min_months` meses
        consecutivos no nulos (valor > 0). Es útil cuando en los casos que hay
        muchos huecos = 0 que no aportan señal para el aprendizaje.

    Retorno
    -------
    int
        1 si cumple el criterio (apta), 0 si no (escape).
    """
    # Vía de escape si la serie es None o vacía
    if x is None or len(x) == 0:
        return 0

    # Normalizamos a float y gestionamos NaN
    x = pd.to_numeric(x, errors="coerce")

    # Criterio simple por longitud total (incluye ceros)
    if not require_consecutive:
        return int(len(x.dropna()) >= min_months)

    # Criterio alternativo: tramo consecutivo con "señal" (valor > 0)
    # Construimos una máscara booleana de "mes con señal"
    mask_signal = (x.fillna(0) > 0).astype(int).values

    # Calculamos la racha máxima (run length) de 1s consecutivos
    max_run = 0
    current = 0
    for v in mask_signal:
        if v == 1:
            current += 1
            max_run = max(max_run, current)
        else:
            current = 0

    return int(max_run >= min_months)

# ---------------------------------------------------------------------
# VAR_CTX DE ELEGIBILIDAD 2: effective_value_ratio
# ---------------------------------------------------------------------
def effective_value_ratio(x: pd.Series, thr: float = 0.0) -> float:
    """
    Proporción de meses "con señal" en la serie x, entendiendo por señal
    un valor estrictamente mayor que `thr`. Por ejemplo, thr=0.0 implica
    contar meses con coste > 0.

    Devuelve un float en [0, 1]. Si la serie está vacía, devuelve np.nan.

    Parámetros
    ----------
    x : pd.Series
        Serie mensual reindexada y numérica.
    thr : float, por defecto 0.0
        Umbral mínimo para considerar que un mes aporta señal.

    Retorno
    -------
    float
        (# meses con valor > thr) / (# meses totales con valor no NaN)
    """
    if x is None or len(x) == 0:
        return np.nan

    x = pd.to_numeric(x, errors="coerce")
    n = x.notna().sum()
    if n == 0:
        return np.nan

    effective = (x.fillna(0) > thr).sum()
    return float(effective) / float(n)

# ---------------------------------------------------------------------
# VAR_CTX DE ELEGIBILIDAD 3: total_cost_guard
# ---------------------------------------------------------------------
def total_cost_guard(x: pd.Series,
                     min_sum: float = 0.0,
                     window_months: Optional[int] = None) -> float:
    """
    Suma total (o suma de la ventana más reciente) de la serie x. Permite
    fijar un umbral mínimo `min_sum` para filtrar series irrelevantes por
    cuantía económica.

    Devuelve la suma (float). Si la serie está vacía, devuelve np.nan.

    Parámetros
    ----------
    x : pd.Series
        Serie mensual reindexada y numérica.
    min_sum : float, por defecto 0.0
        Umbral de suma mínimo recomendado para marcar elegibilidad.
        (La decisión binaria se hace fuera, con la suma devuelta)
    window_months : int opcional
        Si se indica, se suma solo la ventana más reciente de `window_months`
        meses (p.ej., últimos 12 o 24). Si None, se suma toda la serie.

    Retorno
    -------
    float
        Suma (total o en ventana). Útil para comparar contra `min_sum`.
    """
    if x is None or len(x) == 0:
        return np.nan

    x = pd.to_numeric(x, errors="coerce").fillna(0.0)
    if window_months is not None and window_months > 0:
        x = x.tail(window_months)
    return float(x.sum())

# ---------------------------------------------------------------------
# VAR_CTX de ENVOLTORIO: elegibilidad_global
# ---------------------------------------------------------------------
def elegibilidad_global(x: pd.Series,
                        min_months: int = 12,
                        require_consecutive: bool = False,
                        eff_thr: float = 0.0,
                        eff_min_ratio: float = 0.25,
                        min_sum: float = 0.0,
                        sum_window_months: Optional[int] = None) -> Dict[str, Any]:
    """
    Evalúa múltiples criterios de elegibilidad y los devuelve en un dict.
    Está pensado para integrarlo en el pipeline antes de los KPIs de
    comportamiento (tendencia, estacionalidad, intermitencia, etc.).

    Reglas recomendadas (ajustables):
      - has_min_history: al menos `min_months` observaciones (o racha de señal)
      - effective_value_ratio: al menos `eff_min_ratio` de meses con señal
      - total_cost_guard: suma mínima `min_sum` (total o en ventana)

    Parámetros
    ----------
    x : pd.Series
        Serie mensual reindexada y numérica.
    min_months : int
        Longitud mínima exigida.
    require_consecutive : bool
        Si True, exige racha consecutiva de longitud min_months con señal.
    eff_thr : float
        Umbral para definir "señal" en effective_value_ratio (default 0.0).
    eff_min_ratio : float
        Ratio mínimo aceptable de meses con señal (default 0.25 = 25%).
    min_sum : float
        Suma mínima exigida (0.0 desactiva el filtro por cuantía).
    sum_window_months : int opcional
        Ventana reciente para la suma (si None, usa toda la serie).

    Retorno
    -------
    Dict[str, Any]
        {
          'has_min_history': 0/1,
          'effective_value_ratio': float,
          'passes_effective_ratio': 0/1,
          'total_sum': float,
          'passes_min_sum': 0/1,
          'eligible_all': 0/1  # intersección de los criterios activos
        }
    """
    # 1) Historial mínimo
    flag_hist = has_min_history(
        x, min_months=min_months, require_consecutive=require_consecutive
    )

    # 2) Ratio de meses con señal
    eff_ratio = effective_value_ratio(x, thr=eff_thr)
    # Si eff_ratio es NaN, no pasa el criterio
    passes_eff = int((not np.isnan(eff_ratio)) and (eff_ratio >= eff_min_ratio))

    # 3) Guardia por suma total o en ventana
    total_sum = total_cost_guard(x, min_sum=min_sum, window_months=sum_window_months)
    # Si min_sum <= 0, desactivamos el filtro (lo consideramos pasado)
    if min_sum <= 0.0 or np.isnan(total_sum):
        passes_sum = int(min_sum <= 0.0 and not np.isnan(total_sum))
    else:
        passes_sum = int(total_sum >= min_sum)

    # 4) Elegibilidad global: intersección (AND) de los criterios activos
    #    - Historial mínimo siempre activo
    #    - Ratio efectivo activo si eff_min_ratio > 0
    #    - Guardia por suma activa si min_sum > 0
    active_flags = [flag_hist]
    if eff_min_ratio > 0.0:
        active_flags.append(passes_eff)
    if min_sum > 0.0:
        active_flags.append(passes_sum)

    eligible_all = int(all(v == 1 for v in active_flags))

    return {
        "has_min_history": int(flag_hist),
        "effective_value_ratio": float(eff_ratio) if not np.isnan(eff_ratio) else np.nan,
        "passes_effective_ratio": int(passes_eff),
        "total_sum": float(total_sum) if not np.isnan(total_sum) else np.nan,
        "passes_min_sum": int(passes_sum),
        "eligible_all": int(eligible_all)
    }


# ---------------------------------------------------------------------
# EJEMPLO DE USO
# ---------------------------------------------------------------------
# Supongamos que `s` es una serie mensual de costes:
# s.index -> pd.date_range('2021-01-01', periods=36, freq='MS')
# s.values -> floats con costes mensuales (incluye 0 y NaN tras reindex)
#
# 1) Comprobar solo historia mínima:
# has_min = has_min_history(s, min_months=12, require_consecutive=False)
#
# 2) Proporción de meses con señal (>0):
# eff = effective_value_ratio(s, thr=0.0)
#
# 3) Suma total en últimos 12 meses:
# suma12 = total_cost_guard(s, min_sum=0.0, window_months=12)
#
# 4) Chequeo global (recomendado en pipeline):
# flags = elegibilidad_global(
#     s,
#     min_months=12,
#     require_consecutive=False,
#     eff_thr=0.0,
#     eff_min_ratio=0.25,
#     min_sum=0.0,              # ejemplo: desactivado
#     sum_window_months=12      # ejemplo: mirar últimos 12 meses
# )
# if flags["eligible_all"] == 0:
#     # Marcar la serie como "escape" y resolver con pooling/global
#     pass

# ============================================
# Diagnóstico y modelo sugerido
# ============================================
def diagnosticar_y_sugerir_modelo(row: pd.Series) -> Dict[str, str]:
    eps = 1e-9
    zr = float(row.get("zero_ratio", np.nan))
    out_rate = float(row.get("outlier_rate", np.nan))
    slope = float(row.get("slope", 0.0))
    mean = float(row.get("mean", np.nan))
    acf12 = float(row.get("acf12", np.nan))
    seas = float(row.get("seasonality_strength", np.nan))
    p_occ = float(row.get("p_occurrence", np.nan))
    occ_slope = float(row.get("occurrence_slope", np.nan))
    adi_n_over_d = float(row.get("ADI_n_over_d", np.nan))
    neg_share = float(row.get("neg_share", 0.0))

    if int(row.get("eligible_all", 1)) == 0:
        return {"diagnosis": "escape_no_elegible", "model_suggested": "pooling_global|naive_0"}

    if (not np.isnan(zr) and zr >= 0.5) or (not np.isnan(adi_n_over_d) and adi_n_over_d > 1.32):
        if (not np.isnan(p_occ) and not np.isnan(occ_slope)) and (p_occ < 0.8 and occ_slope > 0):
            return {"diagnosis": "intermitente_creciente", "model_suggested": "Croston|SBA"}
        return {"diagnosis": "intermitente", "model_suggested": "Croston|SBA"}

    if not np.isnan(out_rate) and out_rate >= 0.2:
        return {"diagnosis": "picos_outliers", "model_suggested": "transform_log1p+winsor|robusto(ETS/TBATS)"}

    slope_rel = slope / (abs(mean) + eps) if not np.isnan(mean) else 0.0
    if abs(slope_rel) >= 0.05:
        return {"diagnosis": "tendencia", "model_suggested": "Holt(ETS)|ARIMA_drift"}

    if (not np.isnan(seas) and seas >= 0.6) or (not np.isnan(acf12) and acf12 >= 0.3):
        return {"diagnosis": "estacional", "model_suggested": "ETS estacional|SARIMA"}

    if neg_share >= 0.2:
        return {"diagnosis": "abonos_relevantes", "model_suggested": "preproceso_abonos+modelo_robusto"}

    return {"diagnosis": "estable", "model_suggested": "ETS_simple|ARIMA_basico"}

# ============================================
# Pipeline de cálculo KPIs + diagnóstico
# ============================================
def calcular_kpis_y_diagnostico(df_train_fd1_v5_ITE1: pd.DataFrame,
                                min_months: int = 12,
                                eff_min_ratio: float = 0.25,
                                require_consecutive: bool = False,
                                min_sum: float = 0.0,
                                sum_window_months: Optional[int] = None) -> pd.DataFrame:
    kpis = []
    for (bid, ctype), g in df_train_fd1_v5_ITE1.groupby(["ID_BUILDING", "FM_COST_TYPE"]):
        s = g.set_index("FECHA")["cost_float_mod"].sort_index()
        n = len(s)

        zeros_share = float((s == 0).mean()) if n else np.nan
        neg_share = float((s < 0).mean()) if n else np.nan
        mean = float(s.mean()) if n else np.nan
        std = float(s.std(ddof=1)) if n > 1 else np.nan
        cv = float(std/mean) if (n and mean not in [0, np.nan] and not np.isnan(mean)) else np.nan

        acf1 = safe_autocorr(s, 1)
        acf12 = safe_autocorr(s, 12)
        fs = seasonal_strength_stl(s, period=12)
        out_rate = robust_outlier_rate(s)
        slope = trend_slope(s)
        adi = adi_metrics(s)
        cv2_pos = cv2_positivos(s)
        max_zeros = max_zero_run(s)
        occ_stats = occ_prob_and_drift(s)

        zr = zero_ratio(s)
        flags = elegibilidad_global(s, min_months, require_consecutive, 0.0, eff_min_ratio, min_sum, sum_window_months)

        dmin, dmax = (s.index.min(), s.index.max()) if n else (pd.NaT, pd.NaT)

        row = {
            "ID_BUILDING": bid,
            "FM_COST_TYPE": ctype,
            "date_min": dmin, "date_max": dmax,
            "n_months": n,
            "sparsity": zeros_share,
            "zero_ratio": zr,
            "neg_share": neg_share,
            "mean": mean, "std": std, "cv": cv,
            "acf1": acf1, "acf12": acf12,
            "seasonality_strength": fs,
            "outlier_rate": out_rate,
            "slope": slope,
            "ADI_n_over_d": adi.get("ADI_n_over_d", np.nan) if isinstance(adi, dict) else np.nan,
            "ADI_mean_gap": adi.get("ADI_mean_gap", np.nan) if isinstance(adi, dict) else np.nan,
            "demand_months": adi.get("demand_months", np.nan) if isinstance(adi, dict) else np.nan,
            "cv2_positivos": cv2_pos,
            "max_zero_run": max_zeros,
            "p_occurrence": occ_stats.get("p_occurrence", np.nan) if isinstance(occ_stats, dict) else np.nan,
            "occurrence_slope": occ_stats.get("occurrence_slope", np.nan) if isinstance(occ_stats, dict) else np.nan,
            **flags,
            "COUNTRY_DEF": g["COUNTRY_DEF"].mode().iloc[0] if not g["COUNTRY_DEF"].mode().empty else np.nan,
            "ID_REGION_GRUPO": g["ID_REGION_GRUPO"].mode().iloc[0] if not g["ID_REGION_GRUPO"].mode().empty else np.nan,
            "TIPO_USO": g["TIPO_USO"].mode().iloc[0] if not g["TIPO_USO"].mode().empty else np.nan,
            "SUPPLIER_TYPE_MOD_2": g["SUPPLIER_TYPE_MOD_2"].mode().iloc[0] if not g["SUPPLIER_TYPE_MOD_2"].mode().empty else np.nan,
            "FM_RESPONSIBLE_MOD": g["FM_RESPONSIBLE_MOD"].mode().iloc[0] if not g["FM_RESPONSIBLE_MOD"].mode().empty else np.nan,
        }

        decision = diagnosticar_y_sugerir_modelo(pd.Series(row))
        row.update(decision)
        kpis.append(row)

    return pd.DataFrame(kpis).sort_values(["FM_COST_TYPE", "ID_BUILDING"]).reset_index(drop=True)

# === 0) IMPORTS (aseguremos de tener previamente definidas las funciones KPI base) ===
# Deben estar ya definidas en el notebook:
# safe_autocorr, seasonal_strength_stl, robust_outlier_rate, trend_slope,
# adi_metrics, cv2_positivos, max_zero_run, occ_prob_and_drift,
# zero_ratio, has_min_history,
# effective_value_ratio, total_cost_guard, elegibilidad_global,
# diagnosticar_y_sugerir_modelo, calcular_kpis_y_diagnostico


# === 1) CHEQUEOS RÁPIDOS DEL DF (nos evitamos sorpresas) ===
required_cols = {
    "ID_BUILDING","FM_COST_TYPE","FECHA","cost_float_mod",
    "COUNTRY_DEF","ID_REGION_GRUPO","TIPO_USO","SUPPLIER_TYPE_MOD_2","FM_RESPONSIBLE_MOD"
}
missing = required_cols - set(df_train_fd1_v5_ITE1.columns)
assert not missing, f"Faltan columnas en df_train_fd1_v5_ITE1: {missing}"

# Aseguramos tipo datetime
if not np.issubdtype(df_train_fd1_v5_ITE1["FECHA"].dtype, np.datetime64):
    df_train_fd1_v5_ITE1["FECHA"] = pd.to_datetime(df_train_fd1_v5_ITE1["FECHA"], errors="coerce")

assert df_train_fd1_v5_ITE1["FECHA"].notna().all(), "Hay FECHA NaT tras conversión; revisa el dataset."

# Verificamos el rango temporal esperado (2021–2023 para ITE1)
print("Rango fechas ITE1:", df_train_fd1_v5_ITE1["FECHA"].min(), "→", df_train_fd1_v5_ITE1["FECHA"].max())

# === 2) EJECUTAR LA FUNCIÓN CON PARÁMETROS BASE ===
df_result = calcular_kpis_y_diagnostico(
    df_train_fd1_v5_ITE1,
    min_months=12,        # histórico mínimo para modelizar
    eff_min_ratio=0.25,   # % mínimo de meses con señal (>0)
    require_consecutive=False,
    min_sum=0.0,          # activa guardia por cuantía si > 0
    sum_window_months=None
)

print("Columnas devueltas:", list(df_result.columns))
print("Shape:", df_result.shape)
print(df_result.head(10))


# === 3) DIAGNÓSTICO GLOBAL RÁPIDO ===
print("\nDistribución de diagnósticos:")
print(df_result["diagnosis"].value_counts(dropna=False).sort_index())

print("\nModelos sugeridos (top 10):")
print(df_result["model_suggested"].value_counts().head(10))

# Casos no elegibles (escape)
mask_escape = df_result["diagnosis"] == "escape_no_elegible"
print("\n% series no elegibles:", round(100 * mask_escape.mean(), 2), "%")

# Intermitentes (zero_ratio alto o ADI alto)
mask_intermitente = df_result["diagnosis"].str.contains("intermitente", na=False)
print("% series intermitentes:", round(100 * mask_intermitente.mean(), 2), "%")

# Estacionales
mask_estacional = df_result["diagnosis"] == "estacional"
print("% series estacionales:", round(100 * mask_estacional.mean(), 2), "%")


# === 4) SANITY CHECKS (resumen de KPIs clave) ===
kpis_overview = df_result[[
    "zero_ratio","outlier_rate","slope","seasonality_strength",
    "acf12","ADI_n_over_d","neg_share","effective_value_ratio","eligible_all"
]].describe(include="all")
print("\nResumen KPIs clave:\n", kpis_overview)

# Vemos ejemplos de cada diagnóstico para inspección manual
ejemplos = {}
for etiqueta in ["escape_no_elegible","intermitente","intermitente_creciente","picos_outliers","tendencia","estacional","abonos_relevantes","estable"]:
    ejemplos[etiqueta] = df_result[df_result["diagnosis"] == etiqueta].head(3)

# Muestra un ejemplo concreto (cambia la etiqueta por la que te interese)
print("\nEjemplos de 'intermitente':")
print(ejemplos["intermitente"][[
    "ID_BUILDING","FM_COST_TYPE","zero_ratio","ADI_n_over_d","p_occurrence","occurrence_slope","diagnosis","model_suggested"
]])


# === 5) AJUSTES FINOS DE LOS UMBRALES (si fuera necesario) ===
# Si vemos demasiadas series clasificadas como 'picos_outliers', baja o sube el umbral en la función diagnosticar_y_sugerir_modelo
# (por ejemplo, pasar outlier_rate >= 0.2 a 0.25 o 0.15).
# Lo mismo para:
#  - intermitencia: zero_ratio >= 0.5, ADI_n_over_d > 1.32
#  - tendencia relativa: abs(slope_rel) >= 0.05
#  - estacionalidad: seasonality_strength >= 0.6 o acf12 >= 0.3

Rango fechas ITE1: 2021-01-01 00:00:00 → 2023-12-01 00:00:00
Columnas devueltas: ['ID_BUILDING', 'FM_COST_TYPE', 'date_min', 'date_max', 'n_months', 'sparsity', 'zero_ratio', 'neg_share', 'mean', 'std', 'cv', 'acf1', 'acf12', 'seasonality_strength', 'outlier_rate', 'slope', 'ADI_n_over_d', 'ADI_mean_gap', 'demand_months', 'cv2_positivos', 'max_zero_run', 'p_occurrence', 'occurrence_slope', 'has_min_history', 'effective_value_ratio', 'passes_effective_ratio', 'total_sum', 'passes_min_sum', 'eligible_all', 'COUNTRY_DEF', 'ID_REGION_GRUPO', 'TIPO_USO', 'SUPPLIER_TYPE_MOD_2', 'FM_RESPONSIBLE_MOD', 'diagnosis', 'model_suggested']
Shape: (2429, 36)
   ID_BUILDING           FM_COST_TYPE   date_min   date_max  n_months  \
0            2  Eficiencia Energética 2021-12-01 2021-12-01         1
1            9  Eficiencia Energética 2022-10-01 2023-12-01        15
2           18  Eficiencia Energética 2021-01-01 2023-05-01        29
3           57  Eficiencia Energética 2021-04-01 2021-04-01         1
4          126  Eficiencia Energética 2021-04-01 2023-11-01        32
5          129  Eficiencia Energética 2023-11-01 2023-11-01         1
6          135  Eficiencia Energética 2021-04-01 2021-04-01         1
7          136  Eficiencia Energética 2021-04-01 2023-11-01        32
8          137  Eficiencia Energética 2021-04-01 2023-11-01        32
9          146  Eficiencia Energética 2022-05-01 2022-05-01         1

   sparsity  zero_ratio  neg_share         mean          std  ...  total_sum  \
0   1.00000     1.00000        0.0     0.000000          NaN  ...       0.00
1   1.00000     1.00000        0.0     0.000000     0.000000  ...       0.00
2   1.00000     1.00000        0.0     0.000000     0.000000  ...       0.00
3   0.00000     0.00000        0.0   218.780000          NaN  ...     218.78
4   0.84375     0.84375        0.0  1061.812500  2506.910653  ...   33978.00
5   1.00000     1.00000        0.0     0.000000          NaN  ...       0.00
6   0.00000     0.00000        0.0   254.610000          NaN  ...     254.61
7   0.84375     0.84375        0.0  1350.000000  3187.313562  ...   43200.00
8   0.81250     0.81250        0.0   494.483125  1144.964316  ...   15823.46
9   1.00000     1.00000        0.0     0.000000          NaN  ...       0.00

   passes_min_sum  eligible_all  COUNTRY_DEF  ID_REGION_GRUPO      TIPO_USO  \
0               1             0       España                2      Oficinas
1               1             0       España                2      Oficinas
2               1             0       España                2      Oficinas
3               1             0       España                2  Subarrendado
4               1             0       España               12         Bingo
5               1             0       España     Otros_España         Bingo
6               1             0       España                2         Bingo
7               1             0       España                2         Bingo
8               1             0       España                2         Bingo
9               1             0       España                4         Bingo

   SUPPLIER_TYPE_MOD_2     FM_RESPONSIBLE_MOD           diagnosis  \
0              INTERNO  Eficiencia Energética  escape_no_elegible
1              INTERNO  Eficiencia Energética  escape_no_elegible
2              INTERNO  Eficiencia Energética  escape_no_elegible
3              EXTERNO  Eficiencia Energética  escape_no_elegible
4              EXTERNO  Eficiencia Energética  escape_no_elegible
5              INTERNO  Eficiencia Energética  escape_no_elegible
6              EXTERNO  Eficiencia Energética  escape_no_elegible
7              EXTERNO  Eficiencia Energética  escape_no_elegible
8              EXTERNO  Eficiencia Energética  escape_no_elegible
9              EXTERNO  Eficiencia Energética  escape_no_elegible

          model_suggested
0  pooling_global|naive_0
1  pooling_global|naive_0
2  pooling_global|naive_0
3  pooling_global|naive_0
4  pooling_global|naive_0
5  pooling_global|naive_0
6  pooling_global|naive_0
7  pooling_global|naive_0
8  pooling_global|naive_0
9  pooling_global|naive_0

[10 rows x 36 columns]

Distribución de diagnósticos:
diagnosis
escape_no_elegible        870
estable                   500
estacional                489
intermitente              205
intermitente_creciente    274
picos_outliers             60
tendencia                  31
Name: count, dtype: int64

Modelos sugeridos (top 10):
model_suggested
pooling_global|naive_0                       870
ETS_simple|ARIMA_basico                      500
ETS estacional|SARIMA                        489
Croston|SBA                                  479
transform_log1p+winsor|robusto(ETS/TBATS)     60
Holt(ETS)|ARIMA_drift                         31
Name: count, dtype: int64

% series no elegibles: 35.82 %
% series intermitentes: 19.72 %
% series estacionales: 20.13 %

Resumen KPIs clave:
         zero_ratio  outlier_rate        slope  seasonality_strength  \
count  2429.000000   2429.000000  2429.000000           1741.000000
mean      0.358087      0.036275    20.647309              0.610967
std       0.368764      0.079214    86.772806              0.293074
min       0.000000      0.000000 -1191.035000              0.000000
25%       0.000000      0.000000    -0.141832              0.410353
50%       0.166667      0.000000     1.550636              0.646491
75%       0.750000      0.028571    15.062425              0.859926
max       1.000000      0.484848   779.582005              1.000000

             acf12  ADI_n_over_d    neg_share  effective_value_ratio  \
count  1874.000000        2429.0  2429.000000            2429.000000
mean      0.153561           inf     0.000418               0.641495
std       0.437260           NaN     0.010464               0.368543
min      -1.000000           1.0     0.000000               0.000000
25%      -0.156602           1.0     0.000000               0.250000
50%       0.043378           1.2     0.000000               0.833333
75%       0.518020           4.0     0.000000               1.000000
max       1.000000           inf     0.500000               1.000000

       eligible_all
count   2429.000000
mean       0.641828
std        0.479562
min        0.000000
25%        0.000000
50%        1.000000
75%        1.000000
max        1.000000

Ejemplos de 'intermitente':
    ID_BUILDING FM_COST_TYPE  zero_ratio  ADI_n_over_d  p_occurrence  \
70            2    Licencias    0.742857      3.888889      0.257143
71            9    Licencias    0.638889      2.769231      0.361111
72           18    Licencias    0.722222      3.600000      0.277778

    occurrence_slope     diagnosis model_suggested
70         -0.022409  intermitente     Croston|SBA
71         -0.001158  intermitente     Croston|SBA
72         -0.010039  intermitente     Croston|SBA

# Guardamos el diagnóstico y modelado sugerido en un dataframe específico y único para guardar al Drive

df_result_train_fd1_v5_ITE1 = df_result.copy()

# Carpeta de salida en Drive
OUT_DIR = "/content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO"
os.makedirs(OUT_DIR, exist_ok=True)

# Rutas de salida (CSV + Excel único con varias hojas)
csv_path    = os.path.join(OUT_DIR, "df_result_train_fd1_v5_ITE1.csv")
xlsx_path   = os.path.join(OUT_DIR, "df_result_train_fd1_v5_ITE1.xlsx")

# Guardar CSV principal
df_result_train_fd1_v5_ITE1.to_csv(csv_path, sep=";",index=False)

# Resumen por diagnosis y modelo
resumen_por_diagnosis = (
    df_result_train_fd1_v5_ITE1
    .groupby("diagnosis", dropna=False)
    .size()
    .rename("series")
    .reset_index()
    .sort_values("series", ascending=False)
)

resumen_por_modelo = (
    df_result_train_fd1_v5_ITE1
    .groupby(["diagnosis","model_suggested"], dropna=False)
    .size()
    .rename("series")
    .reset_index()
    .sort_values(["diagnosis","series"], ascending=[True,False])
)

# Guardar también resúmenes en CSV separados
csv_diag_sum = os.path.join(OUT_DIR, "resumen_por_diagnosis_ITE1.csv")
csv_mod_sum  = os.path.join(OUT_DIR, "resumen_por_modelo_ITE1.csv")
resumen_por_diagnosis.to_csv(csv_diag_sum, sep=";", index=False)
resumen_por_modelo.to_csv(csv_mod_sum, sep=";", index=False)

# Excel único con varias pestañas (df_result + resúmenes)
# Nota: si alguna columna datetime tuviera tz, Excel podría fallar; en ese caso deberíamos deszonificar antes.
try:
    with pd.ExcelWriter(xlsx_path, engine="openpyxl") as writer:
        df_result_train_fd1_v5_ITE1.to_excel(writer, sheet_name="df_result", index=False)
        resumen_por_diagnosis.to_excel(writer, sheet_name="resumen_diagnosis", index=False)
        resumen_por_modelo.to_excel(writer, sheet_name="resumen_modelo", index=False)
except Exception as e:
    print(f"[AVISO] No se pudo guardar Excel: {type(e).__name__}: {e}")

print("\n[OK] df_result guardado.")
print(" - CSV:", csv_path)
print(" - Excel (multi-hojas):", xlsx_path)
print(" - Resúmenes también en CSV (opcional).")

[OK] df_result guardado.
 - CSV: /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/df_result_train_fd1_v5_ITE1.csv
 - Excel (multi-hojas): /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/df_result_train_fd1_v5_ITE1.xlsx
 - Resúmenes también en CSV (opcional).

# Ruta del notebook .ipynb que vamos a convertir

RUTA_NOTEBOOK = "/content/drive/MyDrive/PROYECTO_NEITH/DOCUMENTOS_PROYECTOS/Neith.ipynb"

# Carpeta de salida y nombre del HTML
OUTPUT_DIR = "/content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO"
OUTPUT_HTML = "Neith_notebook_v5_train_con_KPI_Diagnostico_Modelo.html"

# Aseguramos carpeta de salida (ya en cabecera)
# import os
os.makedirs(OUTPUT_DIR, exist_ok=True)

# Instalamos nbconvert (ya en cabecera)
# !pip install -q "nbconvert>=7.0.0"

# Convertimos notebook a HTML y lo guardamos con el nombre indicado en la carpeta de salida
!jupyter nbconvert --to html "$RUTA_NOTEBOOK" --output "$OUTPUT_HTML" --output-dir "$OUTPUT_DIR"

# Verificamos
salida = os.path.join(OUTPUT_DIR, OUTPUT_HTML)
print("Archivo HTML generado en:", salida, "\nExiste:", os.path.exists(salida))

[NbConvertApp] Converting notebook /content/drive/MyDrive/PROYECTO_NEITH/DOCUMENTOS_PROYECTOS/Neith.ipynb to html
[NbConvertApp] WARNING | Alternative text is missing on 1 image(s).
[NbConvertApp] Writing 1430970 bytes to /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/Neith_notebook_v5_train_con_KPI_Diagnostico_Modelo.html
Archivo HTML generado en: /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/Neith_notebook_v5_train_con_KPI_Diagnostico_Modelo.html
Existe: True

# === CARGAMOS DE df_result_train_fd1_v5_ITE1 (ITE1: 2021–2023) ===
# Intentamos cargar desde Excel (multi-hoja, sheet 'df_result'); si no, desde CSV.

# Ruta de salida usada previamente
OUT_DIR = "/content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO"

xlsx_path = os.path.join(OUT_DIR, "df_result_train_fd1_v5_ITE1.xlsx")
csv_path  = os.path.join(OUT_DIR, "df_result_train_fd1_v5_ITE1.csv")

def _read_df_result(xlsx_path: str, csv_path: str) -> pd.DataFrame:
    # 1) Intentar Excel con hoja 'df_result'
    if os.path.exists(xlsx_path):
        try:
            # Intentamos leer la hoja explícita
            return pd.read_excel(xlsx_path, sheet_name="df_result")
        except ValueError:
            # Si no existe la hoja, leemos la primera disponible
            xls = pd.ExcelFile(xlsx_path)
            first_sheet = xls.sheet_names[0]
            return pd.read_excel(xlsx_path, sheet_name=first_sheet)
    # 2) Fallback: CSV
    if os.path.exists(csv_path):
        return pd.read_csv(csv_path, sep=";")
    raise FileNotFoundError(
        f"No se encontró ni el Excel ni el CSV esperados:\n- {xlsx_path}\n- {csv_path}"
    )

df_result_train_fd1_v5_ITE1 = _read_df_result(xlsx_path, csv_path)

# --- Chequeos suaves y tipificación útil ---
# Columnas esperadas mínimas (ajusta si tu función devuelve otras)
expected_min_cols = {
    "ID_BUILDING", "FM_COST_TYPE", "diagnosis", "model_suggested"
}
missing = expected_min_cols - set(df_result_train_fd1_v5_ITE1.columns)
if missing:
    print(f"[AVISO] Faltan columnas mínimas esperadas en df_result: {missing}")

# Tipos prácticos
if "ID_BUILDING" in df_result_train_fd1_v5_ITE1.columns:
    # coercible a entero
    df_result_train_fd1_v5_ITE1["ID_BUILDING"] = pd.to_numeric(
        df_result_train_fd1_v5_ITE1["ID_BUILDING"], errors="coerce"
    ).astype("Int64")

if "FM_COST_TYPE" in df_result_train_fd1_v5_ITE1.columns:
    df_result_train_fd1_v5_ITE1["FM_COST_TYPE"] = df_result_train_fd1_v5_ITE1["FM_COST_TYPE"].astype(str)

# Si existen KPIs numéricos, los intentamos convertir
kpi_candidates = [
    "zero_ratio","outlier_rate","slope","seasonality_strength",
    "acf12","ADI_n_over_d","neg_share","effective_value_ratio"
]
for c in kpi_candidates:
    if c in df_result_train_fd1_v5_ITE1.columns:
        df_result_train_fd1_v5_ITE1[c] = pd.to_numeric(
            df_result_train_fd1_v5_ITE1[c], errors="coerce"
        )

print("[OK] Cargado df_result_train_fd1_v5_ITE1")
print("Shape:", df_result_train_fd1_v5_ITE1.shape)
print(df_result_train_fd1_v5_ITE1.head(5))

[OK] Cargado df_result_train_fd1_v5_ITE1
Shape: (2429, 36)
   ID_BUILDING           FM_COST_TYPE   date_min   date_max  n_months  \
0            2  Eficiencia Energética 2021-12-01 2021-12-01         1
1            9  Eficiencia Energética 2022-10-01 2023-12-01        15
2           18  Eficiencia Energética 2021-01-01 2023-05-01        29
3           57  Eficiencia Energética 2021-04-01 2021-04-01         1
4          126  Eficiencia Energética 2021-04-01 2023-11-01        32

   sparsity  zero_ratio  neg_share       mean          std  ...  total_sum  \
0   1.00000     1.00000        0.0     0.0000          NaN  ...       0.00
1   1.00000     1.00000        0.0     0.0000     0.000000  ...       0.00
2   1.00000     1.00000        0.0     0.0000     0.000000  ...       0.00
3   0.00000     0.00000        0.0   218.7800          NaN  ...     218.78
4   0.84375     0.84375        0.0  1061.8125  2506.910653  ...   33978.00

   passes_min_sum  eligible_all  COUNTRY_DEF  ID_REGION_GRUPO      TIPO_USO  \
0               1             0       España                2      Oficinas
1               1             0       España                2      Oficinas
2               1             0       España                2      Oficinas
3               1             0       España                2  Subarrendado
4               1             0       España               12         Bingo

   SUPPLIER_TYPE_MOD_2     FM_RESPONSIBLE_MOD           diagnosis  \
0              INTERNO  Eficiencia Energética  escape_no_elegible
1              INTERNO  Eficiencia Energética  escape_no_elegible
2              INTERNO  Eficiencia Energética  escape_no_elegible
3              EXTERNO  Eficiencia Energética  escape_no_elegible
4              EXTERNO  Eficiencia Energética  escape_no_elegible

          model_suggested
0  pooling_global|naive_0
1  pooling_global|naive_0
2  pooling_global|naive_0
3  pooling_global|naive_0
4  pooling_global|naive_0

[5 rows x 36 columns]

# === EXPLORACIÓN DE df_result_train_fd1_v5_ITE1 cargado ===

# Shape general
print("Shape:", df_result_train_fd1_v5_ITE1.shape)

# Columnas disponibles
print("\nColumnas:")
print(list(df_result_train_fd1_v5_ITE1.columns))

# Tipos de datos y nulos
print("\nInfo del dataframe:")
print(df_result_train_fd1_v5_ITE1.info())

# Primeras filas
print("\nPrimeras filas:")
print(df_result_train_fd1_v5_ITE1.head(5))

# Distribución de diagnósticos
if "diagnosis" in df_result_train_fd1_v5_ITE1.columns:
    print("\nDistribución de diagnosis:")
    print(df_result_train_fd1_v5_ITE1["diagnosis"].value_counts(dropna=False).sort_index())

# Modelos sugeridos
if "model_suggested" in df_result_train_fd1_v5_ITE1.columns:
    print("\nModelos sugeridos (top 10):")
    print(df_result_train_fd1_v5_ITE1["model_suggested"].value_counts().head(10))

# Resumen de KPIs clave (si existen en las columnas)
kpi_cols = [
    "zero_ratio","outlier_rate","slope","seasonality_strength",
    "acf12","ADI_n_over_d","neg_share","effective_value_ratio"
]
kpi_existentes = [c for c in kpi_cols if c in df_result_train_fd1_v5_ITE1.columns]
if kpi_existentes:
    print("\nResumen estadístico de KPIs clave:")
    print(df_result_train_fd1_v5_ITE1[kpi_existentes].describe())

Shape: (2429, 36)

Columnas:
['ID_BUILDING', 'FM_COST_TYPE', 'date_min', 'date_max', 'n_months', 'sparsity', 'zero_ratio', 'neg_share', 'mean', 'std', 'cv', 'acf1', 'acf12', 'seasonality_strength', 'outlier_rate', 'slope', 'ADI_n_over_d', 'ADI_mean_gap', 'demand_months', 'cv2_positivos', 'max_zero_run', 'p_occurrence', 'occurrence_slope', 'has_min_history', 'effective_value_ratio', 'passes_effective_ratio', 'total_sum', 'passes_min_sum', 'eligible_all', 'COUNTRY_DEF', 'ID_REGION_GRUPO', 'TIPO_USO', 'SUPPLIER_TYPE_MOD_2', 'FM_RESPONSIBLE_MOD', 'diagnosis', 'model_suggested']

Info del dataframe:
<class 'pandas.core.frame.DataFrame'>
RangeIndex: 2429 entries, 0 to 2428
Data columns (total 36 columns):
 #   Column                  Non-Null Count  Dtype
---  ------                  --------------  -----
 0   ID_BUILDING             2429 non-null   Int64
 1   FM_COST_TYPE            2429 non-null   object
 2   date_min                2429 non-null   datetime64[ns]
 3   date_max                2429 non-null   datetime64[ns]
 4   n_months                2429 non-null   int64
 5   sparsity                2429 non-null   float64
 6   zero_ratio              2429 non-null   float64
 7   neg_share               2429 non-null   float64
 8   mean                    2429 non-null   float64
 9   std                     2241 non-null   float64
 10  cv                      2210 non-null   float64
 11  acf1                    2131 non-null   float64
 12  acf12                   1874 non-null   float64
 13  seasonality_strength    1741 non-null   float64
 14  outlier_rate            2429 non-null   float64
 15  slope                   2429 non-null   float64
 16  ADI_n_over_d            2429 non-null   float64
 17  ADI_mean_gap            2429 non-null   float64
 18  demand_months           2429 non-null   int64
 19  cv2_positivos           2135 non-null   float64
 20  max_zero_run            2429 non-null   int64
 21  p_occurrence            2429 non-null   float64
 22  occurrence_slope        2429 non-null   float64
 23  has_min_history         2429 non-null   int64
 24  effective_value_ratio   2429 non-null   float64
 25  passes_effective_ratio  2429 non-null   int64
 26  total_sum               2429 non-null   float64
 27  passes_min_sum          2429 non-null   int64
 28  eligible_all            2429 non-null   int64
 29  COUNTRY_DEF             2429 non-null   object
 30  ID_REGION_GRUPO         2429 non-null   object
 31  TIPO_USO                2429 non-null   object
 32  SUPPLIER_TYPE_MOD_2     2429 non-null   object
 33  FM_RESPONSIBLE_MOD      2429 non-null   object
 34  diagnosis               2429 non-null   object
 35  model_suggested         2429 non-null   object
dtypes: Int64(1), datetime64[ns](2), float64(18), int64(7), object(8)
memory usage: 685.7+ KB
None

Primeras filas:
   ID_BUILDING           FM_COST_TYPE   date_min   date_max  n_months  \
0            2  Eficiencia Energética 2021-12-01 2021-12-01         1
1            9  Eficiencia Energética 2022-10-01 2023-12-01        15
2           18  Eficiencia Energética 2021-01-01 2023-05-01        29
3           57  Eficiencia Energética 2021-04-01 2021-04-01         1
4          126  Eficiencia Energética 2021-04-01 2023-11-01        32

   sparsity  zero_ratio  neg_share       mean          std  ...  total_sum  \
0   1.00000     1.00000        0.0     0.0000          NaN  ...       0.00
1   1.00000     1.00000        0.0     0.0000     0.000000  ...       0.00
2   1.00000     1.00000        0.0     0.0000     0.000000  ...       0.00
3   0.00000     0.00000        0.0   218.7800          NaN  ...     218.78
4   0.84375     0.84375        0.0  1061.8125  2506.910653  ...   33978.00

   passes_min_sum  eligible_all  COUNTRY_DEF  ID_REGION_GRUPO      TIPO_USO  \
0               1             0       España                2      Oficinas
1               1             0       España                2      Oficinas
2               1             0       España                2      Oficinas
3               1             0       España                2  Subarrendado
4               1             0       España               12         Bingo

   SUPPLIER_TYPE_MOD_2     FM_RESPONSIBLE_MOD           diagnosis  \
0              INTERNO  Eficiencia Energética  escape_no_elegible
1              INTERNO  Eficiencia Energética  escape_no_elegible
2              INTERNO  Eficiencia Energética  escape_no_elegible
3              EXTERNO  Eficiencia Energética  escape_no_elegible
4              EXTERNO  Eficiencia Energética  escape_no_elegible

          model_suggested
0  pooling_global|naive_0
1  pooling_global|naive_0
2  pooling_global|naive_0
3  pooling_global|naive_0
4  pooling_global|naive_0

[5 rows x 36 columns]

Distribución de diagnosis:
diagnosis
escape_no_elegible        870
estable                   500
estacional                489
intermitente              205
intermitente_creciente    274
picos_outliers             60
tendencia                  31
Name: count, dtype: int64

Modelos sugeridos (top 10):
model_suggested
pooling_global|naive_0                       870
ETS_simple|ARIMA_basico                      500
ETS estacional|SARIMA                        489
Croston|SBA                                  479
transform_log1p+winsor|robusto(ETS/TBATS)     60
Holt(ETS)|ARIMA_drift                         31
Name: count, dtype: int64

Resumen estadístico de KPIs clave:
        zero_ratio  outlier_rate        slope  seasonality_strength  \
count  2429.000000   2429.000000  2429.000000           1741.000000
mean      0.358087      0.036275    20.647309              0.610967
std       0.368764      0.079214    86.772806              0.293074
min       0.000000      0.000000 -1191.035000              0.000000
25%       0.000000      0.000000    -0.141832              0.410353
50%       0.166667      0.000000     1.550636              0.646491
75%       0.750000      0.028571    15.062425              0.859926
max       1.000000      0.484848   779.582005              1.000000

             acf12  ADI_n_over_d    neg_share  effective_value_ratio
count  1874.000000        2429.0  2429.000000            2429.000000
mean      0.153561           inf     0.000418               0.641495
std       0.437260           NaN     0.010464               0.368543
min      -1.000000           1.0     0.000000               0.000000
25%      -0.156602           1.0     0.000000               0.250000
50%       0.043378           1.2     0.000000               0.833333
75%       0.518020           4.0     0.000000               1.000000
max       1.000000           inf     0.500000               1.000000

# Preparamos el diagnóstico corto y lo unimos al panel
cols_diag = ["ID_BUILDING","FM_COST_TYPE","diagnosis","model_suggested"]
missing_diag = set(cols_diag) - set(df_result_train_fd1_v5_ITE1.columns)
assert not missing_diag, f"Faltan columnas en df_result_train_fd1_v5_ITE1: {missing_diag}"

# Nos quedamos con las columnas mínimas necesarias del diagnóstico
diag_short = df_result_train_fd1_v5_ITE1[cols_diag].drop_duplicates()

# Comprobamos que el panel de entrenamiento tiene las columnas requeridas
panel_cols_req = ["ID_BUILDING","FM_COST_TYPE","FECHA","cost_float_mod"]
missing_panel = set(panel_cols_req) - set(df_train_fd1_v5_ITE1.columns)
assert not missing_panel, f"Faltan columnas en df_train_fd1_v5_ITE1: {missing_panel}"

# Unimos el panel mensual con el diagnóstico para que cada fila mensual tenga su modelo sugerido
df_panel_model = (
    df_train_fd1_v5_ITE1
    .merge(diag_short, on=["ID_BUILDING","FM_COST_TYPE"], how="inner")
)

# Normalizamos el modelo sugerido a un grupo/sufijo simple

# Definimos una función que mapea cada modelo sugerido a una etiqueta corta (ETS, SARIMA, ARIMA, etc.)
def normalizar_modelo(model_suggested: str) -> str:
    if not isinstance(model_suggested, str):
        return "OTROS"
    s = model_suggested.lower()

    if "pool" in s or "naive" in s:
        return "POOLING"
    if "croston" in s or "sba" in s:
        return "CROSTON"
    if "sarima" in s:
        return "SARIMA"
    if "holt" in s:
        return "HOLT"
    if "arima" in s:
        return "ARIMA"
    if "robusto" in s or "winsor" in s or "tbats" in s:
        return "ROBUSTO"
    if "ets" in s or "holt-winters" in s:
        return "ETS"
    return "OTROS"

# Creamos la nueva columna de grupo de modelo
df_panel_model["model_group"] = df_panel_model["model_suggested"].map(normalizar_modelo)

# === 3) Partimos el panel por grupo y generamos dataframes con el nombre solicitado ===
base_name = "df_train_fd1_v5_ITE1"
grupos = sorted(df_panel_model["model_group"].dropna().unique())

created = {}
for g in grupos:
    # Filtramos el panel para cada grupo de modelo
    df_g = df_panel_model[df_panel_model["model_group"] == g].copy()
    # Construimos el nombre dinámicamente (ej. df_train_fd1_v5_ITE1_ETS)
    var_name = f"{base_name}_{g}"
    # Creamos el dataframe en el entorno global con ese nombre
    globals()[var_name] = df_g
    # Guardamos referencia para mostrar resumen
    created[g] = (var_name, df_g.shape)

# === 4) Visualizamos los resultados de manera agrupada ===
print("Sub-dataframes creados por modelo:")
for g,(name,shape) in created.items():
    # Contamos las parejas únicas en cada sub-bloque para entender su tamaño
    n_pairs = df_panel_model[df_panel_model["model_group"]==g][["ID_BUILDING","FM_COST_TYPE"]].drop_duplicates().shape[0]
    print(f" - {name:32s}: {shape} | parejas únicas = {n_pairs}")

# También mostramos la distribución de grupos por número de series (parejas únicas)
print("\nDistribución de grupos (por parejas únicas):")
pairs_by_group = (
    df_panel_model[["ID_BUILDING","FM_COST_TYPE","model_group"]]
    .drop_duplicates()
    .groupby("model_group").size().rename("series").sort_values(ascending=False)
)
print(pairs_by_group)

Sub-dataframes creados por modelo:
 - df_train_fd1_v5_ITE1_ARIMA      : (17188, 14) | parejas únicas = 500
 - df_train_fd1_v5_ITE1_CROSTON    : (14460, 14) | parejas únicas = 479
 - df_train_fd1_v5_ITE1_HOLT       : (899, 14) | parejas únicas = 31
 - df_train_fd1_v5_ITE1_POOLING    : (13647, 14) | parejas únicas = 870
 - df_train_fd1_v5_ITE1_ROBUSTO    : (2065, 14) | parejas únicas = 60
 - df_train_fd1_v5_ITE1_SARIMA     : (17207, 14) | parejas únicas = 489

Distribución de grupos (por parejas únicas):
model_group
POOLING    870
ARIMA      500
SARIMA     489
CROSTON    479
ROBUSTO     60
HOLT        31
Name: series, dtype: int64

# === GUARDAMOS LOS DATAFRAMES SEGÚN LA SEGREGACIÓN POR MODELO ===

# Definimos la carpeta de salida en Drive
OUT_DIR = "/content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO"
os.makedirs(OUT_DIR, exist_ok=True)

# Recorremos cada grupo de modelos creado en el paso anterior
for g,(var_name,shape) in created.items():
    # Recuperamos el dataframe del entorno global
    df_g = globals()[var_name]

    # Definimos las rutas de salida con el sufijo correspondiente
    csv_path  = os.path.join(OUT_DIR, f"{var_name}.csv")
    xlsx_path = os.path.join(OUT_DIR, f"{var_name}.xlsx")

    # Guardamos en CSV
    df_g.to_csv(csv_path, sep=";", index=False)

    # Guardamos en Excel
    try:
        df_g.to_excel(xlsx_path, index=False)
    except Exception as e:
        print(f"[AVISO] No se pudo guardar {var_name} en Excel: {type(e).__name__}: {e}")

    # Mostramos un mensaje de confirmación por cada dataframe
    print(f"[OK] Guardado {var_name} | filas: {shape[0]} | columnas: {shape[1]}")
    print(" - CSV (;) :", csv_path)
    print(" - Excel:", xlsx_path)

[OK] Guardado df_train_fd1_v5_ITE1_ARIMA | filas: 17188 | columnas: 14
 - CSV (;) : /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/df_train_fd1_v5_ITE1_ARIMA.csv
 - Excel: /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/df_train_fd1_v5_ITE1_ARIMA.xlsx
[OK] Guardado df_train_fd1_v5_ITE1_CROSTON | filas: 14460 | columnas: 14
 - CSV (;) : /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/df_train_fd1_v5_ITE1_CROSTON.csv
 - Excel: /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/df_train_fd1_v5_ITE1_CROSTON.xlsx
[OK] Guardado df_train_fd1_v5_ITE1_HOLT | filas: 899 | columnas: 14
 - CSV (;) : /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/df_train_fd1_v5_ITE1_HOLT.csv
 - Excel: /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/df_train_fd1_v5_ITE1_HOLT.xlsx
[OK] Guardado df_train_fd1_v5_ITE1_POOLING | filas: 13647 | columnas: 14
 - CSV (;) : /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/df_train_fd1_v5_ITE1_POOLING.csv
 - Excel: /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/df_train_fd1_v5_ITE1_POOLING.xlsx
[OK] Guardado df_train_fd1_v5_ITE1_ROBUSTO | filas: 2065 | columnas: 14
 - CSV (;) : /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/df_train_fd1_v5_ITE1_ROBUSTO.csv
 - Excel: /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/df_train_fd1_v5_ITE1_ROBUSTO.xlsx
[OK] Guardado df_train_fd1_v5_ITE1_SARIMA | filas: 17207 | columnas: 14
 - CSV (;) : /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/df_train_fd1_v5_ITE1_SARIMA.csv
 - Excel: /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/df_train_fd1_v5_ITE1_SARIMA.xlsx

# Definimos los nombres clave y los dejamos explícitos
COL_IDB, COL_FMCT = "ID_BUILDING", "FM_COST_TYPE"
COL_FECHA, COL_COST = "FECHA", "cost_float_mod"
CTX_COLS = ["COUNTRY_DEF","ID_REGION_GRUPO","TIPO_USO","SUPPLIER_TYPE_MOD_2","FM_RESPONSIBLE_MOD"]

# Partimos del dataframe de hechos 2024 y añadimos FECHA a partir de YEAR/MONTH

# Creamos FECHA como el primer día de cada mes de 2024 (nos basamos en YEAR y MONTH del propio DF)
df_real_2024 = df_fd1_v5_Real_ITE1_2024.copy()
df_real_2024[COL_FECHA] = pd.to_datetime(
    dict(year=df_real_2024["YEAR"], month=df_real_2024["MONTH"], day=1)
)

# Nos aseguramos de que el coste sea numérico
df_real_2024[COL_COST] = pd.to_numeric(df_real_2024[COL_COST], errors="coerce").fillna(0.0)

# Realizamos las agrupaciones por ID_BUILDING y FM_COST_TYPE con costes agregados por mes (2024)

# Agregamos el coste mensual por pareja y FECHA (sumatorio)
test_agg_2024 = (
    df_real_2024
    .groupby([COL_IDB, COL_FMCT, COL_FECHA], as_index=False)[COL_COST]
    .sum()
)

# Rellenamos los meses faltantes por pareja y reindexamos con nuestra función reindex_group

# Preparamos el dataframe para que cada grupo tenga exactamente las columnas que espera reindex_group
test_agg_2024_for_reindex = test_agg_2024[[COL_IDB, COL_FMCT, COL_FECHA, COL_COST]].copy()

# Aplicamos nuestra función reindex_group por cada pareja (entre su primer y último mes observado en 2024)
# Nota: reindex_group(g) espera columnas: "FECHA", "cost_float_mod", "ID_BUILDING", "FM_COST_TYPE"
df_test_2024 = (
    test_agg_2024_for_reindex
    .groupby([COL_IDB, COL_FMCT], as_index=False, group_keys=False)
    .apply(reindex_group)
    .reset_index(drop=True)
)

# Para cada pareja, calculamos la moda de variables de contexto usando 2021–2023 (panel train)

# Definimos una función de moda simple (tomamos la categoría más frecuente por pareja en el train)
def _mode_safe(s: pd.Series):
    vc = s.dropna().value_counts()
    return vc.index[0] if len(vc) else np.nan

# Calculamos la moda por pareja a partir del panel de train (2021–2023)
ctx_modes = (
    df_train_fd1_v5_ITE1[[COL_IDB, COL_FMCT] + CTX_COLS]
    .groupby([COL_IDB, COL_FMCT], as_index=False)
    .agg({c: _mode_safe for c in CTX_COLS})
)

# Ensamblamos df_test_fd1_v5_ITE1 uniendo la agregación mensual completa con el contexto (moda)

df_test_fd1_v5_ITE1 = df_test_2024.merge(ctx_modes, on=[COL_IDB, COL_FMCT], how="left")

# Añadimos columnas YEAR y MONTH derivadas de FECHA

df_test_fd1_v5_ITE1["YEAR"]  = df_test_fd1_v5_ITE1[COL_FECHA].dt.year.astype(int)
df_test_fd1_v5_ITE1["MONTH"] = df_test_fd1_v5_ITE1[COL_FECHA].dt.month.astype(int)

# Dejamos nombres claros y orden de columnas práctico

ordered_cols = [COL_IDB, COL_FMCT, COL_FECHA, "YEAR", "MONTH", COL_COST] + CTX_COLS
ordered_cols += [c for c in df_test_fd1_v5_ITE1.columns if c not in ordered_cols]
df_test_fd1_v5_ITE1 = df_test_fd1_v5_ITE1[ordered_cols]

# Conciliamos parejas ID_BUILDING y FM_COST_TYPE entre TEST (2024) y TRAIN (2021–2023)

# Queremos identificar si podremos predecir el 100% o si hay parejas nuevas en 2024
pairs_test  = df_test_fd1_v5_ITE1[[COL_IDB, COL_FMCT]].drop_duplicates()
pairs_train = df_train_fd1_v5_ITE1[[COL_IDB, COL_FMCT]].drop_duplicates()

# Parejas nuevas en 2024 (aparecen en test pero no en train)
pairs_new_in_test = (
    pairs_test
    .merge(pairs_train, on=[COL_IDB, COL_FMCT], how="left", indicator=True)
    .query("_merge == 'left_only'")
    .drop(columns="_merge")
)

# Parejas del train sin hechos en 2024 (están en train pero no aparecen en test)
pairs_missing_in_test = (
    pairs_train
    .merge(pairs_test, on=[COL_IDB, COL_FMCT], how="left", indicator=True)
    .query("_merge == 'left_only'")
    .drop(columns="_merge")
)

print("\n=== Conciliación de parejas train vs test (2024) ===")
print(f"Parejas en TRAIN (2021–2023): {len(pairs_train)}")
print(f"Parejas en TEST  (2024)     : {len(pairs_test)}")
print(f"Parejas NUEVAS en TEST (no estaban en TRAIN): {len(pairs_new_in_test)}")
if len(pairs_new_in_test):
    print(pairs_new_in_test.head(10))
print(f"Parejas sin hechos en 2024 (están en TRAIN pero no en TEST): {len(pairs_missing_in_test)}")
if len(pairs_missing_in_test):
    print(pairs_missing_in_test.head(10))

# Realizamos una comprobación rápida del df_test_fd1_v5_ITE1

# Verificamos forma, rango temporal y duplicados (por pareja y mes)
print("\n=== Validación rápida df_test_fd1_v5_ITE1 ===")
print("Shape:", df_test_fd1_v5_ITE1.shape)
print("Rango FECHA:", df_test_fd1_v5_ITE1[COL_FECHA].min().date(), "→", df_test_fd1_v5_ITE1[COL_FECHA].max().date())

dup_mask = df_test_fd1_v5_ITE1.duplicated(subset=[COL_IDB, COL_FMCT, COL_FECHA], keep=False)
print("Duplicados por (ID_BUILDING, FM_COST_TYPE, FECHA):", int(dup_mask.sum()))

print("\nMuestra (5 filas):")
print(df_test_fd1_v5_ITE1.head(5))

=== Conciliación de parejas train vs test (2024) ===
Parejas en TRAIN (2021–2023): 2429
Parejas en TEST  (2024)     : 2692
Parejas NUEVAS en TEST (no estaban en TRAIN): 461
     ID_BUILDING           FM_COST_TYPE
32           104  Eficiencia Energética
39           105  Eficiencia Energética
46           116  Eficiencia Energética
53           117  Eficiencia Energética
57           117                  Obras
61           118  Eficiencia Energética
68           121  Eficiencia Energética
72           121                  Obras
76           122  Eficiencia Energética
108          131  Eficiencia Energética
Parejas sin hechos en 2024 (están en TRAIN pero no en TEST): 198
    ID_BUILDING           FM_COST_TYPE
0             2  Eficiencia Energética
24           57  Eficiencia Energética
25           57              Licencias
27           57       Mtto. Correctivo
28           57        Servicios Ctto.
29           57        Servicios Extra
30           57            Suministros
31           59              Licencias
33           59       Mtto. Correctivo
34           59                  Obras

=== Validación rápida df_test_fd1_v5_ITE1 ===
Shape: (23750, 11)
Rango FECHA: 2024-01-01 → 2024-12-01
Duplicados por (ID_BUILDING, FM_COST_TYPE, FECHA): 0

Muestra (5 filas):
   ID_BUILDING     FM_COST_TYPE      FECHA  YEAR  MONTH  cost_float_mod  \
0            2        Licencias 2024-12-01  2024     12          202.80
1            2  Mtto. Contratos 2024-01-01  2024      1         1874.00
2            2  Mtto. Contratos 2024-02-01  2024      2            0.00
3            2  Mtto. Contratos 2024-03-01  2024      3            0.00
4            2  Mtto. Contratos 2024-04-01  2024      4          175.58

  COUNTRY_DEF ID_REGION_GRUPO  TIPO_USO SUPPLIER_TYPE_MOD_2 FM_RESPONSIBLE_MOD
0      España               2  Oficinas             INTERNO          Licencias
1      España               2  Oficinas             EXTERNO      Mantenimiento
2      España               2  Oficinas             EXTERNO      Mantenimiento
3      España               2  Oficinas             EXTERNO      Mantenimiento
4      España               2  Oficinas             EXTERNO      Mantenimiento

print("Shape df_train_fd1_v5_ITE1:", df_train_fd1_v5_ITE1.shape)
print("Rango FECHA:", df_train_fd1_v5_ITE1[COL_FECHA].min().date(), "→", df_train_fd1_v5_ITE1[COL_FECHA].max().date())
print("\nMuestra (5 filas):")
print(df_train_fd1_v5_ITE1.head(5))

Shape df_train_fd1_v5_ITE1: (65466, 11)
Rango FECHA: 2021-01-01 → 2023-12-01

Muestra (5 filas):
   ID_BUILDING           FM_COST_TYPE      FECHA  YEAR  MONTH  cost_float_mod  \
0            2  Eficiencia Energética 2021-12-01  2021     12            0.00
1            2              Licencias 2021-01-01  2021      1         1145.46
2            2              Licencias 2021-02-01  2021      2           55.95
3            2              Licencias 2021-03-01  2021      3            0.00
4            2              Licencias 2021-04-01  2021      4            0.00

  COUNTRY_DEF ID_REGION_GRUPO  TIPO_USO SUPPLIER_TYPE_MOD_2  \
0      España               2  Oficinas             INTERNO
1      España               2  Oficinas             INTERNO
2      España               2  Oficinas             INTERNO
3      España               2  Oficinas             INTERNO
4      España               2  Oficinas             INTERNO

      FM_RESPONSIBLE_MOD
0  Eficiencia Energética
1              Licencias
2              Licencias
3              Licencias
4              Licencias

# === TEST → Asignamos modelo sugerido desde el train y separamos en grupos ===

# Nos quedamos del diagnóstico del train con lo esencial para mapear al test
cols_diag = ["ID_BUILDING","FM_COST_TYPE","diagnosis","model_suggested"]
diag_short_train = df_result_train_fd1_v5_ITE1[cols_diag].drop_duplicates()

# Unimos el test con el diagnóstico del train (por pareja) para traer el modelo_sugerido
#    - Las parejas nuevas en 2024 quedarán con NaN en 'model_suggested'
#  (puesto no existían en 2021–2023 por lo que no van a tener predicción)
df_test_model = df_test_fd1_v5_ITE1.merge(
    diag_short_train, on=["ID_BUILDING","FM_COST_TYPE"], how="left"
)

# Normalizamos el modelo sugerido a un sufijo de grupo (ETS, SARIMA, ARIMA, POOLING, CROSTON, HOLT, ROBUSTO, OTROS)
def normalizar_modelo(model_suggested: str) -> str:
    if not isinstance(model_suggested, str):
        return "SIN_MODELO"  # parejas nuevas en 2024 (no estaban en train)
    s = model_suggested.lower()
    if "pool" in s or "naive" in s:
        return "POOLING"
    if "croston" in s or "sba" in s:
        return "CROSTON"
    if "sarima" in s:
        return "SARIMA"
    if "holt" in s:
        return "HOLT"
    if "arima" in s:
        return "ARIMA"
    if "robusto" in s or "winsor" in s or "tbats" in s:
        return "ROBUSTO"
    if "ets" in s or "holt-winters" in s:
        return "ETS"
    return "OTROS"

# Creamos la columna de grupo de modelo para el test
df_test_model["model_group"] = df_test_model["model_suggested"].map(normalizar_modelo)

# Separamos el test en sub-dataframes por 'model_group' (mismos sufijos que en train)
base_name_test = "df_test_fd1_v5_ITE1"
grupos_test = sorted(df_test_model["model_group"].dropna().unique())

created_test = {}
for g in grupos_test:
    # Filtramos el panel test para cada grupo de modelo
    df_g = df_test_model[df_test_model["model_group"] == g].copy()
    # Construimos el nombre dinámico (ej. df_test_fd1_v5_ITE1_ETS)
    var_name = f"{base_name_test}_{g}"
    # Creamos el dataframe en el entorno global con ese nombre
    globals()[var_name] = df_g
    # Guardamos referencia para el resumen
    created_test[g] = (var_name, df_g.shape)

# Mostramos un resumen para validar
print("Sub-dataframes de TEST creados por modelo:")
for g,(name,shape) in created_test.items():
    # Contamos las parejas únicas en cada sub-bloque
    n_pairs = globals()[name][["ID_BUILDING","FM_COST_TYPE"]].drop_duplicates().shape[0]
    print(f" - {name:32s}: {shape} | parejas únicas = {n_pairs}")

# Visualizamos las parejas NUEVAS (SIN_MODELO) para decidir política (p.ej. POOLING/NAIVE o marcar para revisión)
if "SIN_MODELO" in created_test:
    name_new, shape_new = created_test["SIN_MODELO"]
    nuevas_pairs = globals()[name_new][["ID_BUILDING","FM_COST_TYPE"]].drop_duplicates()
    print(f"\n[AVISO] Parejas nuevas en 2024 sin modelo del train (SIN_MODELO): {len(nuevas_pairs)}")
    print(nuevas_pairs.head(10))

# 7) (Opcional) Si queremos alinear exactamente la cobertura del test con el train,
#    filtramos el test para solo parejas presentes en el train:
# df_test_model_in_train = df_test_model.merge(
#     df_train_fd1_v5_ITE1[["ID_BUILDING","FM_COST_TYPE"]].drop_duplicates(),
#     on=["ID_BUILDING","FM_COST_TYPE"], how="inner"
# )
# (Entonces repetiríamos la separación por 'model_group' con este subconjunto)

Sub-dataframes de TEST creados por modelo:
 - df_test_fd1_v5_ITE1_ARIMA       : (5835, 14) | parejas únicas = 491
 - df_test_fd1_v5_ITE1_CROSTON     : (4680, 14) | parejas únicas = 472
 - df_test_fd1_v5_ITE1_HOLT        : (368, 14) | parejas únicas = 31
 - df_test_fd1_v5_ITE1_POOLING     : (4881, 14) | parejas únicas = 696
 - df_test_fd1_v5_ITE1_ROBUSTO     : (704, 14) | parejas únicas = 60
 - df_test_fd1_v5_ITE1_SARIMA      : (5648, 14) | parejas únicas = 481
 - df_test_fd1_v5_ITE1_SIN_MODELO  : (1634, 14) | parejas únicas = 461

[AVISO] Parejas nuevas en 2024 sin modelo del train (SIN_MODELO): 461
      ID_BUILDING           FM_COST_TYPE
289           104  Eficiencia Energética
359           105  Eficiencia Energética
428           116  Eficiencia Energética
486           117  Eficiencia Energética
520           117                  Obras
556           118  Eficiencia Energética
621           121  Eficiencia Energética
656           121                  Obras
690           122  Eficiencia Energética
1026          131  Eficiencia Energética

# Filas totales y parejas únicas en el train
n_rows_train = df_train_fd1_v5_ITE1.shape[0]
n_pairs_train = df_train_fd1_v5_ITE1[[COL_IDB, COL_FMCT]].drop_duplicates().shape[0]

print("Filas totales en TRAIN:", n_rows_train)
print("Parejas únicas en TRAIN:", n_pairs_train)
print("Promedio de meses por pareja:", round(n_rows_train / n_pairs_train, 2))

# Distribución exacta: nº de meses por pareja
meses_por_pareja = (
    df_train_fd1_v5_ITE1
    .groupby([COL_IDB, COL_FMCT])[COL_FECHA]
    .nunique()
    .reset_index(name="n_meses")
)

print("\nResumen de nº de meses por pareja en el TRAIN:")
print(meses_por_pareja["n_meses"].describe())

# Opcional: ver cuántas parejas tienen exactamente 36 meses
n_full = (meses_por_pareja["n_meses"] == 36).sum()
print(f"\nParejas con los 36 meses completos: {n_full} de {n_pairs_train}")

Filas totales en TRAIN: 65466
Parejas únicas en TRAIN: 2429
Promedio de meses por pareja: 26.95

Resumen de nº de meses por pareja en el TRAIN:
count    2429.000000
mean       26.951832
std        11.864442
min         1.000000
25%        19.000000
50%        33.000000
75%        36.000000
max        36.000000
Name: n_meses, dtype: float64

Parejas con los 36 meses completos: 925 de 2429

# === GUARDAMOS TEST COMPLETO + SUB-DATAFRAMES POR GRUPO (Excel + CSV con ';') ===

# Definimos la carpeta de salida en Drive
OUT_DIR = "/content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO"
os.makedirs(OUT_DIR, exist_ok=True)

# (Preventivo) Función simple para quitar tz si alguna columna datetime la trae con zona horaria
def _sanitize_for_excel(df: pd.DataFrame) -> pd.DataFrame:
    df2 = df.copy()
    for c in df2.columns:
        if is_datetime64tz_dtype(df2[c]):
            try:
                df2[c] = df2[c].dt.tz_convert(None)
            except Exception:
                df2[c] = df2[c].dt.tz_localize(None)
    return df2

# Guardamos el TEST completo con el modelo sugerido

# Nos aseguramos de que exista df_test_model (test + diagnosis/model_suggested + model_group)
df_master_test = df_test_model.copy()

# Rutas del master
csv_master_path  = os.path.join(OUT_DIR, "df_test_fd1_v5_ITE1__master.csv")
xlsx_master_path = os.path.join(OUT_DIR, "df_test_fd1_v5_ITE1__master.xlsx")

# Guardamos CSV (separador ';')
# Comentario: usamos ';' para compatibilidad con locales que usan coma decimal
df_master_test.to_csv(csv_master_path, index=False, sep=';')

# Guardamos Excel (una hoja)
try:
    _sanitize_for_excel(df_master_test).to_excel(xlsx_master_path, index=False)
except Exception as e:
    print(f"[AVISO] No se pudo guardar el Excel master: {type(e).__name__}: {e}")

print(f"[OK] Guardado TEST master:")
print(" - CSV (;) :", csv_master_path)
print(" - Excel   :", xlsx_master_path)

# Guardamos cada sub-dataframe de TEST por grupo (created_test)

# created_test: dict con {grupo: (var_name, shape)} creado previamente
for g, (var_name, shape) in created_test.items():
    df_g = globals()[var_name].copy()

    # Definimos rutas por grupo
    csv_path  = os.path.join(OUT_DIR, f"{var_name}.csv")
    xlsx_path = os.path.join(OUT_DIR, f"{var_name}.xlsx")

    # Guardamos CSV con separador ';'
    df_g.to_csv(csv_path, index=False, sep=';')

    # Guardamos Excel individual
    try:
        _sanitize_for_excel(df_g).to_excel(xlsx_path, index=False)
    except Exception as e:
        print(f"[AVISO] No se pudo guardar {var_name} en Excel: {type(e).__name__}: {e}")

    # Mostramos confirmación
    print(f"[OK] Guardado {var_name} | filas: {shape[0]} | columnas: {shape[1]}")
    print(" - CSV (;) :", csv_path)
    print(" - Excel   :", xlsx_path)

[OK] Guardado TEST master:
 - CSV (;) : /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/df_test_fd1_v5_ITE1__master.csv
 - Excel   : /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/df_test_fd1_v5_ITE1__master.xlsx
[OK] Guardado df_test_fd1_v5_ITE1_ARIMA | filas: 5835 | columnas: 14
 - CSV (;) : /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/df_test_fd1_v5_ITE1_ARIMA.csv
 - Excel   : /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/df_test_fd1_v5_ITE1_ARIMA.xlsx
[OK] Guardado df_test_fd1_v5_ITE1_CROSTON | filas: 4680 | columnas: 14
 - CSV (;) : /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/df_test_fd1_v5_ITE1_CROSTON.csv
 - Excel   : /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/df_test_fd1_v5_ITE1_CROSTON.xlsx
[OK] Guardado df_test_fd1_v5_ITE1_HOLT | filas: 368 | columnas: 14
 - CSV (;) : /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/df_test_fd1_v5_ITE1_HOLT.csv
 - Excel   : /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/df_test_fd1_v5_ITE1_HOLT.xlsx
[OK] Guardado df_test_fd1_v5_ITE1_POOLING | filas: 4881 | columnas: 14
 - CSV (;) : /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/df_test_fd1_v5_ITE1_POOLING.csv
 - Excel   : /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/df_test_fd1_v5_ITE1_POOLING.xlsx
[OK] Guardado df_test_fd1_v5_ITE1_ROBUSTO | filas: 704 | columnas: 14
 - CSV (;) : /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/df_test_fd1_v5_ITE1_ROBUSTO.csv
 - Excel   : /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/df_test_fd1_v5_ITE1_ROBUSTO.xlsx
[OK] Guardado df_test_fd1_v5_ITE1_SARIMA | filas: 5648 | columnas: 14
 - CSV (;) : /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/df_test_fd1_v5_ITE1_SARIMA.csv
 - Excel   : /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/df_test_fd1_v5_ITE1_SARIMA.xlsx
[OK] Guardado df_test_fd1_v5_ITE1_SIN_MODELO | filas: 1634 | columnas: 14
 - CSV (;) : /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/df_test_fd1_v5_ITE1_SIN_MODELO.csv
 - Excel   : /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/df_test_fd1_v5_ITE1_SIN_MODELO.xlsx

# Ruta del notebook .ipynb que vamos a convertir

RUTA_NOTEBOOK = "/content/drive/MyDrive/PROYECTO_NEITH/DOCUMENTOS_PROYECTOS/Neith.ipynb"

# Carpeta de salida y nombre del HTML
OUTPUT_DIR = "/content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO"
OUTPUT_HTML = "Neith_notebook_v5_Pre_Modelizacion_ST.html"

# Aseguramos carpeta de salida (ya en cabecera)
# import os
os.makedirs(OUTPUT_DIR, exist_ok=True)

# Instalamos nbconvert (ya en cabecera)
# !pip install -q "nbconvert>=7.0.0"

# Convertimos notebook a HTML y lo guardamos con el nombre indicado en la carpeta de salida
!jupyter nbconvert --to html "$RUTA_NOTEBOOK" --output "$OUTPUT_HTML" --output-dir "$OUTPUT_DIR"

# Verificamos
salida = os.path.join(OUTPUT_DIR, OUTPUT_HTML)
print("Archivo HTML generado en:", salida, "\nExiste:", os.path.exists(salida))

[NbConvertApp] Converting notebook /content/drive/MyDrive/PROYECTO_NEITH/DOCUMENTOS_PROYECTOS/Neith.ipynb to html
[NbConvertApp] WARNING | Alternative text is missing on 1 image(s).
[NbConvertApp] Writing 1530363 bytes to /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/Neith_notebook_v5_Pre_Modelizacion_ST.html
Archivo HTML generado en: /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/Neith_notebook_v5_Pre_Modelizacion_ST.html
Existe: True

def mae(y_true, y_pred):
    y_true = np.asarray(y_true, dtype=float)
    y_pred = np.asarray(y_pred, dtype=float)
    return np.mean(np.abs(y_true - y_pred))

def mape(y_true, y_pred):
    y_true = np.asarray(y_true, dtype=float)
    y_pred = np.asarray(y_pred, dtype=float)
    denom = np.where(y_true == 0, 1.0, y_true)
    return np.mean(np.abs((y_true - y_pred) / denom)) * 100.0

def wape(y_true, y_pred):
    denom = np.sum(np.abs(y_true))
    return np.nan if denom == 0 else np.sum(np.abs(y_true - y_pred)) / denom

def smape(y_true, y_pred):
    denom = (np.abs(y_true) + np.abs(y_pred))
    denom = np.where(denom == 0, 1, denom)
    return 100 * np.mean(2*np.abs(y_pred - y_true)/denom)

def mase(y_train, y_true, y_pred, m=1):
    """
    MASE respecto a Naive(m).
    m=1 -> naive simple; si prefieres estacional, usa m=12 para mensual.
    """
    y_train = np.asarray(y_train, dtype=float)
    if len(y_train) <= m:
        m = 1  # salvaguarda
    d = np.mean(np.abs(y_train[m:] - y_train[:-m]))
    d = 1.0 if (d is None or d == 0 or np.isnan(d)) else d
    return float(np.mean(np.abs(np.asarray(y_true) - np.asarray(y_pred))) / d)

# Definimos MASE sin salvaguardas: cuando no hay historia suficiente o d==0/NaN → NaN
def mase_no_fallback(y_train, y_true, y_pred, m=1):
    # Medimos MASE respecto a Naive(m); p.ej. m=1 naive simple, m=12 naive estacional mensual
    import numpy as np
    y_train = np.asarray(y_train, dtype=float)
    y_true  = np.asarray(y_true,  dtype=float)
    y_pred  = np.asarray(y_pred,  dtype=float)

    if len(y_train) <= m:
        return np.nan
    d = np.mean(np.abs(y_train[m:] - y_train[:-m]))
    if not np.isfinite(d) or d == 0:
        return np.nan
    return float(np.mean(np.abs(y_true - y_pred)) / d)

def mase12(y_train, y_true, y_pred):
    # Atajo para MASE estacional mensual, sin fallback
    return mase_no_fallback(y_train, y_true, y_pred, m=12)

def ensure_ts(df_pair):
    """Devuelve la serie mensual indexada por fecha, MS, float, ordenada."""
    s = df_pair[["FECHA", "cost_float_mod"]].copy()
    s["FECHA"] = pd.to_datetime(s["FECHA"])
    s = s.sort_values("FECHA").set_index("FECHA")["cost_float_mod"].asfreq("MS")
    # Relleno mínimo si hay NaN sueltos (no debería, pero por seguridad)
    if s.isna().any():
        s = s.fillna(0.0)
    return s.astype(float)

# Helper: obtenemos la serie mensual continua por pareja usando ensure_ts
def pair_series(df, pair):
    b, t = pair
    df_pair = df[(df["ID_BUILDING"] == b) & (df["FM_COST_TYPE"] == t)][["FECHA", "cost_float_mod"]].copy()
    return ensure_ts(df_pair)  # devuelve serie mensual indexada por fecha (MS), float, ordenada

# Definimos la carpeta de salida en Drive
OUT_DIR = "/content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO"

# --- OPCIÓN 1: Cargar desde CSV (con separador ';') ---
csv_path = f"{OUT_DIR}/df_train_fd1_v5_ITE1_ARIMA.csv"
df_train_fd1_v5_ITE1_ARIMA = pd.read_csv(csv_path, sep=';')

# --- OPCIÓN 2: Cargar desde Excel ---
# xlsx_path = f"{OUT_DIR}/df_train_fd1_v5_ITE1_ARIMA.xlsx"
# df_train_fd1_v5_ITE1_ARIMA = pd.read_excel(xlsx_path)

# Visualizamos forma y primeras filas para validar
print("Shape:", df_train_fd1_v5_ITE1_ARIMA.shape)
print(df_train_fd1_v5_ITE1_ARIMA.head())

Shape: (17188, 14)
   ID_BUILDING FM_COST_TYPE       FECHA  YEAR  MONTH  cost_float_mod  \
0            2  Suministros  2021-01-01  2021      1         6037.69
1            2  Suministros  2021-02-01  2021      2            0.00
2            2  Suministros  2021-03-01  2021      3         9889.47
3            2  Suministros  2021-04-01  2021      4         4908.41
4            2  Suministros  2021-05-01  2021      5            0.00

  COUNTRY_DEF ID_REGION_GRUPO  TIPO_USO SUPPLIER_TYPE_MOD_2  \
0      España               2  Oficinas             EXTERNO
1      España               2  Oficinas             EXTERNO
2      España               2  Oficinas             EXTERNO
3      España               2  Oficinas             EXTERNO
4      España               2  Oficinas             EXTERNO

      FM_RESPONSIBLE_MOD diagnosis          model_suggested model_group
0  Eficiencia Energética   estable  ETS_simple|ARIMA_basico       ARIMA
1  Eficiencia Energética   estable  ETS_simple|ARIMA_basico       ARIMA
2  Eficiencia Energética   estable  ETS_simple|ARIMA_basico       ARIMA
3  Eficiencia Energética   estable  ETS_simple|ARIMA_basico       ARIMA
4  Eficiencia Energética   estable  ETS_simple|ARIMA_basico       ARIMA

# Ruta de salida en Drive
OUT_DIR = "/content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO"

# --- Conjunto TEST para ITE1 (año 2024) ---
csv_test_path = f"{OUT_DIR}/df_test_fd1_v5_ITE1_ARIMA.csv"
df_test_fd1_v5_ITE1_ARIMA = pd.read_csv(csv_test_path, sep=';')

print("Shape TEST:", df_test_fd1_v5_ITE1_ARIMA.shape)
print(df_test_fd1_v5_ITE1_ARIMA.head())

Shape TEST: (5835, 14)
   ID_BUILDING FM_COST_TYPE       FECHA  YEAR  MONTH  cost_float_mod  \
0            2  Suministros  2024-01-01  2024      1         6735.70
1            2  Suministros  2024-02-01  2024      2         5385.96
2            2  Suministros  2024-03-01  2024      3         4701.77
3            2  Suministros  2024-04-01  2024      4         3183.83
4            2  Suministros  2024-05-01  2024      5          759.71

  COUNTRY_DEF ID_REGION_GRUPO  TIPO_USO SUPPLIER_TYPE_MOD_2  \
0      España               2  Oficinas             EXTERNO
1      España               2  Oficinas             EXTERNO
2      España               2  Oficinas             EXTERNO
3      España               2  Oficinas             EXTERNO
4      España               2  Oficinas             EXTERNO

      FM_RESPONSIBLE_MOD diagnosis          model_suggested model_group
0  Eficiencia Energética   estable  ETS_simple|ARIMA_basico       ARIMA
1  Eficiencia Energética   estable  ETS_simple|ARIMA_basico       ARIMA
2  Eficiencia Energética   estable  ETS_simple|ARIMA_basico       ARIMA
3  Eficiencia Energética   estable  ETS_simple|ARIMA_basico       ARIMA
4  Eficiencia Energética   estable  ETS_simple|ARIMA_basico       ARIMA

# Obtenemos las claves únicas (ID_BUILDING, FM_COST_TYPE) en train y test
train_pairs = df_train_fd1_v5_ITE1_ARIMA[["ID_BUILDING", "FM_COST_TYPE"]].drop_duplicates()
test_pairs  = df_test_fd1_v5_ITE1_ARIMA[["ID_BUILDING", "FM_COST_TYPE"]].drop_duplicates()

# Nos quedamos con la intersección
valid_pairs = pd.merge(train_pairs, test_pairs, on=["ID_BUILDING", "FM_COST_TYPE"], how="inner")

print("Series en train:", len(train_pairs))
print("Series en test:", len(test_pairs))
print("Series comunes (válidas):", len(valid_pairs))

# Filtramos el dataset de train para quedarnos solo con esas parejas
df_train_fd1_v5_ITE1_ARIMA_filtered = df_train_fd1_v5_ITE1_ARIMA.merge(
    valid_pairs, on=["ID_BUILDING", "FM_COST_TYPE"], how="inner"
)

print("Shape train original:", df_train_fd1_v5_ITE1_ARIMA.shape)
print("Shape train filtrado:", df_train_fd1_v5_ITE1_ARIMA_filtered.shape)

Series en train: 500
Series en test: 491
Series comunes (válidas): 491
Shape train original: (17188, 14)
Shape train filtrado: (16885, 14)

# Filtramos también el test para quedarnos solo con las parejas válidas
df_test_fd1_v5_ITE1_ARIMA_filtered = df_test_fd1_v5_ITE1_ARIMA.merge(
    valid_pairs, on=["ID_BUILDING", "FM_COST_TYPE"], how="inner"
)

print("Shape test original:", df_test_fd1_v5_ITE1_ARIMA.shape)
print("Shape test filtrado:", df_test_fd1_v5_ITE1_ARIMA_filtered.shape)

Shape test original: (5835, 14)
Shape test filtrado: (5835, 14)

# -------------------------------------------------------------------
# Definimos constantes y parámetros
# -------------------------------------------------------------------
H = 12  # horizonte de predicción mensual (2024 completo)
PAIR_COLS = ["ID_BUILDING", "FM_COST_TYPE"]
TARGET = "cost_float_mod"
DATECOL = "FECHA"

def fit_ets_variants(y):
    """
    Variantes ETS ligeras:
      - SES (simple)
      - Holt aditivo (tendencia)
      - Holt aditivo amortiguado (tendencia amortiguada)
    Selección por AIC aproximado.
    """
    candidates = []

    # SES
    try:
        m_ses = SimpleExpSmoothing(y, initialization_method="estimated").fit(optimized=True)
        aic_ses = -2*m_ses.model.loglikelihood + 2*m_ses.params.shape[0]
        candidates.append(("ETS_SES", m_ses, aic_ses))
    except Exception:
        pass

    # Holt (tendencia aditiva)
    try:
        m_holt = ExponentialSmoothing(y, trend="add", seasonal=None, initialization_method="estimated").fit(optimized=True)
        aic_holt = -2*m_holt.model.loglikelihood + 2*m_holt.params.shape[0]
        candidates.append(("ETS_HOLT", m_holt, aic_holt))
    except Exception:
        pass

    # Holt amortiguado
    try:
        m_holt_d = ExponentialSmoothing(
            y, trend="add", damped_trend=True, seasonal=None, initialization_method="estimated"
        ).fit(optimized=True)
        aic_holt_d = -2*m_holt_d.model.loglikelihood + 2*m_holt_d.params.shape[0]
        candidates.append(("ETS_HOLT_DAMPED", m_holt_d, aic_holt_d))
    except Exception:
        pass

    if not candidates:
        return None, None, np.inf

    best = sorted(candidates, key=lambda x: x[2])[0]
    return best  # (name, fitted_model, aic)

def fit_arima_basic_grid(y):
    """
    'ARIMA_basico': grid muy acotado para no disparar tiempos:
      (p,d,q) en {(0,1,1), (1,1,0), (1,1,1), (1,0,0)}
    Seleccionamos por AIC.
    """
    grid = [(0,1,1), (1,1,0), (1,1,1), (1,0,0)]
    candidates = []
    for order in grid:
        try:
            model = ARIMA(y, order=order)
            res = model.fit(method_kwargs={"warn_convergence": False})
            candidates.append((f"ARIMA{order}", res, res.aic))
        except Exception:
            continue

    if not candidates:
        return None, None, np.inf

    best = sorted(candidates, key=lambda x: x[2])[0]
    return best  # (name, fitted_model, aic)

def forecast_match_index(fitted, start, periods=H):
    """Genera un forecast de 'periods' pasos y devuelve Serie con índice mensual desde 'start'."""
    fc = fitted.forecast(steps=periods)
    # Aseguramos índice temporal correcto:
    idx = pd.date_range(start=start, periods=periods, freq="MS")
    fc = pd.Series(np.asarray(fc, dtype=float), index=idx, name="forecast")
    # Truncado de negativos (en costes no tiene sentido)
    fc = fc.clip(lower=0.0)
    return fc

train_df = df_train_fd1_v5_ITE1_ARIMA_filtered.copy()
test_df  = df_test_fd1_v5_ITE1_ARIMA_filtered.copy()

# -------------------------------------------------------------------
# Loop por parejas -> entrenar ETS y ARIMA sobre 2021-2023, predecir 2024 y evaluar
# -------------------------------------------------------------------
results = []
failed_pairs = []

# Índice esperado del test (12 meses 2024)
test_months = pd.date_range("2024-01-01", periods=H, freq="MS")

# Listado de parejas válidas
pairs = train_df[PAIR_COLS].drop_duplicates().sort_values(PAIR_COLS).values.tolist()

for bid, ctype in pairs:
    tr_pair = train_df[(train_df["ID_BUILDING"] == bid) & (train_df["FM_COST_TYPE"] == ctype)]
    te_pair = test_df [(test_df ["ID_BUILDING"] == bid) & (test_df ["FM_COST_TYPE"] == ctype)]

    try:
        y_train = ensure_ts(tr_pair)
        y_test  = ensure_ts(te_pair).reindex(test_months)

        # Si la serie de train es todo ceros o muy corta, saltamos (se cubrirá con pooling en otro pipeline)
        if (y_train.fillna(0).sum() == 0) or (y_train.dropna().shape[0] < 8):
            failed_pairs.append((bid, ctype, "serie_train_degenerate"))
            continue

        # Ajustes
        ets_name, ets_fit, ets_aic = fit_ets_variants(y_train)
        ar_name,  ar_fit,  ar_aic  = fit_arima_basic_grid(y_train)

        # Selección por AIC (si falta alguno, toma el otro)
        best_name, best_fit, best_aic, second_name, second_fit = None, None, np.inf, None, None
        if ets_fit is None and ar_fit is None:
            failed_pairs.append((bid, ctype, "sin_modelos_validos"))
            continue
        elif ets_fit is None:
            best_name, best_fit, best_aic = ar_name, ar_fit, ar_aic
        elif ar_fit is None:
            best_name, best_fit, best_aic = ets_name, ets_fit, ets_aic
        else:
            if ets_aic <= ar_aic:
                best_name, best_fit, best_aic = ets_name, ets_fit, ets_aic
                second_name, second_fit = ar_name, ar_fit
            else:
                best_name, best_fit, best_aic = ar_name, ar_fit, ar_aic
                second_name, second_fit = ets_name, ets_fit

        # Obtenemos las previsiones o forecasts 2024
        start_2024 = pd.Timestamp("2024-01-01")
        yhat_best = forecast_match_index(best_fit, start_2024, periods=H)

        # Calculamos métricas
        m_mae  = mae(y_test.values, yhat_best.values)
        m_mape = mape(y_test.values, yhat_best.values)

        # También medimos el segundo para diagnóstico
        sec_mae = sec_mape = np.nan
        if second_fit is not None:
            yhat_sec = forecast_match_index(second_fit, start_2024, periods=H)
            sec_mae  = mae(y_test.values, yhat_sec.values)
            sec_mape = mape(y_test.values, yhat_sec.values)

        results.append({
            "ID_BUILDING": bid,
            "FM_COST_TYPE": ctype,
            "best_model": best_name,
            "best_aic": best_aic,
            "MAE_2024": m_mae,
            "MAPE_2024": m_mape,
            "second_model": second_name,
            "sec_MAE_2024": sec_mae,
            "sec_MAPE_2024": sec_mape
        })

    except Exception as e:
        failed_pairs.append((bid, ctype, str(e)))
        continue

res_df = pd.DataFrame(results)
print("Modelos evaluados:", res_df.shape[0], "de", len(pairs))
print(res_df.groupby("best_model").size().sort_values(ascending=False))
print("\nMAE global (media):", res_df["MAE_2024"].mean().round(2))
print("MAPE global (media):", res_df["MAPE_2024"].mean().round(2), "%")

# TOP 10 mejores y peores (por MAE)
print("\nTOP 10 mejores (MAE):")
print(res_df.nsmallest(10, "MAE_2024")[["ID_BUILDING","FM_COST_TYPE","best_model","MAE_2024","MAPE_2024"]])

print("\nTOP 10 peores (MAE):")
print(res_df.nlargest(10, "MAE_2024")[["ID_BUILDING","FM_COST_TYPE","best_model","MAE_2024","MAPE_2024"]])

# Guardamos resultados
# OUT_DIR = "/content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO"
# out_csv = f"{OUT_DIR}/res_ITE1_ARIMA_vs_ETS_por_pareja.csv"
# res_df.to_csv(out_csv, index=False, sep=";")
# print("\nResultados guardados en:", out_csv)

if failed_pairs:
    print("\nParejas omitidas / con error (primeras 20):", failed_pairs[:20])

Modelos evaluados: 491 de 491
best_model
ARIMA(0, 1, 1)    378
ARIMA(1, 1, 0)     59
ARIMA(1, 1, 1)     53
ARIMA(1, 0, 0)      1
dtype: int64

MAE global (media): 684.97
MAPE global (media): 2692.53 %

TOP 10 mejores (MAE):
     ID_BUILDING           FM_COST_TYPE      best_model  MAE_2024  \
189         1170        Mtto. Contratos  ARIMA(1, 1, 0)  0.000000
472      1001210  Eficiencia Energética  ARIMA(1, 0, 0)  0.000005
357      1000608        Mtto. Contratos  ARIMA(1, 1, 1)  0.006274
277      1000310        Servicios Ctto.  ARIMA(0, 1, 1)  3.960000
160         1089  Eficiencia Energética  ARIMA(1, 1, 0)  4.875000
175         1100  Eficiencia Energética  ARIMA(1, 1, 0)  4.881818
151         1084  Eficiencia Energética  ARIMA(1, 1, 0)  5.229545
157         1087  Eficiencia Energética  ARIMA(1, 1, 0)  5.229545
122         1024            Suministros  ARIMA(0, 1, 1)  9.047618
353      1000598            Suministros  ARIMA(0, 1, 1)  9.189820

        MAPE_2024
189  0.000000e+00
472  1.362106e-07
357  3.450481e-03
277  2.436923e+00
160  1.890134e+00
175  1.892778e+00
151  1.776340e+00
157  1.776340e+00
122  2.438515e+01
353  3.118761e+01

TOP 10 peores (MAE):
     ID_BUILDING      FM_COST_TYPE      best_model      MAE_2024  \
415      1001070       Suministros  ARIMA(0, 1, 1)  13380.836462
129         1065       Suministros  ARIMA(0, 1, 1)   9932.522655
116          935       Suministros  ARIMA(0, 1, 1)   8535.036500
454      1001173       Suministros  ARIMA(0, 1, 1)   7859.447983
247         1296       Suministros  ARIMA(1, 1, 0)   7156.328499
257         1572       Suministros  ARIMA(0, 1, 1)   6802.281240
0              2       Suministros  ARIMA(0, 1, 1)   5779.024000
4             18             Obras  ARIMA(0, 1, 1)   5099.201500
107          595       Suministros  ARIMA(0, 1, 1)   4355.257475
413      1001070  Mtto. Correctivo  ARIMA(0, 1, 1)   4278.926669

         MAPE_2024
415  691683.411504
129      82.705817
116      64.981087
454      73.627427
247      26.633430
257      83.781072
0      1016.270921
4      2839.057973
107      99.601470
413     214.060831

# Copiamos train/test filtrados para ETS/ARIMA
train_df = df_train_fd1_v5_ITE1_ARIMA_filtered.copy()
test_df  = df_test_fd1_v5_ITE1_ARIMA_filtered.copy()

# -------------------------------------------------------------------
# Loop por parejas: entrenar ETS/ARIMA (2021-2023) -> predecir 2024 -> evaluar
#     Además, guardamos el pronóstico mensual del mejor modelo.
# -------------------------------------------------------------------
results = []
failed_pairs = []
forecasts_rows = []  # para guardar pronóstico mensual por pareja

test_months = pd.date_range("2024-01-01", periods=H, freq="MS")
pairs = train_df[PAIR_COLS].drop_duplicates().sort_values(PAIR_COLS).values.tolist()

for bid, ctype in pairs:
    tr_pair = train_df[(train_df["ID_BUILDING"] == bid) & (train_df["FM_COST_TYPE"] == ctype)]
    te_pair = test_df [(test_df ["ID_BUILDING"] == bid) & (test_df ["FM_COST_TYPE"] == ctype)]

    try:
        y_train = ensure_ts(tr_pair)
        y_test  = ensure_ts(te_pair).reindex(test_months)

        # Series degeneradas
        if (y_train.fillna(0).sum() == 0) or (y_train.dropna().shape[0] < 8):
            failed_pairs.append((bid, ctype, "serie_train_degenerate"))
            continue

        # Ajustes ETS (3 variantes) y ARIMA (rejilla)
        ets_name, ets_fit, ets_aic = fit_ets_variants(y_train)
        ar_name,  ar_fit,  ar_aic  = fit_arima_basic_grid(y_train)

        # Selección por AIC
        best_name, best_fit, best_aic, second_name, second_fit = None, None, np.inf, None, None
        if ets_fit is None and ar_fit is None:
            failed_pairs.append((bid, ctype, "sin_modelos_validos"))
            continue
        elif ets_fit is None:
            best_name, best_fit, best_aic = ar_name, ar_fit, ar_aic
        elif ar_fit is None:
            best_name, best_fit, best_aic = ets_name, ets_fit, ets_aic
        else:
            if ets_aic <= ar_aic:
                best_name, best_fit, best_aic = ets_name, ets_fit, ets_aic
                second_name, second_fit = ar_name, ar_fit
            else:
                best_name, best_fit, best_aic = ar_name, ar_fit, ar_aic
                second_name, second_fit = ets_name, ets_fit

        # Pronóstico 2024 (mejor modelo)
        start_2024 = pd.Timestamp("2024-01-01")
        yhat_best = forecast_match_index(best_fit, start_2024, periods=H)

        # Métricas del mejor modelo
        m_mae    = mae  (y_test.values, yhat_best.values)
        m_mape   = mape (y_test.values, yhat_best.values)
        m_smape  = smape(y_test.values, yhat_best.values)
        m_wape   = wape (y_test.values, yhat_best.values)
        m_mase1  = mase_no_fallback(y_train.values, y_test.values, yhat_best.values, m=1)   # naive simple
        m_mase12 = mase12          (y_train.values, y_test.values, yhat_best.values)        # naive estacional (m=12)

        # Métricas del segundo mejor (si existe)
        sec_mae = sec_mape = sec_smape = sec_wape = sec_mase1 = sec_mase12 = np.nan
        if second_fit is not None:
            yhat_sec = forecast_match_index(second_fit, start_2024, periods=H)
            sec_mae    = mae  (y_test.values, yhat_sec.values)
            sec_mape   = mape (y_test.values, yhat_sec.values)
            sec_smape  = smape(y_test.values, yhat_sec.values)
            sec_wape   = wape (y_test.values, yhat_sec.values)
            sec_mase1  = mase_no_fallback(y_train.values, y_test.values, yhat_sec.values, m=1)
            sec_mase12 = mase12          (y_train.values, y_test.values, yhat_sec.values)

        results.append({
            "ID_BUILDING": bid,
            "FM_COST_TYPE": ctype,
            "best_model": best_name,
            "best_aic": best_aic,
            "MAE_2024": m_mae,
            "MAPE_2024": m_mape,
            "SMAPE_2024": m_smape,
            "WAPE_2024": m_wape,
            "MASE1_2024": m_mase1,     # nuevo
            "MASE12_2024": m_mase12,   # nuevo
            "second_model": second_name,
            "sec_MAE_2024": sec_mae,
            "sec_MAPE_2024": sec_mape,
            "sec_SMAPE_2024": sec_smape,
            "sec_WAPE_2024": sec_wape,
            "sec_MASE1_2024": sec_mase1,     # nuevo
            "sec_MASE12_2024": sec_mase12    # nuevo
        })

        # Guardamos pronóstico mensual del mejor modelo (junto con el real)
        df_fc = pd.DataFrame({
            "FECHA": test_months,
            "y_true": y_test.values,
            "y_pred_best": yhat_best.values
        })
        df_fc["ID_BUILDING"]  = bid
        df_fc["FM_COST_TYPE"] = ctype
        df_fc["best_model"]   = best_name
        forecasts_rows.append(df_fc)

    except Exception as e:
        failed_pairs.append((bid, ctype, str(e)))
        continue

# Resumen de resultados
res_df = pd.DataFrame(results)
print("Modelos evaluados:", res_df.shape[0], "de", len(pairs))
print(res_df.groupby("best_model").size().sort_values(ascending=False))

# Estadísticos globales (ignorando NaN donde aplique)
print("\nMAE global (media):",   np.nanmean(res_df["MAE_2024"]).round(2))
print("SMAPE global (media):",  np.nanmean(res_df["SMAPE_2024"]).round(2), "%")
print("WAPE global (media):",  (100*np.nanmean(res_df["WAPE_2024"])).round(2), "%")  # mostramos en %
print("MASE1 global (media):",  np.nanmean(res_df["MASE1_2024"]).round(3), "(naive simple)")
print("MASE12 global (media):", np.nanmean(res_df["MASE12_2024"]).round(3), "(naive estacional)")

# TOP 10 mejores y peores (por WAPE, más robusto con ceros)
print("\nTOP 10 mejores (WAPE):")
print(res_df.nsmallest(10, "WAPE_2024")[["ID_BUILDING","FM_COST_TYPE","best_model","WAPE_2024","SMAPE_2024","MAE_2024","MASE1_2024","MASE12_2024"]])

print("\nTOP 10 peores (WAPE):")
print(res_df.nlargest(10, "WAPE_2024")[["ID_BUILDING","FM_COST_TYPE","best_model","WAPE_2024","SMAPE_2024","MAE_2024","MASE1_2024","MASE12_2024"]])

# Guardado opcional (descomentamos si queremos persistir)
# OUT_DIR = "/content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO"
# out_csv_metrics   = f"{OUT_DIR}/res_ITE1_ARIMA_vs_ETS_Metrics.csv"
# out_csv_forecasts = f"{OUT_DIR}/res_ITE1_ARIMA_vs_ETS_Forecasts.csv"
# res_df.to_csv(out_csv_metrics, index=False, sep=";")
# if forecasts_rows:
#     fc_df = pd.concat(forecasts_rows, ignore_index=True)
#     fc_df = fc_df[["ID_BUILDING","FM_COST_TYPE","FECHA","best_model","y_true","y_pred_best"]].sort_values(["ID_BUILDING","FM_COST_TYPE","FECHA"])
#     fc_df.to_csv(out_csv_forecasts, index=False, sep=";")
#     print("\nMétricas guardadas en:", out_csv_metrics)
#     print("Pronósticos mensuales guardados en:", out_csv_forecasts)
# else:
#     print("\nNo se generaron pronósticos (lista vacía).")

if failed_pairs:
    print("\nParejas omitidas / con error (primeras 20):", failed_pairs[:20])

Modelos evaluados: 491 de 491
best_model
ARIMA(0, 1, 1)    378
ARIMA(1, 1, 0)     59
ARIMA(1, 1, 1)     53
ARIMA(1, 0, 0)      1
dtype: int64

MAE global (media): 684.97
SMAPE global (media): 48.38 %
WAPE global (media): 51.35 %
MASE1 global (media): 1.266 (naive simple)
MASE12 global (media): 0.589 (naive estacional)

TOP 10 mejores (WAPE):
     ID_BUILDING           FM_COST_TYPE      best_model     WAPE_2024  \
189         1170        Mtto. Contratos  ARIMA(1, 1, 0)  0.000000e+00
472      1001210  Eficiencia Energética  ARIMA(1, 0, 0)  1.362106e-09
357      1000608        Mtto. Contratos  ARIMA(1, 1, 1)  3.450481e-05
159         1087        Servicios Ctto.  ARIMA(0, 1, 1)  4.398239e-03
132         1075        Servicios Ctto.  ARIMA(0, 1, 1)  8.515159e-03
167         1092        Servicios Ctto.  ARIMA(1, 1, 0)  1.612388e-02
149         1083        Servicios Ctto.  ARIMA(0, 1, 1)  1.627627e-02
138         1078        Servicios Ctto.  ARIMA(1, 1, 0)  1.644988e-02
151         1084  Eficiencia Energética  ARIMA(1, 1, 0)  1.786921e-02
157         1087  Eficiencia Energética  ARIMA(1, 1, 0)  1.786921e-02

       SMAPE_2024    MAE_2024  MASE1_2024  MASE12_2024
189  0.000000e+00    0.000000         NaN          NaN
472  1.362106e-07    0.000005         NaN          NaN
357  3.450542e-03    0.006274    0.000285     0.000380
159  4.402388e-01   28.971608    0.146532     0.064394
132  8.548909e-01   38.041861    0.237467     0.146632
167  1.609209e+00  105.612472    0.540367     0.200763
149  1.624394e+00   83.972311    0.717664     0.247924
138  1.641636e+00   84.977638    0.594734     0.137377
151  1.797628e+00    5.229545    0.158276     0.026691
157  1.797628e+00    5.229545    0.158276     0.026691

TOP 10 peores (WAPE):
     ID_BUILDING      FM_COST_TYPE      best_model  WAPE_2024  SMAPE_2024  \
0              2       Suministros  ARIMA(0, 1, 1)   2.878377  132.319195
211         1230   Servicios Extra  ARIMA(0, 1, 1)   1.912367  119.527376
196         1197  Mtto. Correctivo  ARIMA(0, 1, 1)   1.910372  120.880451
205         1216  Mtto. Correctivo  ARIMA(1, 1, 1)   1.727976  109.445883
240         1283   Servicios Extra  ARIMA(0, 1, 1)   1.724054  115.278783
321      1000480  Mtto. Correctivo  ARIMA(0, 1, 1)   1.683877  110.157581
294      1000402   Servicios Extra  ARIMA(0, 1, 1)   1.617967  124.498400
249         1306  Mtto. Correctivo  ARIMA(0, 1, 1)   1.471232   93.030414
213         1238  Mtto. Correctivo  ARIMA(0, 1, 1)   1.467608   99.511627
417      1001110       Suministros  ARIMA(0, 1, 1)   1.312104   89.083955

        MAE_2024  MASE1_2024  MASE12_2024
0    5779.024000    1.787252     0.605817
211    33.478846    0.514961     0.653429
196   323.488176    0.993703     1.051523
205   228.558071    0.804907     0.628191
240    56.393586    0.604781     0.708419
321   237.180739    0.540120     0.541849
294   464.856749    0.616470     0.643182
249  1781.311863    0.549598     0.405485
213   186.211861    0.827679     0.693953
417   505.486862    1.362183     0.548530

# ================================================================
# Guardado homogéneo para el caso NO estacional (ETS / ARIMA)
#   - Predicciones por pareja (2024)
#   - Métricas por pareja
#   Sufijo: ARIMA_ETS_NS  (Non-Seasonal)
# ================================================================


base_dir = "/content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO"
os.makedirs(base_dir, exist_ok=True)

# Construimos las tablas a guardar con nombres consistentes
# Predicciones por pareja (2024)
if forecasts_rows:
    preds_ARIMA_ETS_2024 = pd.concat(forecasts_rows, ignore_index=True)
    preds_ARIMA_ETS_2024 = preds_ARIMA_ETS_2024[[
        "ID_BUILDING","FM_COST_TYPE","FECHA","best_model","y_true","y_pred_best"
    ]].copy()
else:
    preds_ARIMA_ETS_2024 = pd.DataFrame(columns=[
        "ID_BUILDING","FM_COST_TYPE","FECHA","best_model","y_true","y_pred_best"
    ])

# Métricas por pareja (añadimos campos estándar)
metrics_ARIMA_ETS_2024 = res_df.copy()
metrics_ARIMA_ETS_2024.insert(0, "model", "ARIMA_ETS_NS")  # etiqueta de modelo agregada
# Si queremos contar observaciones del test (12 meses 2024), lo dejamos fijo:
metrics_ARIMA_ETS_2024["n_obs"] = 12

# Rutas de salida (siguiendo el patrón de nombres)
pred_arima_ets_path    = os.path.join(base_dir, "df_pred_fd1_v5_ITE1_ARIMA_ETS_NS.csv")
metrics_arima_ets_path = os.path.join(base_dir, "df_metrics_fd1_v5_ITE1_ARIMA_ETS_NS.csv")

# Guardamos
if not preds_ARIMA_ETS_2024.empty:
    preds_ARIMA_ETS_2024.to_csv(pred_arima_ets_path, sep=";", index=False)
else:
    print("Aviso: no hay predicciones ARIMA_ETS_NS para guardar.")

metrics_ARIMA_ETS_2024.to_csv(metrics_arima_ets_path, index=False)

# Mensaje de confirmación
print("Guardados en Drive (NO estacional):")
print("Predicciones ARIMA_ETS_NS:", pred_arima_ets_path if not preds_ARIMA_ETS_2024.empty else "no disponibles")
print("Métricas ARIMA_ETS_NS   :", metrics_arima_ets_path)

Guardados en Drive (NO estacional):
Predicciones ARIMA_ETS_NS: /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/df_pred_fd1_v5_ITE1_ARIMA_ETS_NS.csv
Métricas ARIMA_ETS_NS   : /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/df_metrics_fd1_v5_ITE1_ARIMA_ETS_NS.csv

# 1) Agregamos el portfolio por mes (y_true, y_pred_best) – asumimos que son datos 2024.
# 2) Construimos el TRAIN agregado SOLO hasta 2023-12 (sin 2024) para el denominador de MASE.
# 3) Calculamos MASE1/MASE12 directamente con los denominadores (evitamos dependencias de estado).
# 4) Graficamos con subtítulo de métricas.


# --- Validaciones y constantes ---
assert "preds_ARIMA_ETS_2024" in globals(), "No encontramos preds_ARIMA_ETS_2024."
assert "train_df" in globals(), "No encontramos train_df."
DATECOL = "FECHA"
TARGET  = "cost_float_mod"

# Nos aseguramos de que FECHA sea datetime
preds_ARIMA_ETS_2024[DATECOL] = pd.to_datetime(preds_ARIMA_ETS_2024[DATECOL], errors="coerce")
train_df[DATECOL]             = pd.to_datetime(train_df[DATECOL], errors="coerce")

# --- 1) Agregado de portfolio por mes (predicciones 2024) ---
agg = (
    preds_ARIMA_ETS_2024
    .groupby(DATECOL, as_index=False)[["y_true","y_pred_best"]]
    .sum()
    .sort_values(DATECOL)
    .reset_index(drop=True)
)
# Reindex a mensual por robustez
idx = pd.date_range(agg[DATECOL].min(), agg[DATECOL].max(), freq="MS")
agg = (agg.set_index(DATECOL).reindex(idx).fillna(0.0).rename_axis(DATECOL).reset_index())

# --- 2) TRAIN agregado del mismo portfolio (solo < 2024-01-01) ---
pairs_port = preds_ARIMA_ETS_2024[["ID_BUILDING","FM_COST_TYPE"]].drop_duplicates()

# (Opcional) alineamos tipos para el merge si hubiera dudas:
# preds_ARIMA_ETS_2024["ID_BUILDING"] = preds_ARIMA_ETS_2024["ID_BUILDING"].astype(str)
# train_df["ID_BUILDING"]             = train_df["ID_BUILDING"].astype(str)

train_port = (
    train_df
    .merge(pairs_port, on=["ID_BUILDING","FM_COST_TYPE"], how="inner")
    [[DATECOL, TARGET]]
    .copy()
)
train_port = train_port[train_port[DATECOL] < pd.Timestamp("2024-01-01")]
s_train = (
    train_port
    .groupby(DATECOL)[TARGET]
    .sum()
    .sort_index()
    .asfreq("MS", fill_value=0.0)
    .astype(float)
)

# --- 3) Métricas de portfolio y MASE directo (sin funciones externas) ---
y_true_port = agg["y_true"].to_numpy(dtype=float)
y_pred_port = agg["y_pred_best"].to_numpy(dtype=float)

MAE_port   = mae  (y_true_port, y_pred_port)
SMAPE_port = smape(y_true_port, y_pred_port)             # %
WAPE_port  = wape (y_true_port, y_pred_port)             # 0–1

# Numerador de MASE (MAE sobre 2024) y denominadores del naive
num   = float(np.mean(np.abs(y_true_port - y_pred_port)))                   # = MAE_port
z     = s_train.to_numpy(dtype=float)
den1  = float(np.mean(np.abs(z[1:]  - z[:-1] ))) if len(z) > 1  else np.nan
den12 = float(np.mean(np.abs(z[12:] - z[:-12]))) if len(z) > 12 else np.nan

MASE1_port  = num / den1  if np.isfinite(den1)  and den1  != 0 else np.nan
MASE12_port = num / den12 if np.isfinite(den12) and den12 != 0 else np.nan

print("Métricas PORTFOLIO (agregado mensual):")
print(f" - MAE     : {MAE_port:,.3f}")
print(f" - SMAPE % : {SMAPE_port:,.3f}")
print(f" - WAPE    : {WAPE_port:,.4f}  ({100*WAPE_port:,.2f}%)")
print(f" - MASE1   : {MASE1_port}")
print(f" - MASE12  : {MASE12_port}")

# --- 4) Gráfico con subtítulo de métricas ---
plt.figure(figsize=(10,5))
plt.plot(agg[DATECOL], agg["y_true"], label="Real (portfolio)", marker="o")
plt.plot(agg[DATECOL], agg["y_pred_best"], label="Predicho (portfolio)", marker="x")

subtitle = (
    f"MAE={num:,.0f} | WAPE={100*WAPE_port:,.1f}% | "
    f"SMAPE={SMAPE_port:,.1f}% | MASE1={MASE1_port:.3f} | MASE12={MASE12_port:.3f}"
)
plt.title("Predicciones agregadas vs Realidad - Portfolio\n" + subtitle)
plt.xlabel("Mes")
plt.ylabel("Coste total")
plt.legend()
plt.grid(True)
plt.tight_layout()
plt.show()

Métricas PORTFOLIO (agregado mensual):
 - MAE     : 131,540.511
 - SMAPE % : 15.871
 - WAPE    : 0.1639  (16.39%)
 - MASE1   : 0.8746391961672845
 - MASE12  : 0.2735416205281805

sns.scatterplot(data=preds_ARIMA_ETS_2024, x="y_true", y="y_pred_best", alpha=0.5)
plt.plot([0, preds_ARIMA_ETS_2024["y_true"].max()], [0, preds_ARIMA_ETS_2024["y_true"].max()], 'r--')
plt.title("Predicciones vs Realidad (todas las parejas)")
plt.xlabel("Real")
plt.ylabel("Predicho")
plt.show()

print("pairs en preds:", preds_ARIMA_ETS_2024[["ID_BUILDING","FM_COST_TYPE"]].drop_duplicates().shape[0])

pairs_port = preds_ARIMA_ETS_2024[["ID_BUILDING","FM_COST_TYPE"]].drop_duplicates()
train_port = (train_df.merge(pairs_port, on=["ID_BUILDING","FM_COST_TYPE"], how="inner")
                      [["FECHA","cost_float_mod"]])
train_port["FECHA"] = pd.to_datetime(train_port["FECHA"])
s_train = (train_port.groupby("FECHA")["cost_float_mod"].sum()
                     .sort_index()
                     .asfreq("MS", fill_value=0.0))

print("len(s_train):", len(s_train), "min:", s_train.index.min(), "max:", s_train.index.max(),
      "sum:", float(s_train.sum()))

import numpy as np
d1  = np.mean(np.abs(s_train[1:].values  - s_train[:-1].values))           if len(s_train) > 1  else np.nan
d12 = np.mean(np.abs(s_train[12:].values - s_train[:-12].values))          if len(s_train) > 12 else np.nan
print("d1:", d1, "d12:", d12)

pairs en preds: 491
len(s_train): 36 min: 2021-01-01 00:00:00 max: 2023-12-01 00:00:00 sum: 33193056.901383586
d1: 150394.02725129225 d12: 480879.3296224533

np.isfinite(y_true_port).all(), np.isfinite(y_pred_port).all()

(np.True_, np.True_)

# Definimos la carpeta de salida en Drive
OUT_DIR = "/content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO"

# --- OPCIÓN 1: Cargar desde CSV (con separador ';') ---
# csv_path = f"{OUT_DIR}/df_train_fd1_v5_ITE1_CROSTON.csv"
# df_train_fd1_v5_ITE1_CROSTON = pd.read_csv(csv_path, sep=';')

# --- OPCIÓN 2: Cargar desde Excel ---
xlsx_path = f"{OUT_DIR}/df_train_fd1_v5_ITE1_CROSTON.xlsx"
df_train_fd1_v5_ITE1_CROSTON = pd.read_excel(xlsx_path)

# Visualizamos forma y primeras filas para validar
print("Shape:", df_train_fd1_v5_ITE1_CROSTON.shape)
print(df_train_fd1_v5_ITE1_CROSTON.head())

# Normalizamos la fecha si existe
if "FECHA" in df_train_fd1_v5_ITE1_CROSTON.columns:
    df_train_fd1_v5_ITE1_CROSTON["FECHA"] = pd.to_datetime(df_train_fd1_v5_ITE1_CROSTON["FECHA"], errors="coerce")

print("Cargado:", df_train_fd1_v5_ITE1_CROSTON.shape)
df_train_fd1_v5_ITE1_CROSTON.head(3)

Shape: (14460, 14)
   ID_BUILDING FM_COST_TYPE      FECHA  YEAR  MONTH  cost_float_mod  \
0            2    Licencias 2021-01-01  2021      1         1145.46
1            2    Licencias 2021-02-01  2021      2           55.95
2            2    Licencias 2021-03-01  2021      3            0.00
3            2    Licencias 2021-04-01  2021      4            0.00
4            2    Licencias 2021-05-01  2021      5          305.95

  COUNTRY_DEF ID_REGION_GRUPO  TIPO_USO SUPPLIER_TYPE_MOD_2  \
0      España               2  Oficinas             INTERNO
1      España               2  Oficinas             INTERNO
2      España               2  Oficinas             INTERNO
3      España               2  Oficinas             INTERNO
4      España               2  Oficinas             INTERNO

  FM_RESPONSIBLE_MOD     diagnosis model_suggested model_group
0          Licencias  intermitente     Croston|SBA     CROSTON
1          Licencias  intermitente     Croston|SBA     CROSTON
2          Licencias  intermitente     Croston|SBA     CROSTON
3          Licencias  intermitente     Croston|SBA     CROSTON
4          Licencias  intermitente     Croston|SBA     CROSTON
Cargado: (14460, 14)

# Definimos la carpeta de salida en Drive
OUT_DIR = "/content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO"

# --- OPCIÓN 1: Cargar desde CSV (con separador ';') ---
# csv_path = f"{OUT_DIR}/df_test_fd1_v5_ITE1_CROSTON.csv"
# df_test_fd1_v5_ITE1_CROSTON = pd.read_csv(csv_path, sep=';')

# --- OPCIÓN 2: Cargar desde Excel ---
xlsx_path = f"{OUT_DIR}/df_test_fd1_v5_ITE1_CROSTON.xlsx"
df_test_fd1_v5_ITE1_CROSTON = pd.read_excel(xlsx_path)

# Visualizamos forma y primeras filas para validar
print("Shape:", df_test_fd1_v5_ITE1_CROSTON.shape)
print(df_test_fd1_v5_ITE1_CROSTON.head())

# Normalizamos fecha
if "FECHA" in df_test_fd1_v5_ITE1_CROSTON.columns:
    df_test_fd1_v5_ITE1_CROSTON["FECHA"] = pd.to_datetime(df_test_fd1_v5_ITE1_CROSTON["FECHA"], errors="coerce")

print("Cargado:", df_test_fd1_v5_ITE1_CROSTON.shape)
df_test_fd1_v5_ITE1_CROSTON.head(3)

Shape: (4680, 14)
   ID_BUILDING     FM_COST_TYPE      FECHA  YEAR  MONTH  cost_float_mod  \
0            2        Licencias 2024-12-01  2024     12           202.8
1            2            Obras 2024-02-01  2024      2            54.6
2            2  Servicios Extra 2024-01-01  2024      1             0.0
3            2  Servicios Extra 2024-02-01  2024      2             0.0
4            9        Licencias 2024-01-01  2024      1          1600.0

  COUNTRY_DEF ID_REGION_GRUPO  TIPO_USO SUPPLIER_TYPE_MOD_2  \
0      España               2  Oficinas             INTERNO
1      España               2  Oficinas             INTERNO
2      España               2  Oficinas             INTERNO
3      España               2  Oficinas             INTERNO
4      España               2  Oficinas             INTERNO

  FM_RESPONSIBLE_MOD     diagnosis model_suggested model_group
0          Licencias  intermitente     Croston|SBA     CROSTON
1   Gestión Espacios  intermitente     Croston|SBA     CROSTON
2      Mantenimiento  intermitente     Croston|SBA     CROSTON
3      Mantenimiento  intermitente     Croston|SBA     CROSTON
4          Licencias  intermitente     Croston|SBA     CROSTON
Cargado: (4680, 14)

# Creamos set de parejas válidas a partir del test
parejas_test = set(zip(
    df_test_fd1_v5_ITE1_CROSTON["ID_BUILDING"],
    df_test_fd1_v5_ITE1_CROSTON["FM_COST_TYPE"]
))

# Filtramos train
df_train_fd1_v5_ITE1_CROSTON_filtrado = df_train_fd1_v5_ITE1_CROSTON[
    df_train_fd1_v5_ITE1_CROSTON.apply(
        lambda r: (r["ID_BUILDING"], r["FM_COST_TYPE"]) in parejas_test, axis=1
    )
].copy()

print("Shape original train:", df_train_fd1_v5_ITE1_CROSTON.shape)
print("Shape filtrado train:", df_train_fd1_v5_ITE1_CROSTON_filtrado.shape)

# Comprobamos que todas las parejas en train filtrado estén en test
faltantes = set(zip(
    df_train_fd1_v5_ITE1_CROSTON_filtrado["ID_BUILDING"],
    df_train_fd1_v5_ITE1_CROSTON_filtrado["FM_COST_TYPE"]
)) - parejas_test
print("Parejas en train filtrado que no están en test:", faltantes)

Shape original train: (14460, 14)
Shape filtrado train: (14272, 14)
Parejas en train filtrado que no están en test: set()

def ensure_1d_array(x):
    arr = np.asarray(x, dtype=float).copy()
    return np.where(np.isnan(arr), 0.0, arr)

def _first_positive(y):
    idx = np.where(y > 0)[0]
    return None if len(idx) == 0 else int(idx[0])

def croston_classic(y, alpha=0.1, h=12):
    y = ensure_1d_array(y)
    n = len(y)
    if n == 0 or np.all(y == 0):
        return np.zeros(h)

    first = _first_positive(y)
    if first is None:
        return np.zeros(h)

    z = y[y > 0]
    intervals = []
    last = -1
    for t, val in enumerate(y):
        if val > 0:
            if last >= 0:
                intervals.append(t - last)
            last = t
    if len(intervals) == 0:
        demand_hat = z.mean()
        interval_hat = n
    else:
        demand_hat = z[0]
        interval_hat = intervals[0]
        k = 0
        for t, val in enumerate(y):
            if val > 0:
                demand_hat = demand_hat + alpha*(val - demand_hat)
                if k < len(intervals):
                    interval_hat = interval_hat + alpha*(intervals[k] - interval_hat)
                    k += 1
    forecast = demand_hat / max(interval_hat, 1e-8)
    return np.full(h, forecast)

def croston_sba(y, alpha=0.1, h=12):
    base = croston_classic(y, alpha=alpha, h=1)[0]
    sba_forecast = (1 - alpha/2.0) * base
    return np.full(h, sba_forecast)

# Aseguramos fechas
df_train = df_train_fd1_v5_ITE1_CROSTON.copy()
df_train["FECHA"] = pd.to_datetime(df_train["FECHA"])
df_test = df_test_fd1_v5_ITE1_CROSTON.copy()
df_test["FECHA"] = pd.to_datetime(df_test["FECHA"])

out_all = []

for (bid, ctype), g_train in df_train.groupby(["ID_BUILDING","FM_COST_TYPE"]):
    g_train = g_train.sort_values("FECHA")
    g_test = df_test[(df_test["ID_BUILDING"]==bid) & (df_test["FM_COST_TYPE"]==ctype)].sort_values("FECHA")

    if g_test.empty:
        continue

    h = len(g_test)  # horizonte = nº meses de test
    y_train = g_train["cost_float_mod"].fillna(0).values

    preds = croston_sba(y_train, alpha=0.1, h=h)  # usamos SBA directo

    df_pred = pd.DataFrame({
        "ID_BUILDING": bid,
        "FM_COST_TYPE": ctype,
        "FECHA": g_test["FECHA"].values,
        "yhat": preds
    })
    df_pred = df_pred.merge(
        g_test[["FECHA","cost_float_mod"]],
        on="FECHA", how="left"
    ).rename(columns={"cost_float_mod":"y"})
    out_all.append(df_pred)

df_pred_all = pd.concat(out_all, ignore_index=True)
print("Predicciones generadas:", df_pred_all.shape)
df_pred_all.head()

Predicciones generadas: (4680, 5)

metrics = (df_pred_all.dropna(subset=["y"])
           .groupby(["ID_BUILDING","FM_COST_TYPE"], as_index=False)
           .apply(lambda g: pd.Series({
               "WAPE": wape(g["y"].values, g["yhat"].values),
               "SMAPE": smape(g["y"].values, g["yhat"].values)
           }))
           .reset_index(drop=True))

print("Métricas calculadas:", metrics.shape)
metrics.head()

Métricas calculadas: (472, 4)

base_dir = "/content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO"

# Guardamos predicciones con reales
pred_path = os.path.join(base_dir, "df_pred_fd1_v5_ITE1_croston.csv")
df_pred_all.to_csv(pred_path, index=False)

# Guardamos métricas
metrics_path = os.path.join(base_dir, "df_metrics_fd1_v5_ITE1_croston.csv")
metrics.to_csv(metrics_path, index=False)

print("Guardados en Drive:")
print("Predicciones:", pred_path)
print("Métricas:", metrics_path)

Guardados en Drive:
Predicciones: /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/df_pred_fd1_v5_ITE1_croston.csv
Métricas: /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/df_metrics_fd1_v5_ITE1_croston.csv

print("Resumen global CROSTON/SBA")
print("WAPE medio:", metrics["WAPE"].mean())
print("SMAPE medio:", metrics["SMAPE"].mean())
print("WAPE mediana:", metrics["WAPE"].median())
print("SMAPE mediana:", metrics["SMAPE"].median())

Resumen global CROSTON/SBA
WAPE medio: 1.163753551902763
SMAPE medio: 132.87407307842736
WAPE mediana: 0.9888819586917396
SMAPE mediana: 136.28199442702885

metrics["WAPE"].hist(bins=30)

<Axes: >

metrics["SMAPE"].hist(bins=30)

<Axes: >

# Mejores 10 (por WAPE)
top10 = metrics.sort_values("WAPE").head(10)
print("TOP 10 mejores series (WAPE bajo):")
print(top10)

# Peores 10 (por WAPE)
worst10 = metrics.sort_values("WAPE", ascending=False).head(10)
print("\nTOP 10 peores series (WAPE alto):")
print(worst10)

TOP 10 mejores series (WAPE bajo):
     ID_BUILDING      FM_COST_TYPE      WAPE      SMAPE
6             57   Mtto. Contratos  0.001871   0.186971
112          681   Servicios Extra  0.051078   4.980576
286      1000667   Mtto. Contratos  0.159152  14.742060
178         1573  Mtto. Correctivo  0.183697  20.227607
357      1001123   Servicios Extra  0.232948  26.365772
241      1000461   Mtto. Contratos  0.264129  18.269656
295      1000764   Mtto. Contratos  0.293021  34.332114
298      1000765   Mtto. Contratos  0.298461  35.081329
291      1000758   Mtto. Contratos  0.308637  36.495657
228      1000434   Servicios Extra  0.383595  32.186263

TOP 10 peores series (WAPE alto):
     ID_BUILDING      FM_COST_TYPE       WAPE       SMAPE
7             59   Servicios Ctto.  21.826612  184.806609
390      1001151  Mtto. Correctivo   8.786301  173.397598
212      1000409         Licencias   8.508665  183.460664
356      1001123  Mtto. Correctivo   5.391055  162.180979
120         1007         Licencias   4.173347  147.943091
449      1001344  Mtto. Correctivo   3.996184  169.024318
471      1001400   Servicios Extra   3.967161  146.298757
122         1015  Mtto. Correctivo   3.654344  146.367398
9            105         Licencias   3.418017  193.509385
172         1317  Mtto. Correctivo   3.082614  140.917842

# Definimos la carpeta de salida en Drive
OUT_DIR = "/content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO"

# --- OPCIÓN 1: Cargar desde CSV (con separador ';') ---
# csv_path = f"{OUT_DIR}/df_train_fd1_v5_ITE1_SARIMA.csv"
# df_train_fd1_v5_ITE1_SARIMA = pd.read_csv(csv_path, sep=';')

# --- OPCIÓN 2: Cargar desde Excel ---
xlsx_path = f"{OUT_DIR}/df_train_fd1_v5_ITE1_SARIMA.xlsx"
df_train_fd1_v5_ITE1_SARIMA = pd.read_excel(xlsx_path)

# Visualizamos forma y primeras filas para validar
print("Shape:", df_train_fd1_v5_ITE1_SARIMA.shape)
print(df_train_fd1_v5_ITE1_SARIMA.head())

# Normalizamos la fecha si existe
if "FECHA" in df_train_fd1_v5_ITE1_SARIMA.columns:
    df_train_fd1_v5_ITE1_SARIMA["FECHA"] = pd.to_datetime(df_train_fd1_v5_ITE1_SARIMA["FECHA"], errors="coerce")

print("Cargado:", df_train_fd1_v5_ITE1_SARIMA.shape)
df_train_fd1_v5_ITE1_SARIMA.head(3)

Shape: (17207, 14)
   ID_BUILDING     FM_COST_TYPE      FECHA  YEAR  MONTH  cost_float_mod  \
0            2  Mtto. Contratos 2021-01-01  2021      1          412.50
1            2  Mtto. Contratos 2021-02-01  2021      2            0.00
2            2  Mtto. Contratos 2021-03-01  2021      3            0.00
3            2  Mtto. Contratos 2021-04-01  2021      4          950.92
4            2  Mtto. Contratos 2021-05-01  2021      5         1206.55

  COUNTRY_DEF ID_REGION_GRUPO  TIPO_USO SUPPLIER_TYPE_MOD_2  \
0      España               2  Oficinas             EXTERNO
1      España               2  Oficinas             EXTERNO
2      España               2  Oficinas             EXTERNO
3      España               2  Oficinas             EXTERNO
4      España               2  Oficinas             EXTERNO

  FM_RESPONSIBLE_MOD   diagnosis        model_suggested model_group
0      Mantenimiento  estacional  ETS estacional|SARIMA      SARIMA
1      Mantenimiento  estacional  ETS estacional|SARIMA      SARIMA
2      Mantenimiento  estacional  ETS estacional|SARIMA      SARIMA
3      Mantenimiento  estacional  ETS estacional|SARIMA      SARIMA
4      Mantenimiento  estacional  ETS estacional|SARIMA      SARIMA
Cargado: (17207, 14)

# Definimos la carpeta de salida en Drive
OUT_DIR = "/content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO"

# --- OPCIÓN 1: Cargar desde CSV (con separador ';') ---
# csv_path = f"{OUT_DIR}/df_test_fd1_v5_ITE1_SARIMA.csv"
# df_test_fd1_v5_ITE1_SARIMA = pd.read_csv(csv_path, sep=';')

# --- OPCIÓN 2: Cargar desde Excel ---
xlsx_path = f"{OUT_DIR}/df_test_fd1_v5_ITE1_SARIMA.xlsx"
df_test_fd1_v5_ITE1_SARIMA = pd.read_excel(xlsx_path)

# Visualizamos forma y primeras filas para validar
print("Shape:", df_test_fd1_v5_ITE1_SARIMA.shape)
print(df_test_fd1_v5_ITE1_SARIMA.head())

# Normalizamos la fecha si existe
if "FECHA" in df_test_fd1_v5_ITE1_SARIMA.columns:
    df_test_fd1_v5_ITE1_SARIMA["FECHA"] = pd.to_datetime(df_test_fd1_v5_ITE1_SARIMA["FECHA"], errors="coerce")

print("Cargado:", df_test_fd1_v5_ITE1_SARIMA.shape)
df_test_fd1_v5_ITE1_SARIMA.head(3)

Shape: (5648, 14)
   ID_BUILDING     FM_COST_TYPE      FECHA  YEAR  MONTH  cost_float_mod  \
0            2  Mtto. Contratos 2024-01-01  2024      1         1874.00
1            2  Mtto. Contratos 2024-02-01  2024      2            0.00
2            2  Mtto. Contratos 2024-03-01  2024      3            0.00
3            2  Mtto. Contratos 2024-04-01  2024      4          175.58
4            2  Mtto. Contratos 2024-05-01  2024      5          175.58

  COUNTRY_DEF ID_REGION_GRUPO  TIPO_USO SUPPLIER_TYPE_MOD_2  \
0      España               2  Oficinas             EXTERNO
1      España               2  Oficinas             EXTERNO
2      España               2  Oficinas             EXTERNO
3      España               2  Oficinas             EXTERNO
4      España               2  Oficinas             EXTERNO

  FM_RESPONSIBLE_MOD   diagnosis        model_suggested model_group
0      Mantenimiento  estacional  ETS estacional|SARIMA      SARIMA
1      Mantenimiento  estacional  ETS estacional|SARIMA      SARIMA
2      Mantenimiento  estacional  ETS estacional|SARIMA      SARIMA
3      Mantenimiento  estacional  ETS estacional|SARIMA      SARIMA
4      Mantenimiento  estacional  ETS estacional|SARIMA      SARIMA
Cargado: (5648, 14)

# En este bloque preparamos el sub-dataframe _SARIMA ya cargado:
# 1) Aseguramos tipos y columnas clave
# 2) Filtramos TEST a 2024
# 3) Obtenemos la intersección de parejas presentes en TRAIN y TEST-2024
# 4) Dejamos listas las constantes para modelización (ETS/SARIMA) en pasos siguientes


# Comprobamos que los dataframes esperados existen
assert "df_train_fd1_v5_ITE1_SARIMA" in globals(), "No está cargado df_train_fd1_v5_ITE1_SARIMA"
assert "df_test_fd1_v5_ITE1_SARIMA" in globals(), "No está cargado df_test_fd1_v5_ITE1_SARIMA"

# Normalizamos FECHA por seguridad
for _df in [df_train_fd1_v5_ITE1_SARIMA, df_test_fd1_v5_ITE1_SARIMA]:
    if "FECHA" in _df.columns:
        _df["FECHA"] = pd.to_datetime(_df["FECHA"], errors="coerce")

# Definimos claves y variable target
KEY = ["ID_BUILDING", "FM_COST_TYPE"]
TARGET = "cost_float_mod"

# Validamos presencia de columnas clave
faltan = [c for c in (KEY + ["FECHA", TARGET]) if c not in df_train_fd1_v5_ITE1_SARIMA.columns]
assert not faltan, f"Faltan columnas en TRAIN: {faltan}"
faltan = [c for c in (KEY + ["FECHA", TARGET]) if c not in df_test_fd1_v5_ITE1_SARIMA.columns]
assert not faltan, f"Faltan columnas en TEST: {faltan}"

# Filtramos TEST a 2024
df_test_2024 = df_test_fd1_v5_ITE1_SARIMA[df_test_fd1_v5_ITE1_SARIMA["FECHA"].dt.year == 2024].copy()

# Calculamos parejas presentes en ambos conjuntos
parejas_train = set(map(tuple, df_train_fd1_v5_ITE1_SARIMA[KEY].drop_duplicates().values))
parejas_test  = set(map(tuple, df_test_2024[KEY].drop_duplicates().values))
parejas = sorted(parejas_train.intersection(parejas_test))

print("Resumen conjuntos _SARIMA:")
print(" - TRAIN shape:", df_train_fd1_v5_ITE1_SARIMA.shape)
print(" - TEST  shape:", df_test_fd1_v5_ITE1_SARIMA.shape)
print(" - TEST 2024 shape:", df_test_2024.shape)
print(" - Parejas en TRAIN:", len(parejas_train))
print(" - Parejas en TEST-2024:", len(parejas_test))
print(" - Parejas comunes a modelizar:", len(parejas))
print("Ejemplo de parejas:", parejas[:5])

# Dejamos constantes de frecuencia estacional para pasos siguientes
FREQ = "MS"                  # frecuencia mensual (Month Start)
SEASONAL_PERIODS = 12        # estacionalidad anual

Resumen conjuntos _SARIMA:
 - TRAIN shape: (17207, 14)
 - TEST  shape: (5648, 14)
 - TEST 2024 shape: (5648, 14)
 - Parejas en TRAIN: 489
 - Parejas en TEST-2024: 481
 - Parejas comunes a modelizar: 481
Ejemplo de parejas: [(2, 'Mtto. Contratos'), (2, 'Mtto. Correctivo'), (2, 'Servicios Ctto.'), (9, 'Mtto. Contratos'), (9, 'Servicios Extra')]

# Definimos constantes de proyecto
TARGET   = "cost_float_mod"
DATECOL  = "FECHA"
FREQ     = "MS"               # frecuencia mensual
SEASONAL_PERIODS = 12         # estacionalidad anual

# Comprobamos dataframes en memoria
assert "df_train_fd1_v5_ITE1_SARIMA" in globals(), "No está cargado df_train_fd1_v5_ITE1_SARIMA"
assert "df_test_fd1_v5_ITE1_SARIMA"  in globals(), "No está cargado df_test_fd1_v5_ITE1_SARIMA"

# Normalizamos FECHA por seguridad
for _df in [df_train_fd1_v5_ITE1_SARIMA, df_test_fd1_v5_ITE1_SARIMA]:
    if DATECOL in _df.columns:
        _df[DATECOL] = pd.to_datetime(_df[DATECOL], errors="coerce")

# Verificamos columnas clave
for c in ["ID_BUILDING", "FM_COST_TYPE", DATECOL, TARGET]:
    assert c in df_train_fd1_v5_ITE1_SARIMA.columns, f"Falta {c} en TRAIN"
    assert c in df_test_fd1_v5_ITE1_SARIMA.columns,  f"Falta {c} en TEST"

# Verificamos que las métricas y la utilidad ensure_ts ya existen
for fnm in ["wape", "smape", "mae", "mape", "mase"]:
    assert fnm in globals(), f"No encontramos la función {fnm}() en el entorno."
for fnu in ["ensure_ts", "pair_series"]:
    assert fnu in globals(), f"No encontramos la utilidad {fnu}() en el entorno."

# Obtenemos la serie mensual continua por pareja usando ensure_ts -> guardamos a utilidades
# def pair_series(df, pair):
#    b, t = pair
#    df_pair = df[(df["ID_BUILDING"] == b) & (df["FM_COST_TYPE"] == t)][[DATECOL, TARGET]].copy()
#    return ensure_ts(df_pair)  # devuelve serie mensual indexada por fecha (MS), float, ordenada

# Preparamos índice mensual 2024 y filtramos TEST (si procede)
idx_2024 = pd.date_range("2024-01-01", "2024-12-01", freq="MS")
df_test_2024 = df_test_fd1_v5_ITE1_SARIMA[df_test_fd1_v5_ITE1_SARIMA[DATECOL].dt.year == 2024].copy()

# Calculamos parejas comunes
parejas_train = set(map(tuple, df_train_fd1_v5_ITE1_SARIMA[["ID_BUILDING","FM_COST_TYPE"]].drop_duplicates().values))
parejas_test  = set(map(tuple, df_test_2024[["ID_BUILDING","FM_COST_TYPE"]].drop_duplicates().values))
parejas = sorted(parejas_train.intersection(parejas_test))

print("Resumen inicial:")
print(" - TRAIN:", df_train_fd1_v5_ITE1_SARIMA.shape)
print(" - TEST :", df_test_fd1_v5_ITE1_SARIMA.shape)
print(" - TEST 2024:", df_test_2024.shape)
print(" - Parejas comunes:", len(parejas))

Resumen inicial:
 - TRAIN: (17207, 14)
 - TEST : (5648, 14)
 - TEST 2024: (5648, 14)
 - Parejas comunes: 481

def predecir_ets(y_train, n_steps, seasonal_periods=SEASONAL_PERIODS):
    # Requerimos al menos dos ciclos estacionales para estabilidad
    if y_train is None or len(y_train) < seasonal_periods * 2:
        return None
    try:
        model = ExponentialSmoothing(
            y_train,
            trend="add",
            seasonal="add",
            seasonal_periods=seasonal_periods,
            initialization_method="estimated"
        ).fit(optimized=True, use_brute=False)
        return np.asarray(model.forecast(n_steps), dtype=float)
    except Exception:
        return None

registros_ets = []
for (b, t) in parejas:
    y_tr = serie_mensual(df_train_fd1_v5_ITE1_SARIMA, (b, t), col_y=TARGET)
    y_te = serie_mensual(df_test_2024, (b, t), col_y=TARGET)
    y_te_2024 = y_te.reindex(idx_2024).fillna(0.0)
    yhat = predecir_ets(y_tr.values, n_steps=len(idx_2024))
    if yhat is None:
        continue
    for fecha, real, pred in zip(idx_2024, y_te_2024.values, yhat):
        registros_ets.append({
            "ID_BUILDING": b,
            "FM_COST_TYPE": t,
            "FECHA": fecha,
            "y_real": float(real),
            "y_hat": float(pred),
            "model": "ETS"
        })

preds_ETS_2024 = pd.DataFrame(registros_ets)
print("Predicciones ETS 2024 generadas:", preds_ETS_2024.shape)
print(preds_ETS_2024.head(10))

/usr/local/lib/python3.12/dist-packages/statsmodels/tsa/holtwinters/model.py:903: ConvergenceWarning: Optimization failed to converge. Check mle_retvals.
  warnings.warn(
/usr/local/lib/python3.12/dist-packages/statsmodels/tsa/holtwinters/model.py:903: ConvergenceWarning: Optimization failed to converge. Check mle_retvals.
  warnings.warn(
/usr/local/lib/python3.12/dist-packages/statsmodels/tsa/holtwinters/model.py:903: ConvergenceWarning: Optimization failed to converge. Check mle_retvals.
  warnings.warn(
/usr/local/lib/python3.12/dist-packages/statsmodels/tsa/holtwinters/model.py:903: ConvergenceWarning: Optimization failed to converge. Check mle_retvals.
  warnings.warn(

Predicciones ETS 2024 generadas: (5700, 6)
   ID_BUILDING     FM_COST_TYPE      FECHA   y_real        y_hat model
0            2  Mtto. Contratos 2024-01-01  1874.00  3408.995613   ETS
1            2  Mtto. Contratos 2024-02-01     0.00  1786.092913   ETS
2            2  Mtto. Contratos 2024-03-01     0.00  1786.086291   ETS
3            2  Mtto. Contratos 2024-04-01   175.58  2598.503450   ETS
4            2  Mtto. Contratos 2024-05-01   175.58  3035.778366   ETS
5            2  Mtto. Contratos 2024-06-01   175.58  1786.088753   ETS
6            2  Mtto. Contratos 2024-07-01   175.58  2543.745587   ETS
7            2  Mtto. Contratos 2024-08-01   175.58  1860.816923   ETS
8            2  Mtto. Contratos 2024-09-01   175.58  1946.786141   ETS
9            2  Mtto. Contratos 2024-10-01   175.58  2410.544310   ETS

SARIMA_PDQs  = [(0,1,1), (1,1,1)]
SARIMA_sPDQs = [(0,1,1,SEASONAL_PERIODS), (1,1,1,SEASONAL_PERIODS)]

def predecir_sarima(y_train, n_steps, pdqs=SARIMA_PDQs, spdqs=SARIMA_sPDQs):
    # Exigimos al menos dos ciclos estacionales para estabilidad
    if y_train is None or len(y_train) < SEASONAL_PERIODS * 2:
        return None
    best_aic = np.inf
    best_res = None
    for (p,d,q) in pdqs:
        for (P,D,Q,s) in spdqs:
            try:
                model = SARIMAX(
                    y_train,
                    order=(p,d,q),
                    seasonal_order=(P,D,Q,s),
                    enforce_stationarity=False,
                    enforce_invertibility=False
                )
                res = model.fit(disp=False)
                if res.aic < best_aic:
                    best_aic = res.aic
                    best_res = res
            except Exception:
                continue
    if best_res is None:
        return None
    try:
        return np.asarray(best_res.get_forecast(steps=n_steps).predicted_mean, dtype=float)
    except Exception:
        return None

registros_sarima = []
for (b, t) in parejas:
    y_tr = serie_mensual(df_train_fd1_v5_ITE1_SARIMA, (b, t), col_y=TARGET)
    y_te = serie_mensual(df_test_2024, (b, t), col_y=TARGET)
    y_te_2024 = y_te.reindex(idx_2024).fillna(0.0)
    yhat = predecir_sarima(y_tr.values, n_steps=len(idx_2024))
    if yhat is None:
        continue
    for fecha, real, pred in zip(idx_2024, y_te_2024.values, yhat):
        registros_sarima.append({
            "ID_BUILDING": b,
            "FM_COST_TYPE": t,
            "FECHA": fecha,
            "y_real": float(real),
            "y_hat": float(pred),
            "model": "SARIMA"
        })

preds_SARIMA_2024 = pd.DataFrame(registros_sarima)
print("Predicciones SARIMA 2024 generadas:", preds_SARIMA_2024.shape)
print(preds_SARIMA_2024.head(10))

/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "
/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "

Predicciones SARIMA 2024 generadas: (5700, 6)
   ID_BUILDING     FM_COST_TYPE      FECHA   y_real        y_hat   model
0            2  Mtto. Contratos 2024-01-01  1874.00  3342.385380  SARIMA
1            2  Mtto. Contratos 2024-02-01     0.00   581.211202  SARIMA
2            2  Mtto. Contratos 2024-03-01     0.00   556.790898  SARIMA
3            2  Mtto. Contratos 2024-04-01   175.58  1232.007561  SARIMA
4            2  Mtto. Contratos 2024-05-01   175.58  1908.993900  SARIMA
5            2  Mtto. Contratos 2024-06-01   175.58   551.560840  SARIMA
6            2  Mtto. Contratos 2024-07-01   175.58  1307.669885  SARIMA
7            2  Mtto. Contratos 2024-08-01   175.58   542.239712  SARIMA
8            2  Mtto. Contratos 2024-09-01   175.58   839.468846  SARIMA
9            2  Mtto. Contratos 2024-10-01   175.58  1234.156158  SARIMA

/usr/local/lib/python3.12/dist-packages/statsmodels/base/model.py:607: ConvergenceWarning: Maximum Likelihood optimization failed to converge. Check mle_retvals
  warnings.warn("Maximum Likelihood optimization failed to "

# Unificamos predicciones (si no hay ETS, usamos solo SARIMA)
if "preds_ETS_2024" in globals() and not preds_ETS_2024.empty:
    preds_all_2024 = pd.concat([preds_ETS_2024, preds_SARIMA_2024], ignore_index=True)
else:
    preds_all_2024 = preds_SARIMA_2024.copy()

# Añadimos MASE a las métricas por pareja y modelo (usamos m=12 para naive estacional mensual)
# Requisitos: tener disponible pair_series() y df_train_fd1_v5_ITE1_SARIMA
assert "pair_series" in globals(), "Falta pair_series(); ejecútese el bloque donde la definimos."
assert "df_train_fd1_v5_ITE1_SARIMA" in globals(), "Falta df_train_fd1_v5_ITE1_SARIMA en memoria."

registros_metricas = []
for (b, t, m), g in preds_all_2024.groupby(["ID_BUILDING", "FM_COST_TYPE", "model"]):
    y_true = g["y_real"].values
    y_hat  = g["y_hat"].values
    # Obtenemos la serie de train para calcular MASE respecto a naive estacional
    y_tr = pair_series(df_train_fd1_v5_ITE1_SARIMA, (b, t)).values
    registros_metricas.append({
        "ID_BUILDING": b,
        "FM_COST_TYPE": t,
        "model": m,
        "MAE": mae(y_true, y_hat),                 # escala del dato
        "MAPE_pct": mape(y_true, y_hat),           # porcentaje según función del notebook
        "WAPE": wape(y_true, y_hat),               # 0-1 (ponderado por nivel)
        "SMAPE_pct": smape(y_true, y_hat),         # porcentaje según función del notebook
        "MASE": mase(y_tr, y_true, y_hat, m=12),   # respecto a naive estacional mensual
        "n_obs": len(g)
    })

df_metricas_2024 = (
    pd.DataFrame(registros_metricas)
    .sort_values(["model", "WAPE", "MAE"])
    .reset_index(drop=True)
)

# Resumen global por modelo (colapsamos por meses y parejas)
# Para MASE_global usamos la media ponderada por n_obs de la MASE por pareja (no tiene sentido recalcular MASE global sin train por pareja)
resumen_global = []
for m, g in preds_all_2024.groupby("model"):
    y_true = g["y_real"].values
    y_hat  = g["y_hat"].values

    metricas_m = df_metricas_2024[df_metricas_2024["model"] == m]
    peso = metricas_m["n_obs"].to_numpy()
    mase_vals = metricas_m["MASE"].to_numpy()
    valid = (~np.isnan(mase_vals)) & (peso > 0)
    mase_global = np.nan if valid.sum() == 0 else np.nansum(mase_vals[valid] * peso[valid]) / np.sum(peso[valid])

    resumen_global.append({
        "model": m,
        "MAE_global": mae(y_true, y_hat),
        "MAPE_global_pct": mape(y_true, y_hat),
        "WAPE_global": wape(y_true, y_hat),
        "SMAPE_global_pct": smape(y_true, y_hat),
        "MASE_global": mase_global,
        "n_filas": len(g)
    })

df_resumen_global_2024 = (
    pd.DataFrame(resumen_global)
    .sort_values("WAPE_global")
    .reset_index(drop=True)
)

print("Predicciones unificadas:", preds_all_2024.shape)
print("Métricas por pareja y modelo:", df_metricas_2024.shape)
print("Resumen global por modelo:")
print(df_resumen_global_2024)

Predicciones unificadas: (11400, 6)
Métricas por pareja y modelo: (950, 9)
Resumen global por modelo:
    model    MAE_global  MAPE_global_pct  WAPE_global  SMAPE_global_pct  \
0     ETS   1338.020293      9526.005176     0.302550         51.835090
1  SARIMA  15658.454548     78220.399089     3.540659         56.277935

   MASE_global  n_filas
0     2.103296     5700
1    13.020476     5700

# Resumen global por modelo (agregamos sobre todas las parejas y meses)
resumen_global = []
for m, g in preds_all_2024.groupby("model"):
    y_true = g["y_real"].values
    y_hat  = g["y_hat"].values
    # Para MASE global usamos la media ponderada por n_obs sobre df_metricas_2024
    metricas_m = df_metricas_2024[df_metricas_2024["model"] == m]
    peso = metricas_m["n_obs"].values
    mase_vals = metricas_m["MASE"].values
    mase_global = np.nan if peso.sum() == 0 else np.nansum(mase_vals * peso) / np.sum(peso)

    resumen_global.append({
        "model": m,
        "MAE_global": mae(y_true, y_hat),
        "MAPE_global_pct": mape(y_true, y_hat),
        "WAPE_global": wape(y_true, y_hat),
        "SMAPE_global_pct": smape(y_true, y_hat),
        "MASE_global": mase_global,
        "n_filas": len(g)
    })

df_resumen_global_2024 = pd.DataFrame(resumen_global).sort_values("WAPE_global").reset_index(drop=True)
print("Resumen global por modelo:")
print(df_resumen_global_2024)

Resumen global por modelo:
    model    MAE_global  MAPE_global_pct  WAPE_global  SMAPE_global_pct  \
0     ETS   1338.020293      9526.005176     0.302550         51.835090
1  SARIMA  15658.454548     78220.399089     3.540659         56.277935

   MASE_global  n_filas
0     2.103296     5700
1    13.020476     5700

# Guardamos files con sufijo ETS_S (ETS estacional) y SARIMA.
# 1) Guardamos predicciones SARIMA y ETS_S con sus reales
# 2) Guardamos métricas filtradas por modelo (SARIMA y ETS) desde df_metricas_2024
# 3) Forzamos el sufijo ETS_S en los ficheros de salida para no confundir con EST

import os

# Definimos la carpeta base en Drive y nos aseguramos de que exista
base_dir = "/content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO"
os.makedirs(base_dir, exist_ok=True)

# Comprobamos que tenemos en memoria las tablas esperadas
assert "df_metricas_2024" in globals(), "No encontramos df_metricas_2024"

# Filtramos métricas por modelo
metrics_sarima = df_metricas_2024[df_metricas_2024["model"] == "SARIMA"].copy()
metrics_ets_s  = df_metricas_2024[df_metricas_2024["model"] == "ETS"].copy()

# Rutas de salida (mantenemos el estilo de nombres e imponemos ETS_S)
pred_sarima_path   = os.path.join(base_dir, "df_pred_fd1_v5_ITE1_sarima.csv")
metrics_sarima_path= os.path.join(base_dir, "df_metrics_fd1_v5_ITE1_sarima.csv")

pred_ets_s_path    = os.path.join(base_dir, "df_pred_fd1_v5_ITE1_ETS_S.csv")
metrics_ets_s_path = os.path.join(base_dir, "df_metrics_fd1_v5_ITE1_ETS_S.csv")

# Guardamos predicciones SARIMA (si existen)
if "preds_SARIMA_2024" in globals() and not preds_SARIMA_2024.empty:
    preds_SARIMA_2024.to_csv(pred_sarima_path, index=False)
else:
    print("Aviso: no encontramos preds_SARIMA_2024 o está vacío; no guardamos predicciones SARIMA.")

# Guardamos predicciones ETS_S (si existen)
if "preds_ETS_2024" in globals() and not preds_ETS_2024.empty:
    preds_ETS_2024.to_csv(pred_ets_s_path, index=False)
else:
    print("Aviso: no encontramos preds_ETS_2024 o está vacío; no guardamos predicciones ETS_S.")

# Guardamos métricas por modelo
metrics_sarima.to_csv(metrics_sarima_path, sep=";", index=False)
metrics_ets_s.to_csv(metrics_ets_s_path,sep=";", index=False)

# Mostramos confirmación de guardado
print("Guardados en Drive:")
if "preds_SARIMA_2024" in globals() and not preds_SARIMA_2024.empty:
    print("Predicciones SARIMA:", pred_sarima_path)
else:
    print("Predicciones SARIMA: no disponibles")

if "preds_ETS_2024" in globals() and not preds_ETS_2024.empty:
    print("Predicciones ETS_S :", pred_ets_s_path)
else:
    print("Predicciones ETS_S : no disponibles")

print("Métricas SARIMA   :", metrics_sarima_path)
print("Métricas ETS_S    :", metrics_ets_s_path)

Guardados en Drive:
Predicciones SARIMA: /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/df_pred_fd1_v5_ITE1_sarima.csv
Predicciones ETS_S : /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/df_pred_fd1_v5_ITE1_ETS_S.csv
Métricas SARIMA   : /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/df_metrics_fd1_v5_ITE1_sarima.csv
Métricas ETS_S    : /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/df_metrics_fd1_v5_ITE1_ETS_S.csv

# Definimos carpeta y sufijo a cargar (cambiar TAG según el sub_dataframe)
OUT_DIR = "/content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO"
TAG = "HOLT"  # ejemplos: "SARIMA", "ARIMA_filtered", "ETS_S"

# Rutas de ficheros esperados
train_xlsx = os.path.join(OUT_DIR, f"df_train_fd1_v5_ITE1_{TAG}.xlsx")
test_xlsx  = os.path.join(OUT_DIR, f"df_test_fd1_v5_ITE1_{TAG}.xlsx")
train_csv  = os.path.join(OUT_DIR, f"df_train_fd1_v5_ITE1_{TAG}.csv")
test_csv   = os.path.join(OUT_DIR, f"df_test_fd1_v5_ITE1_{TAG}.csv")

def _load_df(xlsx_path, csv_path):
    # Intentamos Excel y, si no existe, probamos CSV con sep=';'
    if os.path.exists(xlsx_path):
        df = pd.read_excel(xlsx_path)
        origen = xlsx_path
    elif os.path.exists(csv_path):
        df = pd.read_csv(csv_path, sep=";")
        origen = csv_path
    else:
        raise FileNotFoundError(f"No encontramos ni Excel ni CSV:\n - {xlsx_path}\n - {csv_path}")
    # Normalizamos FECHA si existe
    if "FECHA" in df.columns:
        df["FECHA"] = pd.to_datetime(df["FECHA"], errors="coerce")
    return df, origen

# Cargamos TRAIN y TEST
df_train, origen_train = _load_df(train_xlsx, train_csv)
df_test,  origen_test  = _load_df(test_xlsx,  test_csv)

print("Cargado TRAIN desde:", origen_train, "| Shape:", df_train.shape)
print("Cargado TEST  desde:", origen_test,  "| Shape:", df_test.shape)

# Registramos con nombre estándar en globals() para mantener la convención del proyecto
globals()[f"df_train_fd1_v5_ITE1_{TAG}"] = df_train
globals()[f"df_test_fd1_v5_ITE1_{TAG}"]  = df_test

# Mostramos una vista rápida
print("\nTRAIN.head():")
print(df_train.head(3))
print("\nTEST.head():")
print(df_test.head(3))
print("\nTRAIN.shape:")
print(df_train.shape)
print("\nTEST.shape:")
print(df_test.shape)

Cargado TRAIN desde: /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/df_train_fd1_v5_ITE1_HOLT.xlsx | Shape: (899, 14)
Cargado TEST  desde: /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/df_test_fd1_v5_ITE1_HOLT.xlsx | Shape: (368, 14)

TRAIN.head():
   ID_BUILDING      FM_COST_TYPE      FECHA  YEAR  MONTH  cost_float_mod  \
0          121  Mtto. Correctivo 2021-01-01  2021      1          663.69
1          121  Mtto. Correctivo 2021-02-01  2021      2          549.00
2          121  Mtto. Correctivo 2021-03-01  2021      3         1349.58

  COUNTRY_DEF ID_REGION_GRUPO TIPO_USO SUPPLIER_TYPE_MOD_2 FM_RESPONSIBLE_MOD  \
0      España              17    Bingo             INTERNO      Mantenimiento
1      España              17    Bingo             INTERNO      Mantenimiento
2      España              17    Bingo             INTERNO      Mantenimiento

   diagnosis        model_suggested model_group
0  tendencia  Holt(ETS)|ARIMA_drift        HOLT
1  tendencia  Holt(ETS)|ARIMA_drift        HOLT
2  tendencia  Holt(ETS)|ARIMA_drift        HOLT

TEST.head():
   ID_BUILDING      FM_COST_TYPE      FECHA  YEAR  MONTH  cost_float_mod  \
0          121  Mtto. Correctivo 2024-01-01  2024      1          140.42
1          121  Mtto. Correctivo 2024-02-01  2024      2          407.78
2          121  Mtto. Correctivo 2024-03-01  2024      3         1419.69

  COUNTRY_DEF ID_REGION_GRUPO TIPO_USO SUPPLIER_TYPE_MOD_2 FM_RESPONSIBLE_MOD  \
0      España              17    Bingo             INTERNO      Mantenimiento
1      España              17    Bingo             INTERNO      Mantenimiento
2      España              17    Bingo             INTERNO      Mantenimiento

   diagnosis        model_suggested model_group
0  tendencia  Holt(ETS)|ARIMA_drift        HOLT
1  tendencia  Holt(ETS)|ARIMA_drift        HOLT
2  tendencia  Holt(ETS)|ARIMA_drift        HOLT

TRAIN.shape:
(899, 14)

TEST.shape:
(368, 14)

# ------------------------------------------------------------
# 1) Función de pronóstico HOLT con fallbacks mínimos
# ------------------------------------------------------------
def forecast_holt(y_train: pd.Series, h: int) -> np.ndarray:
    """
    y_train: Serie mensual (índice 'MS'), float.
    h: pasos a predecir (nº meses en el test de esa pareja)
    """
    n = len(y_train)
    if n == 0:
        return np.zeros(h, dtype=float)

    # Fallbacks para series cortas
    if n < 2:
        last = float(y_train.iloc[-1])
        return np.repeat(last, h)

    if n == 2:
        try:
            model = SimpleExpSmoothing(y_train, initialization_method="estimated").fit(optimized=True)
            return model.forecast(h)
        except Exception:
            last = float(y_train.iloc[-1])
            return np.repeat(last, h)

    # Holt con tendencia amortiguada (sin estacionalidad)
    try:
        model = ExponentialSmoothing(
            y_train,
            trend="add",
            damped_trend=True,
            seasonal=None,
            initialization_method="estimated",
        ).fit(optimized=True)
        return model.forecast(h)
    except Exception:
        # Último fallback robusto
        try:
            model = SimpleExpSmoothing(y_train, initialization_method="estimated").fit(optimized=True)
            return model.forecast(h)
        except Exception:
            last = float(y_train.iloc[-1])
            return np.repeat(last, h)

# ------------------------------------------------------------
# 2) Preparación ligera y pares a evaluar (usando df_train / df_test)
# ------------------------------------------------------------
# Aseguramos tipos de fecha y orden
df_train = df_train.copy()
df_test  = df_test.copy()
df_train["FECHA"] = pd.to_datetime(df_train["FECHA"])
df_test["FECHA"]  = pd.to_datetime(df_test["FECHA"])

df_train.sort_values(["ID_BUILDING","FM_COST_TYPE", "FECHA"], inplace=True)
df_test.sort_values (["ID_BUILDING","FM_COST_TYPE", "FECHA"], inplace=True)

# Sólo iteramos por parejas presentes en train y test
pairs_train = df_train[["ID_BUILDING","FM_COST_TYPE"]].drop_duplicates()
pairs_test  = df_test[["ID_BUILDING","FM_COST_TYPE"]].drop_duplicates()
pairs = pairs_train.merge(pairs_test, on=["ID_BUILDING","FM_COST_TYPE"])

# ------------------------------------------------------------
# 3) Loop principal por pareja
# ------------------------------------------------------------
preds_rows = []
metricas_rows = []

for bid, cty in pairs.itertuples(index=False, name=None):
    # Serie mensual continua train/test usando tus helpers
    y_tr = pair_series(df_train, (bid, cty))  # Serie mensual (índice MS), float
    y_te = pair_series(df_test,  (bid, cty))  # Índice de meses reales y y_true

    if y_tr.empty or y_te.empty:
        continue

    h = len(y_te)
    y_hat = forecast_holt(y_tr, h)

    # Alineamos a las fechas de test
    df_pair_pred = pd.DataFrame({
        "ID_BUILDING": bid,
        "FM_COST_TYPE": cty,
        "FECHA": y_te.index,              # índice mensual del test
        "y_real": y_te.values.astype(float),
        "y_hat":  np.asarray(y_hat, dtype=float),
        "model": "Holt(ETS)_damped"
    })
    preds_rows.append(df_pair_pred)

# Métricas por pareja (reutilizamos tus funciones)
    y_true = df_pair_pred["y_real"].values
    y_pred = df_pair_pred["y_hat"].values

    metricas_rows.append({
        "ID_BUILDING": bid,
        "FM_COST_TYPE": cty,
        "n_train": len(y_tr),
        "n_test":  h,
        "MAE": mae(y_true, y_pred),
        "MAPE_pct": mape(y_true, y_pred),
        "WAPE": wape(y_true, y_pred),
        "SMAPE_pct": smape(y_true, y_pred),
        "MASE1": mase_no_fallback(y_tr.values, y_true, y_pred, m=1),   # baseline naive simple
        "MASE12": mase12(y_tr.values, y_true, y_pred),                 # baseline naive estacional mensual
        "model": "Holt(ETS)_damped"
    })

# ------------------------------------------------------------
# 4) Salidas finales
# ------------------------------------------------------------
preds_holt = pd.concat(preds_rows, ignore_index=True) if preds_rows else pd.DataFrame(
    columns=["ID_BUILDING","FM_COST_TYPE","FECHA","y_real","y_hat","model"]
)
metricas_holt = pd.DataFrame(metricas_rows)

# Resumen global (WAPE ya es ponderado por suma de |y|)
if not preds_holt.empty:
    y_all = preds_holt["y_real"].values
    f_all = preds_holt["y_hat"].values
    resumen_global = {
        "model": "Holt(ETS)_damped",
        "MAE_global": mae(y_all, f_all),
        "MAPE_global_pct": mape(y_all, f_all),
        "WAPE_global": wape(y_all, f_all),
        "SMAPE_global_pct": smape(y_all, f_all),
        "MASE1_global_mean": float(np.nanmean(metricas_holt["MASE1"])) if not metricas_holt.empty else np.nan,
        "MASE12_global_mean": float(np.nanmean(metricas_holt["MASE12"])) if not metricas_holt.empty else np.nan
    }
else:
    resumen_global = {
        "model": "Holt(ETS)_damped",
        "MAE_global": np.nan,
        "MAPE_global_pct": np.nan,
        "WAPE_global": np.nan,
        "SMAPE_global_pct": np.nan,
        "MASE1_global_mean": np.nan,
        "MASE12_global_mean": np.nan
    }

print("Resumen global HOLT:", resumen_global)

# === Tabla de resumen global (1 fila) ===
resumen_cols = ["model","MAE_global","MAPE_global_pct","WAPE_global","SMAPE_global_pct","MASE1_global_mean", "MASE12_global_mean"]
resumen_df = pd.DataFrame([resumen_global])[resumen_cols]

# Formateo bonito (2 decimales en %)
resumen_fmt = resumen_df.copy()
for c in ["MAPE_global_pct","SMAPE_global_pct"]:
    resumen_fmt[c] = resumen_fmt[c].map(lambda x: f"{x:.2f}%" if pd.notnull(x) else "NaN")
for c in ["MAE_global","WAPE_global","MASE1_global_mean", "MASE12_global_mean"]:
    resumen_fmt[c] = resumen_fmt[c].map(lambda x: f"{x:.4f}" if pd.notnull(x) else "NaN")

display(resumen_fmt)  # En Jupyter/Colab se verá como tabla

Resumen global HOLT: {'model': 'Holt(ETS)_damped', 'MAE_global': np.float64(867.3833794374615), 'MAPE_global_pct': np.float64(2418.3251645925106), 'WAPE_global': np.float64(0.5507451663956814), 'SMAPE_global_pct': np.float64(55.610707459262244), 'MASE1_global_mean': 1.9531936831932684, 'MASE12_global_mean': 1.082819284692108}

# --- TOP y BOTTOM 10 por WAPE ---
if not metricas_holt.empty:
    # Ordenamos
    top10 = metricas_holt.sort_values("WAPE").head(10).copy()
    bot10 = metricas_holt.sort_values("WAPE").tail(10).copy()

    def _fmt(df):
        df2 = df[[
            "ID_BUILDING","FM_COST_TYPE","n_train","n_test",
            "MAE","MAPE_pct","WAPE","SMAPE_pct","MASE1","MASE12","model"
        ]].copy()
        # Formateo
        df2["MAPE_pct"] = df2["MAPE_pct"].map(lambda x: f"{x:.2f}%" if pd.notnull(x) else "NaN")
        df2["SMAPE_pct"] = df2["SMAPE_pct"].map(lambda x: f"{x:.2f}%" if pd.notnull(x) else "NaN")
        for c in ["MAE","WAPE","MASE1","MASE12"]:
            df2[c] = df2[c].map(lambda x: f"{x:.4f}" if pd.notnull(x) else "NaN")
        return df2

    print("\n=== TOP 10 (mejor WAPE) ===")
    print(_fmt(top10).to_string(index=False))

    print("\n=== BOTTOM 10 (peor WAPE) ===")
    print(_fmt(bot10).to_string(index=False))
else:
    print("metricas_holt está vacío; no hay tablas por pareja que mostrar.")

=== TOP 10 (mejor WAPE) ===
 ID_BUILDING          FM_COST_TYPE  n_train  n_test      MAE MAPE_pct   WAPE SMAPE_pct  MASE1 MASE12            model
        1078 Eficiencia Energética       18      12  10.5190    4.58% 0.0463     4.71% 0.3727 0.0616 Holt(ETS)_damped
     1001207       Servicios Ctto.       36      12 323.3000    5.96% 0.0595     6.06% 2.1772 0.1541 Holt(ETS)_damped
        1099 Eficiencia Energética       18      12  14.8568    6.43% 0.0640     6.30% 0.4622 0.0789 Holt(ETS)_damped
        1091 Eficiencia Energética       18      12  23.8365    8.77% 0.0872     8.55% 0.6358 0.1045 Holt(ETS)_damped
        1079 Eficiencia Energética       18      12  27.1089   10.46% 0.1048    11.05% 0.8309 0.1314 Holt(ETS)_damped
        1086 Eficiencia Energética       18      12  22.6732   10.48% 0.1050    11.08% 0.8603 0.1384 Holt(ETS)_damped
     1000632 Eficiencia Energética       18      12  18.8121   11.08% 0.1111    11.75% 0.7548 0.1107 Holt(ETS)_damped
        1080 Eficiencia Energética       18      12  10.8663   13.13% 0.1322    14.13% 0.7713 0.1059 Holt(ETS)_damped
     1000872       Mtto. Contratos       33      12  52.2789   14.41% 0.1441    13.21% 5.6797 0.4102 Holt(ETS)_damped
     1001208       Servicios Extra       34      12 641.9988   15.55% 0.1493    14.78% 1.1185 0.6240 Holt(ETS)_damped

=== BOTTOM 10 (peor WAPE) ===
 ID_BUILDING          FM_COST_TYPE  n_train  n_test       MAE  MAPE_pct   WAPE SMAPE_pct  MASE1 MASE12            model
     1000437       Servicios Extra       34      12 1995.1925   236.95% 0.8555    89.33% 2.2467 2.4625 Holt(ETS)_damped
     1000587      Mtto. Correctivo       33       9  335.2452 13389.23% 1.1477    95.04% 1.5610 1.0620 Holt(ETS)_damped
         150      Mtto. Correctivo       34      12 1398.1271 20830.73% 1.2976    94.94% 3.8533 3.0593 Holt(ETS)_damped
     1001339 Eficiencia Energética       18      12  487.1168   137.95% 1.4175   176.97% 0.7121 0.4424 Holt(ETS)_damped
         124      Mtto. Correctivo       36      12 1207.3085 13073.95% 1.9500   121.53% 2.3830 2.2572 Holt(ETS)_damped
     1000481       Mtto. Contratos       36      12 1379.8152   201.02% 1.9764    97.24% 7.9489 4.2488 Holt(ETS)_damped
        1242      Mtto. Correctivo       36      12  471.6437  6824.00% 2.3121   125.31% 1.6858 1.3818 Holt(ETS)_damped
         141      Mtto. Correctivo       36      12 1580.5699 21008.99% 2.3635   135.75% 2.4908 2.6915 Holt(ETS)_damped
         131      Mtto. Correctivo       36      12 1502.6033   595.10% 2.4494   118.66% 2.1231 1.9583 Holt(ETS)_damped
         121      Mtto. Correctivo       36      12 2319.8337   459.61% 2.6402   119.02% 2.4434 2.4216 Holt(ETS)_damped

import matplotlib.pyplot as plt

# --- Elegimos un TOP y un BOTTOM ---
best_pair  = metricas_holt.sort_values("WAPE").iloc[0]   # mejor WAPE
worst_pair = metricas_holt.sort_values("WAPE").iloc[-1]  # peor WAPE

def plot_pair(pair_row, preds_df):
    bid, cty = pair_row["ID_BUILDING"], pair_row["FM_COST_TYPE"]
    dfp = preds_df[(preds_df.ID_BUILDING==bid) & (preds_df.FM_COST_TYPE==cty)].copy()
    dfp = dfp.sort_values("FECHA")

    plt.figure(figsize=(10,5))
    plt.plot(dfp["FECHA"], dfp["y_real"], marker="o", label="Real")
    plt.plot(dfp["FECHA"], dfp["y_hat"], marker="x", label="Predicho (HOLT)")
    plt.title(f"ID_BUILDING={bid} | {cty}")
    plt.xlabel("Fecha")
    plt.ylabel("Coste")
    plt.legend()
    plt.grid(True)
    plt.show()

# --- Gráficas ---
print("=== Caso TOP (mejor WAPE) ===")
plot_pair(best_pair, preds_holt)

print("=== Caso BOTTOM (peor WAPE) ===")
plot_pair(worst_pair, preds_holt)

=== Caso TOP (mejor WAPE) ===

=== Caso BOTTOM (peor WAPE) ===

# Cargamos datasets df_fd1_v5_ITE1 y df_fd1_v5_Real_ITE1_2024

# Ruta base donde guardaste los outputs
ruta_base = "/content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/"

# Dataset de entrenamiento (Iteración 1: 2021-2023)
df_fd1_v5_ITE1 = pd.read_csv(ruta_base + "df_fd1_v5_ITE1.csv", sep=";")

# Dataset de reales (Iteración 1: datos del 2024)
df_fd1_v5_Real_ITE1_2024 = pd.read_csv(ruta_base + "df_fd1_v5_Real_ITE1_2024.csv", sep=";")

# Confirmamos tamaños y columnas
print("ITE1 shape:", df_fd1_v5_ITE1.shape)
print("Real_ITE1_2024 shape:", df_fd1_v5_Real_ITE1_2024.shape)
print("\nColumnas ITE1:", df_fd1_v5_ITE1.columns.tolist())
print("Columnas Real_ITE1_2024:", df_fd1_v5_Real_ITE1_2024.columns.tolist())

ITE1 shape: (174626, 12)
Real_ITE1_2024 shape: (69290, 12)

Columnas ITE1: ['ID_ORDER', 'ID_BUILDING', 'FM_COST_TYPE', 'MONTH', 'YEAR', 'cost_float_mod', 'SUPPLIER_TYPE_MOD_2', 'FM_RESPONSIBLE_MOD', 'TIPO_USO', 'COUNTRY_CATALOGO', 'ID_REGION_GRUPO', 'COUNTRY_DEF']
Columnas Real_ITE1_2024: ['ID_ORDER', 'ID_BUILDING', 'FM_COST_TYPE', 'MONTH', 'YEAR', 'cost_float_mod', 'SUPPLIER_TYPE_MOD_2', 'FM_RESPONSIBLE_MOD', 'TIPO_USO', 'COUNTRY_CATALOGO', 'ID_REGION_GRUPO', 'COUNTRY_DEF']

# ===============================================================
# Helper 0) Funciones auxiliares
# ===============================================================

def crear_fecha(df):
    """
    Creamos la columna FECHA como primer día de mes a partir de YEAR y MONTH.
    Controlamos tipos y meses válidos.
    """
    df = df.copy()
    df["YEAR"] = pd.to_numeric(df["YEAR"], errors="coerce").astype("Int64")
    df["MONTH"] = pd.to_numeric(df["MONTH"], errors="coerce").astype("Int64")
    df = df[(df["MONTH"] >= 1) & (df["MONTH"] <= 12)]
    df["FECHA"] = pd.to_datetime(
        dict(year=df["YEAR"].astype(int), month=df["MONTH"].astype(int), day=1),
        errors="coerce"
    )
    return df

def serie_mensual_con_huecos(df, year_ini, year_fin):
    """
    A partir del detalle (varias líneas por mes), agregamos por pareja y FECHA.
    Devolvemos la serie mensual agregada (suma de cost_float_mod) con los meses observados.
    """
    df = df.copy()
    df = df[(df["FECHA"] >= f"{year_ini}-01-01") & (df["FECHA"] <= f"{year_fin}-12-31")]
    agg = (df.groupby(["ID_BUILDING","FM_COST_TYPE","FECHA"], as_index=False)["cost_float_mod"]
             .sum())
    return agg

def completar_meses_por_pareja(agg_df):
    """
    Para cada pareja (ID_BUILDING, FM_COST_TYPE) reindexamos entre su primer y último mes observado
    y rellenamos los meses ausentes con 0. No extendemos fuera del rango observado de esa pareja.
    """
    pares = agg_df[["ID_BUILDING","FM_COST_TYPE"]].drop_duplicates()
    out = []

    # Recorremos cada pareja y reindexamos su rango observado
    for _, row in pares.iterrows():
        bid, ctype = row["ID_BUILDING"], row["FM_COST_TYPE"]
        sub = agg_df[(agg_df["ID_BUILDING"]==bid) & (agg_df["FM_COST_TYPE"]==ctype)].copy()
        sub = sub.set_index("FECHA").sort_index()
        # Rango observado de esa pareja
        fmin, fmax = sub.index.min(), sub.index.max()
        # Reindexación a frecuencia mensual únicamente entre fmin y fmax
        idx = pd.date_range(fmin, fmax, freq="MS")
        sub = sub.reindex(idx)
        sub["ID_BUILDING"] = bid
        sub["FM_COST_TYPE"] = ctype
        # Rellenamos NaN de coste con 0
        sub["cost_float_mod"] = sub["cost_float_mod"].fillna(0.0)
        sub = sub.reset_index().rename(columns={"index":"FECHA"})
        out.append(sub)

    out = pd.concat(out, ignore_index=True)
    # Ordenamos
    out = out[["ID_BUILDING","FM_COST_TYPE","FECHA","cost_float_mod"]].sort_values(
        ["ID_BUILDING","FM_COST_TYPE","FECHA"]
    ).reset_index(drop=True)
    return out

def moda_segura(s):
    """
    Calculamos el modo (valor más frecuente) de una serie, devolviendo el primero en caso de empate.
    Si todo es NaN, devolvemos NaN.
    """
    s = s.dropna()
    if s.empty:
        return np.nan
    m = s.mode()
    return m.iloc[0] if len(m) > 0 else np.nan

def contexto_por_pareja(df_base_2021_2023):
    """
    A partir del detalle 2021–2023, calculamos el modo por pareja de las variables de contexto.
    """
    cols_ctx = ["COUNTRY_DEF","ID_REGION_GRUPO","TIPO_USO","SUPPLIER_TYPE_MOD_2","FM_RESPONSIBLE_MOD"]
    grp = df_base_2021_2023.groupby(["ID_BUILDING","FM_COST_TYPE"], as_index=False)
    ctx = grp.agg({col: moda_segura for col in cols_ctx})
    return ctx

def ensamblar_final(serie_completa, ctx):
    """
    Unimos la serie mensual completa con el contexto por pareja y añadimos YEAR y MONTH.
    Validamos duplicados.
    """
    df = serie_completa.merge(ctx, on=["ID_BUILDING","FM_COST_TYPE"], how="left")
    df["YEAR"] = df["FECHA"].dt.year.astype(int)
    df["MONTH"] = df["FECHA"].dt.month.astype(int)
    # Validamos unicidad por pareja y FECHA
    dup_mask = df.duplicated(subset=["ID_BUILDING","FM_COST_TYPE","FECHA"], keep=False)
    assert not dup_mask.any(), "Hay duplicados por pareja y FECHA tras el ensamblado."
    # Orden práctico de columnas
    ordered_cols = [
        "ID_BUILDING","FM_COST_TYPE","FECHA","YEAR","MONTH","cost_float_mod",
        "COUNTRY_DEF","ID_REGION_GRUPO","TIPO_USO","SUPPLIER_TYPE_MOD_2","FM_RESPONSIBLE_MOD"
    ]
    # Algunas columnas de contexto pueden no existir si venían vacías; controlamos
    for c in ordered_cols:
        if c not in df.columns:
            df[c] = np.nan
    df = df[ordered_cols].sort_values(["ID_BUILDING","FM_COST_TYPE","FECHA"]).reset_index(drop=True)
    return df

def chequeo_rapido(nombre, df, anio_ini, anio_fin, n_muestra=5):
    """
    Hacemos un chequeo de forma, rango temporal y mostramos una pequeña muestra para validar.
    """
    print(f"=== {nombre} ===")
    print("Shape:", df.shape)
    print("FECHA min:", df["FECHA"].min(), "| FECHA max:", df["FECHA"].max())
    # Comprobamos rango esperado
    assert df["FECHA"].min() >= pd.to_datetime(f"{anio_ini}-01-01"), "FECHA mínima fuera de rango"
    assert df["FECHA"].max() <= pd.to_datetime(f"{anio_fin}-12-31"), "FECHA máxima fuera de rango"
    # Duplicados
    dups = df.duplicated(subset=["ID_BUILDING","FM_COST_TYPE","FECHA"]).sum()
    print("Duplicados por pareja/FECHA:", dups)
    # Muestra
    print("\nMuestra de filas:")
    print(df.head(n_muestra))
    print("="*60, "\n")

# ===============================================================
# Secuencia ITE1) Generamos df_fd1_v5_ITE1_train y df_fd1_v5_ITE1_test
# Partimos de:
#   - df_fd1_v5_ITE1               (detalle 2021–2023)
#   - df_fd1_v5_Real_ITE1_2024     (detalle 2024)
# ===============================================================

# 1) Creamos FECHA en ambos orígenes
df_ite1_detalle_train = crear_fecha(df_fd1_v5_ITE1)
df_ite1_detalle_test  = crear_fecha(df_fd1_v5_Real_ITE1_2024)

# 2) Filtramos por YEAR explícitamente
df_ite1_detalle_train = df_ite1_detalle_train[(df_ite1_detalle_train["YEAR"]>=2021) & (df_ite1_detalle_train["YEAR"]<=2023)]
df_ite1_detalle_test  = df_ite1_detalle_test[(df_ite1_detalle_test["YEAR"]==2024)]

# 3) Agregamos a nivel mensual por pareja (suma de cost_float_mod)
agg_train = serie_mensual_con_huecos(df_ite1_detalle_train, 2021, 2023)
agg_test  = serie_mensual_con_huecos(df_ite1_detalle_test,  2024, 2024)

# 4) Rellenamos meses faltantes entre el primer y último mes observado de cada pareja
serie_train_completa = completar_meses_por_pareja(agg_train)
serie_test_completa  = completar_meses_por_pareja(agg_test)

# 5) Calculamos contexto por pareja usando SOLO 2021–2023 (moda)
ctx_2021_2023 = contexto_por_pareja(df_ite1_detalle_train)

# 6) Ensamblamos finales con contexto y añadimos YEAR/MONTH
df_fd1_v5_ITE1_train = ensamblar_final(serie_train_completa, ctx_2021_2023)
df_fd1_v5_ITE1_test  = ensamblar_final(serie_test_completa,  ctx_2021_2023)  # usamos el mismo contexto

# 7) Comprobaciones rápidas
chequeo_rapido("df_fd1_v5_ITE1_train (2021–2023)", df_fd1_v5_ITE1_train, 2021, 2023)
chequeo_rapido("df_fd1_v5_ITE1_test (2024)",       df_fd1_v5_ITE1_test,  2024, 2024)

=== df_fd1_v5_ITE1_train (2021–2023) ===
Shape: (65466, 11)
FECHA min: 2021-01-01 00:00:00 | FECHA max: 2023-12-01 00:00:00
Duplicados por pareja/FECHA: 0

Muestra de filas:
   ID_BUILDING           FM_COST_TYPE      FECHA  YEAR  MONTH  cost_float_mod  \
0            2  Eficiencia Energética 2021-12-01  2021     12            0.00
1            2              Licencias 2021-01-01  2021      1         1145.46
2            2              Licencias 2021-02-01  2021      2           55.95
3            2              Licencias 2021-03-01  2021      3            0.00
4            2              Licencias 2021-04-01  2021      4            0.00

  COUNTRY_DEF ID_REGION_GRUPO  TIPO_USO SUPPLIER_TYPE_MOD_2  \
0      España               2  Oficinas             INTERNO
1      España               2  Oficinas             INTERNO
2      España               2  Oficinas             INTERNO
3      España               2  Oficinas             INTERNO
4      España               2  Oficinas             INTERNO

      FM_RESPONSIBLE_MOD
0  Eficiencia Energética
1              Licencias
2              Licencias
3              Licencias
4              Licencias
============================================================

=== df_fd1_v5_ITE1_test (2024) ===
Shape: (23750, 11)
FECHA min: 2024-01-01 00:00:00 | FECHA max: 2024-12-01 00:00:00
Duplicados por pareja/FECHA: 0

Muestra de filas:
   ID_BUILDING     FM_COST_TYPE      FECHA  YEAR  MONTH  cost_float_mod  \
0            2        Licencias 2024-12-01  2024     12          202.80
1            2  Mtto. Contratos 2024-01-01  2024      1         1874.00
2            2  Mtto. Contratos 2024-02-01  2024      2            0.00
3            2  Mtto. Contratos 2024-03-01  2024      3            0.00
4            2  Mtto. Contratos 2024-04-01  2024      4          175.58

  COUNTRY_DEF ID_REGION_GRUPO  TIPO_USO SUPPLIER_TYPE_MOD_2 FM_RESPONSIBLE_MOD
0      España               2  Oficinas             INTERNO          Licencias
1      España               2  Oficinas             EXTERNO      Mantenimiento
2      España               2  Oficinas             EXTERNO      Mantenimiento
3      España               2  Oficinas             EXTERNO      Mantenimiento
4      España               2  Oficinas             EXTERNO      Mantenimiento
============================================================

# 1) Definimos los dataframes de trabajo
#    Filtramos por rango por seguridad utilizando YEAR (ya viene en los agregados)
train_df = df_fd1_v5_ITE1_train[
    (df_fd1_v5_ITE1_train["YEAR"] >= 2021) & (df_fd1_v5_ITE1_train["YEAR"] <= 2023)
].copy()

real_df = df_fd1_v5_ITE1_test[
    (df_fd1_v5_ITE1_test["YEAR"] == 2024)
].copy()

# 2) Comprobamos columnas mínimas
cols_req = ["ID_BUILDING","FM_COST_TYPE","FECHA","cost_float_mod"]
for name, df in [("train_df", train_df), ("real_df", real_df)]:
    faltan = set(cols_req) - set(df.columns)
    assert not faltan, f"Faltan columnas en {name}: {faltan}"

# 3) Identificamos parejas comunes entre ambos conjuntos
pairs_train = train_df[["ID_BUILDING","FM_COST_TYPE"]].drop_duplicates()
pairs_real  = real_df[["ID_BUILDING","FM_COST_TYPE"]].drop_duplicates()
pairs_both  = pairs_train.merge(pairs_real, on=["ID_BUILDING","FM_COST_TYPE"])
assert len(pairs_both) > 0, "No hay parejas en común entre 2021-2023 y 2024."

# 4) Elegimos una pareja aleatoria presente en ambos (si queremos reproducibilidad, fijamos random_state)
# pair = pairs_both.sample(1, random_state=None).iloc[0]
# bid, ctype = pair["ID_BUILDING"], pair["FM_COST_TYPE"]
bid, ctype = 1000480, "Mtto. Correctivo"  # para ser coherente con el resto del notebook lo fijamos
print(f"Pareja seleccionada -> ID_BUILDING: {bid} | FM_COST_TYPE: {ctype}")



# 5) Extraemos las dos series (train y real) para esa pareja
#    Como ya están agregadas por mes, cada FECHA es única por pareja.
s_train = (train_df[(train_df["ID_BUILDING"]==bid) & (train_df["FM_COST_TYPE"]==ctype)]
           .loc[:, ["FECHA","cost_float_mod"]]
           .rename(columns={"cost_float_mod":"valor"})
           .sort_values("FECHA")
           .reset_index(drop=True))

s_real = (real_df[(real_df["ID_BUILDING"]==bid) & (real_df["FM_COST_TYPE"]==ctype)]
          .loc[:, ["FECHA","cost_float_mod"]]
          .rename(columns={"cost_float_mod":"valor"})
          .sort_values("FECHA")
          .reset_index(drop=True))

# 6) Graficamos cada serie por separado para una lectura clara
plt.figure()
plt.plot(s_train["FECHA"], s_train["valor"])
plt.title(f"Serie 2021–2023 | ID {bid} · {ctype}")
plt.xlabel("Fecha"); plt.ylabel("Valor")
plt.tight_layout()
plt.show()

plt.figure()
plt.plot(s_real["FECHA"], s_real["valor"])
plt.title(f"Serie 2024 (Real) | ID {bid} · {ctype}")
plt.xlabel("Fecha"); plt.ylabel("Valor")
plt.tight_layout()
plt.show()

# 7) Preparamos salida en CSV simple
def to_simple_csv(df):
    out = df.copy()
    out["FECHA"] = out["FECHA"].dt.strftime("%Y-%m-%d")
    return out.to_csv(index=False)

print("\n=== SERIE ENTRENAMIENTO 2021-2023 ===")
print(to_simple_csv(s_train))

print("\n=== SERIE REAL 2024 ===")
print(to_simple_csv(s_real))

Pareja seleccionada -> ID_BUILDING: 1000480 | FM_COST_TYPE: Mtto. Correctivo

=== SERIE ENTRENAMIENTO 2021-2023 ===
FECHA,valor
2021-01-01,349.3419384615381
2021-02-01,1314.0121972027978
2021-03-01,268.4100153846158
2021-04-01,402.1224461538466
2021-05-01,728.7152615384616
2021-06-01,0.0
2021-07-01,695.6266769230766
2021-08-01,53.7296615384615
2021-09-01,453.01681538461565
2021-10-01,25.09999999999999
2021-11-01,36.70769230769229
2021-12-01,865.7773076923067
2022-01-01,101.11558461538458
2022-02-01,7.26769230769231
2022-03-01,215.92163076923072
2022-04-01,56.91538461538464
2022-05-01,375.5271999999998
2022-06-01,692.0130769230775
2022-07-01,248.4935999999999
2022-08-01,142.0847076923076
2022-09-01,380.8114307692304
2022-10-01,2189.4840307692307
2022-11-01,207.2112153846158
2022-12-01,439.4492000000002
2023-01-01,204.6260307692308
2023-02-01,430.231461538462
2023-03-01,19.3737230769231
2023-04-01,175.47186153846138
2023-05-01,263.4378000000001
2023-06-01,262.2343846153846
2023-07-01,79.1907538461539
2023-08-01,251.1000000000001
2023-09-01,354.1569692307689
2023-10-01,0.0
2023-11-01,69.92215384615385
2023-12-01,562.8817846153843


=== SERIE REAL 2024 ===
FECHA,valor
2024-01-01,481.277415384615
2024-02-01,137.23230769230773
2024-03-01,188.23076923076928
2024-04-01,30.710769230769213
2024-05-01,342.0891846153844
2024-06-01,158.61043076923073
2024-07-01,41.97695384615386
2024-08-01,48.961538461538396
2024-09-01,170.70307692307685
2024-10-01,30.423076923076902
2024-11-01,60.03152307692304
2024-12-01,0.0

# ============================================================
# Funciones auxiliares
# ============================================================

def _ensure_monthly(df, date_col="FECHA"):
    """
    Aseguramos que la serie tenga frecuencia mensual continua entre FECHA min y max.
    Rellenamos huecos con 0. No extrapolamos fuera del rango observado.
    """
    df = df.copy()
    df[date_col] = pd.to_datetime(df[date_col])
    df = df.sort_values(date_col)
    fmin, fmax = df[date_col].min(), df[date_col].max()
    idx = pd.date_range(fmin, fmax, freq="MS")
    df = df.set_index(date_col).reindex(idx)
    df.index.name = date_col
    df = df.reset_index()
    return df

def _iqr_limits(x: pd.Series):
    """
    Calculamos Q1, Q3, IQR y devolvemos límites inferior y superior para IQR.
    """
    Q1 = x.quantile(0.25)
    Q3 = x.quantile(0.75)
    IQR = Q3 - Q1
    lim_inf = Q1 - 1.5 * IQR
    lim_sup = Q3 + 1.5 * IQR
    return float(lim_inf), float(lim_sup)

def _redistribuir_deficit_sin_negativos(vals: np.ndarray, deficit: float) -> np.ndarray:
    """
    Rebajamos 'vals' (>=0) para absorber 'deficit' (>=0) de forma proporcional sin dejar nada < 0.
    Implementamos un bucle con topes (caps) para respetar el mínimo 0.
    """
    s = vals.astype(float).copy()
    rem = float(deficit)
    if rem <= 0:
        return s

    # Iteramos hasta absorber el déficit o quedarnos sin masa positiva
    while rem > 1e-12:
        pos_mask = s > 0
        total_pos = s[pos_mask].sum()
        if total_pos <= 0:
            # No hay masa positiva; salimos
            break

        # Proporción de reducción
        reduccion = np.zeros_like(s)
        reduccion[pos_mask] = rem * (s[pos_mask] / total_pos)

        # Tope: nadie puede quedar por debajo de 0
        cap = s.copy()
        exceso_cap = np.maximum(reduccion - cap, 0.0)
        reduccion_efectiva = reduccion - exceso_cap

        # Aplicamos
        s = s - reduccion_efectiva
        s[s < 0] = 0.0

        # Actualizamos remanente
        rem = float(exceso_cap.sum())
        if rem <= 1e-12:
            rem = 0.0
            break
    return s

# ============================================================
# Transformación de UNA SERIE (FECHA + valor)
# ============================================================

def transformar_serie_outliers(
    df_serie: pd.DataFrame,
    date_col: str = "FECHA",
    value_col: str = "cost_float_mod",  # si nuestra serie trae 'valor', lo indicamos aquí
    asegurar_mensual: bool = True
) -> pd.DataFrame:
    """
    Dado un DataFrame con columnas [FECHA, value_col], generamos:
      - is_outlier (IQR global de la serie)
      - cost_float_mod_2 (ajustada con reglas para outliers +/-)
      - cost_float_mean_y (media anual original si el año tiene outliers; NaN si no)
      - cost_float_mod_3 (original con outliers sustituidos por cost_float_mean_y)
    Devolvemos un DataFrame ordenado por FECHA con YEAR y las nuevas columnas.
    """
    # Copiamos y normalizamos columnas
    st = df_serie[[date_col, value_col]].copy()
    st[date_col] = pd.to_datetime(st[date_col])
    st = st.sort_values(date_col)

    # Aseguramos mensualidad continua entre min y max
    if asegurar_mensual:
        st = _ensure_monthly(st, date_col=date_col)
        # Rellenamos NaN del valor con 0 después de reindexar
        if value_col not in st.columns:
            # si reindexó sin la columna, la añadimos a 0
            st[value_col] = 0.0
        st[value_col] = st[value_col].fillna(0.0)

    # Añadimos YEAR y alineamos tipo float
    st["YEAR"] = st[date_col].dt.year
    st[value_col] = st[value_col].astype(float)

    # Calculamos límites IQR sobre la serie completa
    lim_inf, lim_sup = _iqr_limits(st[value_col])

    # Etiquetamos outliers
    st["is_outlier"] = ((st[value_col] < lim_inf) | (st[value_col] > lim_sup)).astype(int)

    # Inicializamos la serie ajustada con los valores originales
    st["cost_float_mod"] = st[value_col]  # alias explícito
    st["cost_float_mod_2"] = st["cost_float_mod"].astype(float)

    # 1) Tratamos OUTLIERS POSITIVOS: aplanamos al lim_sup y redistribuimos exceso anual
    for y in sorted(st["YEAR"].unique()):
        mask_y = st["YEAR"] == y
        vals_y = st.loc[mask_y, "cost_float_mod_2"].values

        # Exceso sobre el límite superior
        exceso_vec = np.maximum(vals_y - lim_sup, 0.0)
        exceso_total = float(exceso_vec.sum())

        if exceso_total > 0:
            # Aplanamos al límite superior
            st.loc[mask_y, "cost_float_mod_2"] = np.minimum(vals_y, lim_sup)
            # Redistribución uniforme del exceso anual
            n_meses = int(mask_y.sum())
            st.loc[mask_y, "cost_float_mod_2"] += exceso_total / n_meses

    # 2) Tratamos OUTLIERS NEGATIVOS con las dos reglas
    for y in sorted(st["YEAR"].unique()):
        mask_y = st["YEAR"] == y
        # Valores ORIGINALES para detectar outliers negativos
        vals_y_orig = st.loc[mask_y, "cost_float_mod"].values
        # Valores AJUSTADOS tras positivos
        vals_y_adj  = st.loc[mask_y, "cost_float_mod_2"].values

        out_neg_mask = vals_y_orig < lim_inf
        if not np.any(out_neg_mask):
            continue

        suma_anual = float(vals_y_orig.sum())

        if suma_anual < 0:
            # Regla 1: suma anual negativa -> eliminamos outlier (a 0) sin redistribuir
            vals_y_adj[out_neg_mask] = 0.0
            st.loc[mask_y, "cost_float_mod_2"] = vals_y_adj
        else:
            # Regla 2: suma anual positiva -> subimos outliers negativos a 0 y
            # rebajamos meses positivos proporcionalmente sin dejar nada < 0
            deficit = float(np.abs(vals_y_adj[out_neg_mask]).sum())
            vals_y_adj[out_neg_mask] = 0.0
            if deficit > 0:
                vals_y_adj = _redistribuir_deficit_sin_negativos(vals_y_adj, deficit)
            st.loc[mask_y, "cost_float_mod_2"] = vals_y_adj

    # 3) cost_float_mean_y: media anual ORIGINAL en años con al menos un outlier
    st["cost_float_mean_y"] = np.nan
    years_con_outliers = (
        st.groupby("YEAR")["is_outlier"].max().pipe(lambda s: s[s == 1].index.tolist())
    )
    if years_con_outliers:
        media_anual = st.groupby("YEAR")["cost_float_mod"].mean()
        for y in years_con_outliers:
            st.loc[st["YEAR"] == y, "cost_float_mean_y"] = media_anual.loc[y]

    # 4) cost_float_mod_3: original, sustituyendo outliers por la media anual de su año
    st["cost_float_mod_3"] = st["cost_float_mod"]
    m_out = st["is_outlier"] == 1
    st.loc[m_out, "cost_float_mod_3"] = st.loc[m_out, "cost_float_mean_y"]

    # Salida ordenada
    out = st[[date_col, "YEAR", "cost_float_mod", "is_outlier",
              "cost_float_mod_2", "cost_float_mean_y", "cost_float_mod_3"]].copy()
    out = out.sort_values(date_col).reset_index(drop=True)

    # Información útil de control (opcional)
    print(f"Límites IQR -> lim_inf: {lim_inf:.4f} | lim_sup: {lim_sup:.4f}")
    return out

# ============================================================
# Transformación de TODO UN PANEL (por parejas)
# ============================================================

def transformar_panel_outliers(
    df_panel: pd.DataFrame,
    id_cols=("ID_BUILDING","FM_COST_TYPE"),
    date_col="FECHA",
    value_col="cost_float_mod",
    asegurar_mensual=True
) -> pd.DataFrame:
    """
    Aplicamos la transformación a todo el panel agrupando por id_cols.
    Devolvemos un DataFrame con las columnas originales mínimas (ids y FECHA)
    y las nuevas variables generadas por serie.
    """
    # Nos quedamos con columnas mínimas
    cols_min = list(id_cols) + [date_col, value_col]
    missing = set(cols_min) - set(df_panel.columns)
    if missing:
        raise ValueError(f"Faltan columnas en df_panel: {missing}")

    df_panel = df_panel[cols_min].copy()
    df_panel[date_col] = pd.to_datetime(df_panel[date_col])

    # Aplicamos por grupo
    out_list = []
    for keys, g in df_panel.groupby(list(id_cols), sort=False):
        g = g.sort_values(date_col)
        # Transformamos la serie del grupo
        t = transformar_serie_outliers(
            g[[date_col, value_col]],
            date_col=date_col,
            value_col=value_col,
            asegurar_mensual=asegurar_mensual
        )
        # Reinyectamos las claves del grupo
        for i, col in enumerate(id_cols):
            t[col] = keys[i] if isinstance(keys, tuple) else keys
        # Ordenamos columnas
        out_list.append(t)

    out = pd.concat(out_list, ignore_index=True)
    out = out[list(id_cols) + [date_col, "YEAR",
                               "cost_float_mod","is_outlier",
                               "cost_float_mod_2","cost_float_mean_y","cost_float_mod_3"]]
    out = out.sort_values(list(id_cols) + [date_col]).reset_index(drop=True)
    return out

# ============================================================
# Ejemplos de uso
# ============================================================

# 1) Para UNA serie (por ejemplo, s_train que ya tenemos con columnas FECHA y 'valor'):
#    Si nuestra columna se llama 'valor', pasamos value_col='valor'
# serie_enriquecida = transformar_serie_outliers(s_train.rename(columns={"valor":"cost_float_mod"}),
#                                                date_col="FECHA", value_col="cost_float_mod")

# 2) Para TODO el panel agregado mensual (p. ej., df_fd1_v5_ITE1_train):
# panel_enriquecido = transformar_panel_outliers(df_fd1_v5_ITE1_train,
#                                                id_cols=("ID_BUILDING","FM_COST_TYPE"),
#                                                date_col="FECHA",
#                                                value_col="cost_float_mod",
#                                                asegurar_mensual=True)

# ============================================================
# 2) Aplicamos a la serie TRAIN pegada
# ============================================================

# En nuestro caso, la columna se llama 'valor' -> la mapeamos a 'cost_float_mod'
serie_train_enriquecida = transformar_serie_outliers(
    s_train.rename(columns={"valor": "cost_float_mod"}),
    date_col="FECHA",
    value_col="cost_float_mod",
    asegurar_mensual=True
)

print("Resumen serie enriquecida (primeros 12 meses):")
print(serie_train_enriquecida.head(12))

# ============================================================
# 3) Diagnóstico de cambio estructural vs 2024
# ============================================================

# Calculamos la media de 2024 (real)
media_2024 = s_real["valor"].mean()

# Función auxiliar para evaluar y mostrar diagnóstico
def diagnostico_cambio(nombre, serie_train, media_test, umbral=0.30):
    media_train = serie_train.mean()
    ratio = abs(media_test - media_train) / (media_train if media_train != 0 else 1.0)
    print(f"\n== Diagnóstico {nombre} ==")
    print(f"Media 2021–2023: {media_train:.2f}")
    print(f"Media 2024     : {media_test:.2f}")
    print(f"Cambio relativo: {ratio:.2%}")
    if ratio > umbral:
        print("Conclusión: detectamos cambio estructural (caída/subida de nivel relevante).")
    else:
        print("Conclusión: no parece haber cambio estructural relevante en medias.")

# Aplicamos a las tres variantes del conjunto train enriquecido
diagnostico_cambio("Original (cost_float_mod)",
                   serie_train_enriquecida["cost_float_mod"], media_2024)

diagnostico_cambio("Ajustada (cost_float_mod_2)",
                   serie_train_enriquecida["cost_float_mod_2"], media_2024)

diagnostico_cambio("Outliers sustituidos (cost_float_mod_3)",
                   serie_train_enriquecida["cost_float_mod_3"], media_2024)

Límites IQR -> lim_inf: -456.6198 | lim_sup: 966.0293
Resumen serie enriquecida (primeros 12 meses):
        FECHA  YEAR  cost_float_mod  is_outlier  cost_float_mod_2  \
0  2021-01-01  2021      349.341938           0        378.340510
1  2021-02-01  2021     1314.012197           1        995.027906
2  2021-03-01  2021      268.410015           0        297.408587
3  2021-04-01  2021      402.122446           0        431.121018
4  2021-05-01  2021      728.715262           0        757.713833
5  2021-06-01  2021        0.000000           0         28.998572
6  2021-07-01  2021      695.626677           0        724.625249
7  2021-08-01  2021       53.729662           0         82.728233
8  2021-09-01  2021      453.016815           0        482.015387
9  2021-10-01  2021       25.100000           0         54.098572
10 2021-11-01  2021       36.707692           0         65.706264
11 2021-12-01  2021      865.777308           0        894.775880

    cost_float_mean_y  cost_float_mod_3
0          432.713334        349.341938
1          432.713334        432.713334
2          432.713334        268.410015
3          432.713334        402.122446
4          432.713334        728.715262
5          432.713334          0.000000
6          432.713334        695.626677
7          432.713334         53.729662
8          432.713334        453.016815
9          432.713334         25.100000
10         432.713334         36.707692
11         432.713334        865.777308

== Diagnóstico Original (cost_float_mod) ==
Media 2021–2023: 358.93
Media 2024     : 140.85
Cambio relativo: 60.76%
Conclusión: detectamos cambio estructural (caída/subida de nivel relevante).

== Diagnóstico Ajustada (cost_float_mod_2) ==
Media 2021–2023: 358.93
Media 2024     : 140.85
Cambio relativo: 60.76%
Conclusión: detectamos cambio estructural (caída/subida de nivel relevante).

== Diagnóstico Outliers sustituidos (cost_float_mod_3) ==
Media 2021–2023: 285.33
Media 2024     : 140.85
Cambio relativo: 50.64%
Conclusión: detectamos cambio estructural (caída/subida de nivel relevante).

# Graficamos la serie original y las variantes junto con los valores reales de 2024
plt.figure(figsize=(12,6))

# Original
plt.plot(serie_train_enriquecida["FECHA"],
         serie_train_enriquecida["cost_float_mod"],
         label="Train original (cost_float_mod)", color="blue")

# Ajustada
plt.plot(serie_train_enriquecida["FECHA"],
         serie_train_enriquecida["cost_float_mod_2"],
         label="Train ajustada (cost_float_mod_2)", color="green")

# Outliers sustituidos
plt.plot(serie_train_enriquecida["FECHA"],
         serie_train_enriquecida["cost_float_mod_3"],
         label="Train outliers→media (cost_float_mod_3)", color="orange")

# Serie real 2024
plt.plot(s_real["FECHA"],
         s_real["valor"],
         label="Real 2024", color="red", linestyle="--")

plt.title("Comparativa serie train (2021–2023) vs real 2024\nID_BUILDING 1000480 | FM_COST_TYPE Mtto. Correctivo")
plt.xlabel("Fecha")
plt.ylabel("Coste")
plt.legend()
plt.grid(True)
plt.tight_layout()
plt.show()

# --- 1) Preparamos la serie real 2024 ---
serie_real = s_real.set_index("FECHA")["valor"]

# --- 2) Definimos una función auxiliar para entrenar y evaluar SES ---
def evaluar_ses(serie_train, serie_real, nombre=""):
    # Ajustamos SES
    modelo = SimpleExpSmoothing(serie_train).fit(optimized=True)
    pred = modelo.forecast(len(serie_real))
    pred.index = serie_real.index

    # Métricas
    mae = mean_absolute_error(serie_real, pred)
    wape = np.sum(np.abs(serie_real - pred)) / np.sum(np.abs(serie_real))
    smape = (100/len(serie_real)) * np.sum(
        2 * np.abs(pred - serie_real) / (np.abs(serie_real) + np.abs(pred))
    )
    # MASE con naïve
    naive1 = serie_train.shift(1).dropna()
    mae_naive1 = np.mean(np.abs(serie_train[1:] - naive1))
    mase = mae / mae_naive1

    # Precisión anual
    suma_pred = pred.sum()
    suma_real = serie_real.sum()
    error_abs = abs(suma_pred - suma_real)
    error_rel = error_abs / abs(suma_real) if suma_real != 0 else np.nan

    print(f"\n=== RESULTADOS SES ({nombre}) ===")
    print(f"MAE   : {mae:.2f}")
    print(f"WAPE  : {wape:.2%}")
    print(f"SMAPE : {smape:.2f}%")
    print(f"MASE  : {mase:.3f}")
    print(f"Suma predicha: {suma_pred:.2f} | Suma real: {suma_real:.2f}")
    print(f"Error anual absoluto: {error_abs:.2f}")
    print(f"Error anual relativo: {error_rel:.2%}")

    return pred


# --- 3) Evaluamos con cada variable de coste ---
serie_train_mod   = serie_train_enriquecida.set_index("FECHA")["cost_float_mod"]
serie_train_mod2  = serie_train_enriquecida.set_index("FECHA")["cost_float_mod_2"]
serie_train_mod3  = serie_train_enriquecida.set_index("FECHA")["cost_float_mod_3"]

pred_mod  = evaluar_ses(serie_train_mod,  serie_real, nombre="Original (mod)")
pred_mod2 = evaluar_ses(serie_train_mod2, serie_real, nombre="Ajustada (mod_2)")
pred_mod3 = evaluar_ses(serie_train_mod3, serie_real, nombre="Outliers→Media (mod_3)")

# --- 4) Visualizamos comparaciones ---
import matplotlib.pyplot as plt

plt.figure(figsize=(10,5))
plt.plot(serie_train_mod.index, serie_train_mod.values, label="Entrenamiento (mod)", color="blue")
plt.plot(serie_real.index, serie_real.values, label="Real 2024", color="black")
plt.plot(pred_mod.index, pred_mod.values, label="Pred mod", color="red", linestyle="--")
plt.plot(pred_mod2.index, pred_mod2.values, label="Pred mod_2", color="green", linestyle="--")
plt.plot(pred_mod3.index, pred_mod3.values, label="Pred mod_3", color="orange", linestyle="--")
plt.title("Comparación predicciones SES con 3 transformaciones de costes")
plt.xlabel("Fecha"); plt.ylabel("Coste mensual (€)")
plt.legend()
plt.tight_layout()
plt.show()

/usr/local/lib/python3.12/dist-packages/statsmodels/tsa/base/tsa_model.py:473: ValueWarning: No frequency information was provided, so inferred frequency MS will be used.
  self._init_dates(dates, freq)
/usr/local/lib/python3.12/dist-packages/statsmodels/tsa/base/tsa_model.py:473: ValueWarning: No frequency information was provided, so inferred frequency MS will be used.
  self._init_dates(dates, freq)
/usr/local/lib/python3.12/dist-packages/statsmodels/tsa/base/tsa_model.py:473: ValueWarning: No frequency information was provided, so inferred frequency MS will be used.
  self._init_dates(dates, freq)

=== RESULTADOS SES (Original (mod)) ===
MAE   : 230.48
WAPE  : 163.63%
SMAPE : 109.14%
MASE  : 0.525
Suma predicha: 4192.10 | Suma real: 1690.25
Error anual absoluto: 2501.86
Error anual relativo: 148.02%

=== RESULTADOS SES (Ajustada (mod_2)) ===
MAE   : 254.64
WAPE  : 180.78%
SMAPE : 112.65%
MASE  : 0.727
Suma predicha: 4540.09 | Suma real: 1690.25
Error anual absoluto: 2849.84
Error anual relativo: 168.60%

=== RESULTADOS SES (Outliers→Media (mod_3)) ===
MAE   : 188.59
WAPE  : 133.89%
SMAPE : 102.92%
MASE  : 0.655
Suma predicha: 3459.83 | Suma real: 1690.25
Error anual absoluto: 1769.58
Error anual relativo: 104.69%

# Creamos versión enriquecida del TRAIN si no existen columnas
if not {"cost_float_mod_2","cost_float_mod_3"}.issubset(df_fd1_v5_ITE1_train.columns):
    print("Generamos columnas mod_2 y mod_3 en TRAIN...")
    df_train_enr = transformar_panel_outliers(df_fd1_v5_ITE1_train,
                                              id_cols=("ID_BUILDING","FM_COST_TYPE"),
                                              date_col="FECHA",
                                              value_col="cost_float_mod",
                                              asegurar_mensual=True)
else:
    df_train_enr = df_fd1_v5_ITE1_train.copy()

Generamos columnas mod_2 y mod_3 en TRAIN...
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -41.9625 | lim_sup: 69.9375
Límites IQR -> lim_inf: -379.8725 | lim_sup: 2816.6275
Límites IQR -> lim_inf: -876.8238 | lim_sup: 1632.3063
Límites IQR -> lim_inf: -378.7650 | lim_sup: 631.2750
Límites IQR -> lim_inf: 5657.1087 | lim_sup: 6909.1587
Límites IQR -> lim_inf: -220.5225 | lim_sup: 367.5375
Límites IQR -> lim_inf: -2230.7325 | lim_sup: 23705.3475
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -356.4937 | lim_sup: 594.1562
Límites IQR -> lim_inf: -594.8237 | lim_sup: 3707.6663
Límites IQR -> lim_inf: -928.1637 | lim_sup: 3528.2463
Límites IQR -> lim_inf: -627.0000 | lim_sup: 1045.0000
Límites IQR -> lim_inf: -3514.8425 | lim_sup: 48546.1775
Límites IQR -> lim_inf: -287.1112 | lim_sup: 656.3587
Límites IQR -> lim_inf: -1677.3825 | lim_sup: 18622.1775
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -67.5000 | lim_sup: 112.5000
Límites IQR -> lim_inf: -754.1700 | lim_sup: 5000.4700
Límites IQR -> lim_inf: -930.0275 | lim_sup: 1741.0925
Límites IQR -> lim_inf: -7137.7688 | lim_sup: 12088.4213
Límites IQR -> lim_inf: -1511.4300 | lim_sup: 14135.4100
Límites IQR -> lim_inf: -448.7275 | lim_sup: 848.7125
Límites IQR -> lim_inf: 52.8275 | lim_sup: 10907.8075
Límites IQR -> lim_inf: 218.7800 | lim_sup: 218.7800
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -393.7500 | lim_sup: 656.2500
Límites IQR -> lim_inf: -61.2000 | lim_sup: 102.0000
Límites IQR -> lim_inf: -301.2600 | lim_sup: 502.1000
Límites IQR -> lim_inf: -140.0000 | lim_sup: 380.0000
Límites IQR -> lim_inf: -1803.1650 | lim_sup: 3005.2750
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -380.8425 | lim_sup: 634.7375
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -1232.8050 | lim_sup: 2054.6750
Límites IQR -> lim_inf: -515.7525 | lim_sup: 859.5875
Límites IQR -> lim_inf: -3355.9925 | lim_sup: 6157.4875
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 267.9700 | lim_sup: 267.9700
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -910.2000 | lim_sup: 1517.0000
Límites IQR -> lim_inf: 4350.0000 | lim_sup: 4350.0000
Límites IQR -> lim_inf: 683.3637 | lim_sup: 1135.7538
Límites IQR -> lim_inf: -1346.1637 | lim_sup: 2243.6062
Límites IQR -> lim_inf: -686.0088 | lim_sup: 3411.4813
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -302.5250 | lim_sup: 865.9150
Límites IQR -> lim_inf: -634.7625 | lim_sup: 1535.4575
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 2915.2050 | lim_sup: 4882.6850
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -1380.3787 | lim_sup: 7470.6512
Límites IQR -> lim_inf: -140.6250 | lim_sup: 234.3750
Límites IQR -> lim_inf: 165.3000 | lim_sup: 224.1000
Límites IQR -> lim_inf: -1068.5925 | lim_sup: 2577.4275
Límites IQR -> lim_inf: 2897.3450 | lim_sup: 7979.3050
Límites IQR -> lim_inf: -259.5000 | lim_sup: 432.5000
Límites IQR -> lim_inf: -2755.9150 | lim_sup: 12519.8850
Límites IQR -> lim_inf: -322.5000 | lim_sup: 537.5000
Límites IQR -> lim_inf: -39.7450 | lim_sup: 367.5750
Límites IQR -> lim_inf: -688.9863 | lim_sup: 1608.4038
Límites IQR -> lim_inf: 294.7500 | lim_sup: 294.7500
Límites IQR -> lim_inf: 1466.3050 | lim_sup: 3399.2250
Límites IQR -> lim_inf: -325.5000 | lim_sup: 542.5000
Límites IQR -> lim_inf: -1355.3275 | lim_sup: 7331.5125
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -298.7350 | lim_sup: 799.2250
Límites IQR -> lim_inf: -1230.4925 | lim_sup: 2269.7075
Límites IQR -> lim_inf: 914.8950 | lim_sup: 2662.8550
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -939.2462 | lim_sup: 5479.2237
Límites IQR -> lim_inf: -285.0000 | lim_sup: 475.0000
Límites IQR -> lim_inf: -701.6488 | lim_sup: 1538.1613
Límites IQR -> lim_inf: -853.4075 | lim_sup: 2636.0725
Límites IQR -> lim_inf: 2622.3300 | lim_sup: 4943.0500
Límites IQR -> lim_inf: -483.4800 | lim_sup: 805.8000
Límites IQR -> lim_inf: -1770.5225 | lim_sup: 11209.4375
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 48.6250 | lim_sup: 454.0250
Límites IQR -> lim_inf: -692.5350 | lim_sup: 1742.0250
Límites IQR -> lim_inf: 2604.4050 | lim_sup: 4919.0850
Límites IQR -> lim_inf: -703.8300 | lim_sup: 1173.0500
Límites IQR -> lim_inf: -1008.6700 | lim_sup: 9450.7300
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -400.0875 | lim_sup: 745.3525
Límites IQR -> lim_inf: -784.2725 | lim_sup: 1877.4275
Límites IQR -> lim_inf: 250.0000 | lim_sup: 250.0000
Límites IQR -> lim_inf: 1457.1650 | lim_sup: 2714.6450
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -499.2138 | lim_sup: 6577.9563
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 68.4763 | lim_sup: 296.3862
Límites IQR -> lim_inf: -703.5688 | lim_sup: 1302.3613
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 1485.7375 | lim_sup: 2767.2775
Límites IQR -> lim_inf: 360.0000 | lim_sup: 360.0000
Límites IQR -> lim_inf: -682.7875 | lim_sup: 3893.2525
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -216.2075 | lim_sup: 601.0125
Límites IQR -> lim_inf: -1064.6600 | lim_sup: 2703.9600
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 2839.7450 | lim_sup: 5509.4650
Límites IQR -> lim_inf: -98.6362 | lim_sup: 164.3937
Límites IQR -> lim_inf: -4045.5300 | lim_sup: 11732.5900
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -506.8950 | lim_sup: 938.8650
Límites IQR -> lim_inf: -543.7063 | lim_sup: 1149.6037
Límites IQR -> lim_inf: 250.0000 | lim_sup: 250.0000
Límites IQR -> lim_inf: 1460.5188 | lim_sup: 2700.4087
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -230.6900 | lim_sup: 4303.6500
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 502.6350 | lim_sup: 774.2750
Límites IQR -> lim_inf: -1447.8250 | lim_sup: 2500.1750
Límites IQR -> lim_inf: -139.6050 | lim_sup: 232.6750
Límites IQR -> lim_inf: 101.0000 | lim_sup: 101.0000
Límites IQR -> lim_inf: -827.2375 | lim_sup: 9166.1625
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 257.7450 | lim_sup: 290.1050
Límites IQR -> lim_inf: -1035.0150 | lim_sup: 2589.5850
Límites IQR -> lim_inf: 165.0000 | lim_sup: 165.0000
Límites IQR -> lim_inf: -1056.4650 | lim_sup: 1760.7750
Límites IQR -> lim_inf: -2657.9825 | lim_sup: 10456.9375
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -183.5625 | lim_sup: 305.9375
Límites IQR -> lim_inf: -75.0000 | lim_sup: 125.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -396.8862 | lim_sup: 2326.6437
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 1560.0000 | lim_sup: 3800.0000
Límites IQR -> lim_inf: -708.0100 | lim_sup: 1575.2300
Límites IQR -> lim_inf: -97.9200 | lim_sup: 163.2000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -3436.4763 | lim_sup: 12803.4938
Límites IQR -> lim_inf: 254.6100 | lim_sup: 254.6100
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -156.6250 | lim_sup: 1486.0550
Límites IQR -> lim_inf: -922.7700 | lim_sup: 3113.5900
Límites IQR -> lim_inf: -2483.8275 | lim_sup: 4139.7125
Límites IQR -> lim_inf: 4072.4275 | lim_sup: 5568.6875
Límites IQR -> lim_inf: -201.3750 | lim_sup: 335.6250
Límites IQR -> lim_inf: -2274.8512 | lim_sup: 14995.8587
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -276.2250 | lim_sup: 1366.4550
Límites IQR -> lim_inf: -1078.9388 | lim_sup: 2852.0312
Límites IQR -> lim_inf: 2272.9337 | lim_sup: 4303.0838
Límites IQR -> lim_inf: -175.4062 | lim_sup: 292.3438
Límites IQR -> lim_inf: -3455.7387 | lim_sup: 15252.5712
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -731.9462 | lim_sup: 2017.7238
Límites IQR -> lim_inf: -1183.9475 | lim_sup: 2862.1525
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 3127.7600 | lim_sup: 4067.8400
Límites IQR -> lim_inf: -116.4375 | lim_sup: 194.0625
Límites IQR -> lim_inf: -3657.9050 | lim_sup: 15227.6350
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 190.9100 | lim_sup: 190.9100
Límites IQR -> lim_inf: -269.6063 | lim_sup: 734.9838
Límites IQR -> lim_inf: 467.6650 | lim_sup: 5800.8250
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -4000.5525 | lim_sup: 11137.1075
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -1086.6075 | lim_sup: 7493.8125
Límites IQR -> lim_inf: -45.0000 | lim_sup: 75.0000
Límites IQR -> lim_inf: 384.0000 | lim_sup: 384.0000
Límites IQR -> lim_inf: -557.3100 | lim_sup: 977.2700
Límites IQR -> lim_inf: 798.3600 | lim_sup: 2377.8000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -690.2550 | lim_sup: 9532.3450
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 384.0000 | lim_sup: 384.0000
Límites IQR -> lim_inf: -145.1100 | lim_sup: 241.8500
Límites IQR -> lim_inf: 947.4800 | lim_sup: 2119.5600
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -1903.1737 | lim_sup: 9158.9562
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 120.0000 | lim_sup: 120.0000
Límites IQR -> lim_inf: -982.7175 | lim_sup: 2256.2225
Límites IQR -> lim_inf: 654.2525 | lim_sup: 3516.1925
Límites IQR -> lim_inf: -414.7425 | lim_sup: 691.2375
Límites IQR -> lim_inf: -1143.3300 | lim_sup: 8137.9500
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 120.0000 | lim_sup: 120.0000
Límites IQR -> lim_inf: -411.6937 | lim_sup: 686.1562
Límites IQR -> lim_inf: 709.9488 | lim_sup: 1640.2187
Límites IQR -> lim_inf: -63.6000 | lim_sup: 106.0000
Límites IQR -> lim_inf: -646.8938 | lim_sup: 4234.3963
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -459.2288 | lim_sup: 1786.9012
Límites IQR -> lim_inf: -708.0437 | lim_sup: 1646.5262
Límites IQR -> lim_inf: 1345.2800 | lim_sup: 1345.2800
Límites IQR -> lim_inf: 1230.0600 | lim_sup: 4694.3800
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -1201.6600 | lim_sup: 9330.9000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -289.6675 | lim_sup: 1761.7125
Límites IQR -> lim_inf: -808.2862 | lim_sup: 2520.2837
Límites IQR -> lim_inf: 3566.3400 | lim_sup: 5867.7000
Límites IQR -> lim_inf: -55.5000 | lim_sup: 92.5000
Límites IQR -> lim_inf: -1928.6787 | lim_sup: 11553.4312
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 565.5800 | lim_sup: 565.5800
Límites IQR -> lim_inf: -471.3825 | lim_sup: 980.1375
Límites IQR -> lim_inf: 1595.2137 | lim_sup: 3395.4637
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -1462.4262 | lim_sup: 7508.9037
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -338.1225 | lim_sup: 563.5375
Límites IQR -> lim_inf: 293.7750 | lim_sup: 330.6950
Límites IQR -> lim_inf: -34.9500 | lim_sup: 316.9700
Límites IQR -> lim_inf: -227.8025 | lim_sup: 1070.7375
Límites IQR -> lim_inf: 180.0000 | lim_sup: 180.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -211.4250 | lim_sup: 352.3750
Límites IQR -> lim_inf: 276.4200 | lim_sup: 311.1400
Límites IQR -> lim_inf: -166.8000 | lim_sup: 278.0000
Límites IQR -> lim_inf: -226.7063 | lim_sup: 993.3838
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 326.0700 | lim_sup: 326.0700
Límites IQR -> lim_inf: -626.7750 | lim_sup: 1148.6250
Límites IQR -> lim_inf: -1785.0888 | lim_sup: 11668.3013
Límites IQR -> lim_inf: 2800.0400 | lim_sup: 3916.0400
Límites IQR -> lim_inf: -236.6250 | lim_sup: 394.3750
Límites IQR -> lim_inf: -231.9625 | lim_sup: 3304.2375
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -677.2463 | lim_sup: 1128.7438
Límites IQR -> lim_inf: 323.7850 | lim_sup: 364.4650
Límites IQR -> lim_inf: -502.0575 | lim_sup: 836.7625
Límites IQR -> lim_inf: -96.8550 | lim_sup: 904.1050
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -696.3787 | lim_sup: 1352.2313
Límites IQR -> lim_inf: 450.0000 | lim_sup: 450.0000
Límites IQR -> lim_inf: 661.2250 | lim_sup: 883.0250
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -600.9975 | lim_sup: 2440.6825
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 282.9950 | lim_sup: 378.2350
Límites IQR -> lim_inf: -420.0000 | lim_sup: 700.0000
Límites IQR -> lim_inf: -322.1038 | lim_sup: 1501.6263
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -88.2375 | lim_sup: 147.0625
Límites IQR -> lim_inf: 53.9250 | lim_sup: 60.6850
Límites IQR -> lim_inf: 20.6025 | lim_sup: 185.4225
Límites IQR -> lim_inf: -55.0475 | lim_sup: 412.3725
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -159.0000 | lim_sup: 265.0000
Límites IQR -> lim_inf: -600.7500 | lim_sup: 1001.2500
Límites IQR -> lim_inf: 431.0700 | lim_sup: 485.2300
Límites IQR -> lim_inf: -615.2400 | lim_sup: 1457.4000
Límites IQR -> lim_inf: -119.6812 | lim_sup: 1053.2887
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -387.0000 | lim_sup: 645.0000
Límites IQR -> lim_inf: 282.6200 | lim_sup: 377.4200
Límites IQR -> lim_inf: -330.0000 | lim_sup: 550.0000
Límites IQR -> lim_inf: -389.9350 | lim_sup: 1539.8650
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -461.8838 | lim_sup: 769.8063
Límites IQR -> lim_inf: -1520.6925 | lim_sup: 3177.6875
Límites IQR -> lim_inf: 291.6850 | lim_sup: 389.5650
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -213.2588 | lim_sup: 1376.7313
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -333.3525 | lim_sup: 555.5875
Límites IQR -> lim_inf: 678.0000 | lim_sup: 678.0000
Límites IQR -> lim_inf: 307.6100 | lim_sup: 410.8100
Límites IQR -> lim_inf: -420.0000 | lim_sup: 700.0000
Límites IQR -> lim_inf: -63.0100 | lim_sup: 1189.0100
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -1369.3788 | lim_sup: 2493.5513
Límites IQR -> lim_inf: 813.1550 | lim_sup: 1693.7550
Límites IQR -> lim_inf: -202.5000 | lim_sup: 337.5000
Límites IQR -> lim_inf: -334.2188 | lim_sup: 1492.8913
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -376.9350 | lim_sup: 628.2250
Límites IQR -> lim_inf: -2833.3700 | lim_sup: 8500.1100
Límites IQR -> lim_inf: 164.6900 | lim_sup: 219.9700
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -123.9825 | lim_sup: 786.9575
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -226.8000 | lim_sup: 378.0000
Límites IQR -> lim_inf: 320.1650 | lim_sup: 427.5650
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -304.3800 | lim_sup: 987.1600
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -156.6000 | lim_sup: 261.0000
Límites IQR -> lim_inf: -2065.5500 | lim_sup: 10120.4500
Límites IQR -> lim_inf: 268.4050 | lim_sup: 302.1250
Límites IQR -> lim_inf: -165.0000 | lim_sup: 355.0000
Límites IQR -> lim_inf: -372.8038 | lim_sup: 745.1862
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 89.6200 | lim_sup: 119.7000
Límites IQR -> lim_inf: -6.2862 | lim_sup: 85.9837
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -105.9000 | lim_sup: 176.5000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 68.0000 | lim_sup: 68.0000
Límites IQR -> lim_inf: -162.9500 | lim_sup: 1635.8700
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -317.4000 | lim_sup: 529.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -194.2687 | lim_sup: 2093.8212
Límites IQR -> lim_inf: -270.0000 | lim_sup: 450.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -263.8500 | lim_sup: 439.7500
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -905.5350 | lim_sup: 2747.3850
Límites IQR -> lim_inf: -135.0000 | lim_sup: 225.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -253.8375 | lim_sup: 423.0625
Límites IQR -> lim_inf: 650.0000 | lim_sup: 650.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -802.7500 | lim_sup: 1381.2500
Límites IQR -> lim_inf: -123.4525 | lim_sup: 2331.3675
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -113.1187 | lim_sup: 188.5312
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -351.6412 | lim_sup: 1699.8487
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -207.7500 | lim_sup: 346.2500
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -204.7925 | lim_sup: 504.6275
Límites IQR -> lim_inf: -718.6987 | lim_sup: 2729.5112
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -284.4937 | lim_sup: 521.8362
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -46.4600 | lim_sup: 168.1000
Límites IQR -> lim_inf: -254.6175 | lim_sup: 2052.6225
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -311.1000 | lim_sup: 518.5000
Límites IQR -> lim_inf: 110.0000 | lim_sup: 110.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -501.9112 | lim_sup: 2819.1988
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -398.0888 | lim_sup: 663.4813
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 65.0000 | lim_sup: 65.0000
Límites IQR -> lim_inf: -393.2337 | lim_sup: 2369.3763
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -309.0000 | lim_sup: 515.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 65.0000 | lim_sup: 65.0000
Límites IQR -> lim_inf: -1167.2550 | lim_sup: 3330.9250
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -522.3750 | lim_sup: 870.6250
Límites IQR -> lim_inf: 43.7500 | lim_sup: 43.7500
Límites IQR -> lim_inf: 110.0000 | lim_sup: 110.0000
Límites IQR -> lim_inf: -513.8238 | lim_sup: 2589.5463
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -621.2025 | lim_sup: 1284.5975
Límites IQR -> lim_inf: 110.0000 | lim_sup: 110.0000
Límites IQR -> lim_inf: 68.0000 | lim_sup: 68.0000
Límites IQR -> lim_inf: -405.8225 | lim_sup: 3253.6175
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -218.9625 | lim_sup: 364.9375
Límites IQR -> lim_inf: 110.0000 | lim_sup: 110.0000
Límites IQR -> lim_inf: -257.2662 | lim_sup: 1885.3837
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -277.4287 | lim_sup: 462.3812
Límites IQR -> lim_inf: 110.0000 | lim_sup: 110.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -265.7537 | lim_sup: 2623.0563
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -445.9425 | lim_sup: 743.2375
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -10.3475 | lim_sup: 1368.5725
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -402.0000 | lim_sup: 670.0000
Límites IQR -> lim_inf: 110.0000 | lim_sup: 110.0000
Límites IQR -> lim_inf: -504.0213 | lim_sup: 1884.6688
Límites IQR -> lim_inf: -41.5688 | lim_sup: 69.2812
Límites IQR -> lim_inf: -135.0000 | lim_sup: 225.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -154.3875 | lim_sup: 1532.1325
Límites IQR -> lim_inf: -41.5688 | lim_sup: 69.2812
Límites IQR -> lim_inf: -115.3800 | lim_sup: 192.3000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -110.6637 | lim_sup: 1935.4262
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -407.7075 | lim_sup: 679.5125
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 68.0000 | lim_sup: 68.0000
Límites IQR -> lim_inf: -219.6225 | lim_sup: 2161.5575
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -134.7375 | lim_sup: 224.5625
Límites IQR -> lim_inf: -448.8850 | lim_sup: 830.2750
Límites IQR -> lim_inf: -316.7700 | lim_sup: 527.9500
Límites IQR -> lim_inf: 165.0000 | lim_sup: 165.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -1455.5550 | lim_sup: 4580.1050
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -120.1200 | lim_sup: 200.2000
Límites IQR -> lim_inf: -277.7550 | lim_sup: 462.9250
Límites IQR -> lim_inf: 110.0000 | lim_sup: 110.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -421.3262 | lim_sup: 2452.9637
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -277.1400 | lim_sup: 461.9000
Límites IQR -> lim_inf: 110.0000 | lim_sup: 110.0000
Límites IQR -> lim_inf: -58.7500 | lim_sup: 271.2500
Límites IQR -> lim_inf: -305.1225 | lim_sup: 2079.1175
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -422.3100 | lim_sup: 703.8500
Límites IQR -> lim_inf: 110.0000 | lim_sup: 110.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 15.8300 | lim_sup: 1913.5300
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -317.5500 | lim_sup: 529.2500
Límites IQR -> lim_inf: -96.0000 | lim_sup: 160.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -54.9125 | lim_sup: 1795.3075
Límites IQR -> lim_inf: -481.3125 | lim_sup: 802.1875
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -394.2900 | lim_sup: 657.1500
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -543.7500 | lim_sup: 2567.4300
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -57.7125 | lim_sup: 96.1875
Límites IQR -> lim_inf: -403.6500 | lim_sup: 672.7500
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 110.0000 | lim_sup: 110.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -786.4062 | lim_sup: 2792.2238
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -57.7125 | lim_sup: 96.1875
Límites IQR -> lim_inf: -71.2463 | lim_sup: 118.7438
Límites IQR -> lim_inf: 704.4100 | lim_sup: 704.4100
Límites IQR -> lim_inf: 55.0000 | lim_sup: 55.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -597.8275 | lim_sup: 2081.1725
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -424.5000 | lim_sup: 835.5000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -304.2500 | lim_sup: 593.7500
Límites IQR -> lim_inf: -654.7275 | lim_sup: 2517.9525
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -360.0750 | lim_sup: 682.1250
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -97.5000 | lim_sup: 162.5000
Límites IQR -> lim_inf: -522.6288 | lim_sup: 2367.1213
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -418.0500 | lim_sup: 696.7500
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -160.9200 | lim_sup: 268.2000
Límites IQR -> lim_inf: -328.1275 | lim_sup: 2081.9925
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -378.9075 | lim_sup: 631.5125
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -503.5825 | lim_sup: 2976.8375
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -261.6000 | lim_sup: 436.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -407.2225 | lim_sup: 1996.2775
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -336.0000 | lim_sup: 560.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -371.0150 | lim_sup: 1867.0650
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -575.5800 | lim_sup: 959.3000
Límites IQR -> lim_inf: 110.0000 | lim_sup: 110.0000
Límites IQR -> lim_inf: -390.6587 | lim_sup: 3360.1312
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -41.5688 | lim_sup: 69.2812
Límites IQR -> lim_inf: -103.0050 | lim_sup: 171.6750
Límites IQR -> lim_inf: -375.0000 | lim_sup: 625.0000
Límites IQR -> lim_inf: 98.9300 | lim_sup: 132.2100
Límites IQR -> lim_inf: 35.0000 | lim_sup: 35.0000
Límites IQR -> lim_inf: -74.7625 | lim_sup: 2500.5575
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -323.8350 | lim_sup: 539.7250
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -406.7187 | lim_sup: 2227.2512
Límites IQR -> lim_inf: 1112.0000 | lim_sup: 1112.0000
Límites IQR -> lim_inf: -200.2200 | lim_sup: 333.7000
Límites IQR -> lim_inf: -186.0975 | lim_sup: 310.1625
Límites IQR -> lim_inf: -1282.3675 | lim_sup: 2910.6125
Límites IQR -> lim_inf: 361.1850 | lim_sup: 482.3450
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -21.3800 | lim_sup: 403.1000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -489.0000 | lim_sup: 815.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -165.2963 | lim_sup: 1854.5538
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 225.0000 | lim_sup: 225.0000
Límites IQR -> lim_inf: -663.4000 | lim_sup: 1415.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -398.7600 | lim_sup: 664.6000
Límites IQR -> lim_inf: 110.0000 | lim_sup: 110.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -106.8938 | lim_sup: 1707.1363
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -326.0362 | lim_sup: 574.6337
Límites IQR -> lim_inf: 110.0000 | lim_sup: 110.0000
Límites IQR -> lim_inf: -231.6675 | lim_sup: 2179.6725
Límites IQR -> lim_inf: -90.7500 | lim_sup: 151.2500
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -401.8875 | lim_sup: 669.8125
Límites IQR -> lim_inf: 110.0000 | lim_sup: 110.0000
Límites IQR -> lim_inf: -490.0850 | lim_sup: 2468.5150
Límites IQR -> lim_inf: -90.7500 | lim_sup: 151.2500
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -253.3625 | lim_sup: 504.9375
Límites IQR -> lim_inf: 110.0000 | lim_sup: 110.0000
Límites IQR -> lim_inf: -804.8450 | lim_sup: 3200.7350
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -470.2988 | lim_sup: 783.8313
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -139.2663 | lim_sup: 1974.6638
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -290.5088 | lim_sup: 484.1813
Límites IQR -> lim_inf: 110.0000 | lim_sup: 110.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -217.3800 | lim_sup: 1819.3000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -171.7500 | lim_sup: 286.2500
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -57.9763 | lim_sup: 1875.6337
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -438.8550 | lim_sup: 731.4250
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 715.0000 | lim_sup: 715.0000
Límites IQR -> lim_inf: -384.5400 | lim_sup: 2583.2000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -147.2250 | lim_sup: 245.3750
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 116.1200 | lim_sup: 116.1200
Límites IQR -> lim_inf: -170.2088 | lim_sup: 1872.6413
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -244.9650 | lim_sup: 408.2750
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -25.5000 | lim_sup: 42.5000
Límites IQR -> lim_inf: -185.3225 | lim_sup: 2029.8775
Límites IQR -> lim_inf: -217.5000 | lim_sup: 362.5000
Límites IQR -> lim_inf: -1745.6400 | lim_sup: 2909.4000
Límites IQR -> lim_inf: -648.5825 | lim_sup: 1682.1975
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 1985.5800 | lim_sup: 4568.7800
Límites IQR -> lim_inf: -76.5000 | lim_sup: 127.5000
Límites IQR -> lim_inf: -1950.3713 | lim_sup: 8340.7388
Límites IQR -> lim_inf: -150.0000 | lim_sup: 250.0000
Límites IQR -> lim_inf: -150.1200 | lim_sup: 1139.0800
Límites IQR -> lim_inf: -484.8188 | lim_sup: 1827.6713
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 4402.8200 | lim_sup: 7047.5400
Límites IQR -> lim_inf: -217.8750 | lim_sup: 363.1250
Límites IQR -> lim_inf: -4084.8625 | lim_sup: 21461.5575
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -189.1200 | lim_sup: 315.2000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -594.4575 | lim_sup: 2971.7425
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -328.8750 | lim_sup: 548.1250
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 165.0000 | lim_sup: 165.0000
Límites IQR -> lim_inf: -493.0750 | lim_sup: 2582.6650
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -251.0625 | lim_sup: 418.4375
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -309.3663 | lim_sup: 2965.3238
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 90.0000 | lim_sup: 90.0000
Límites IQR -> lim_inf: -30.7413 | lim_sup: 266.3088
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -78.2000 | lim_sup: 363.4000
Límites IQR -> lim_inf: 250.0000 | lim_sup: 250.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -162.6675 | lim_sup: 271.1125
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -102.0000 | lim_sup: 170.0000
Límites IQR -> lim_inf: -0.3975 | lim_sup: 210.3425
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -504.2700 | lim_sup: 840.4500
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -229.1025 | lim_sup: 2663.2575
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -308.0138 | lim_sup: 513.3563
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 79.9600 | lim_sup: 79.9600
Límites IQR -> lim_inf: -281.6938 | lim_sup: 2438.6562
Límites IQR -> lim_inf: 259.1000 | lim_sup: 259.1000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -398.7000 | lim_sup: 664.5000
Límites IQR -> lim_inf: 110.0000 | lim_sup: 110.0000
Límites IQR -> lim_inf: 68.0000 | lim_sup: 68.0000
Límites IQR -> lim_inf: -283.3388 | lim_sup: 1736.5312
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -41.5688 | lim_sup: 69.2812
Límites IQR -> lim_inf: -556.9462 | lim_sup: 928.2437
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -433.7588 | lim_sup: 2504.6912
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -3044.0887 | lim_sup: 7257.6812
Límites IQR -> lim_inf: -2746.3637 | lim_sup: 10400.0863
Límites IQR -> lim_inf: 15986.7687 | lim_sup: 20945.9988
Límites IQR -> lim_inf: -2686.6788 | lim_sup: 5146.9113
Límites IQR -> lim_inf: -11694.2663 | lim_sup: 66165.1237
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 108.0987 | lim_sup: 113.1687
Límites IQR -> lim_inf: -92.3238 | lim_sup: 2135.4262
Límites IQR -> lim_inf: -337.5000 | lim_sup: 562.5000
Límites IQR -> lim_inf: 64.0000 | lim_sup: 64.0000
Límites IQR -> lim_inf: -32.9962 | lim_sup: 54.9937
Límites IQR -> lim_inf: 408.7300 | lim_sup: 408.7300
Límites IQR -> lim_inf: -218.3550 | lim_sup: 1027.4250
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -438.9375 | lim_sup: 731.5625
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 68.0000 | lim_sup: 68.0000
Límites IQR -> lim_inf: -408.0212 | lim_sup: 2328.5687
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -168.2550 | lim_sup: 280.4250
Límites IQR -> lim_inf: -765.9750 | lim_sup: 2981.5050
Límites IQR -> lim_inf: -2160.8600 | lim_sup: 3973.5400
Límites IQR -> lim_inf: -120.5100 | lim_sup: 200.8500
Límites IQR -> lim_inf: 3274.5938 | lim_sup: 4526.4237
Límites IQR -> lim_inf: -1174.4175 | lim_sup: 2515.7225
Límites IQR -> lim_inf: -8174.7375 | lim_sup: 25244.0825
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -459.1500 | lim_sup: 765.2500
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -141.4512 | lim_sup: 844.9187
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 122.2650 | lim_sup: 163.2650
Límites IQR -> lim_inf: -39.3025 | lim_sup: 208.9575
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 180.7850 | lim_sup: 241.4650
Límites IQR -> lim_inf: -36.4087 | lim_sup: 326.0612
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 46.9800 | lim_sup: 46.9800
Límites IQR -> lim_inf: -159.0987 | lim_sup: 338.7312
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 113.3500 | lim_sup: 127.5900
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -284.4425 | lim_sup: 762.1975
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -242.1112 | lim_sup: 403.5187
Límites IQR -> lim_inf: 110.0000 | lim_sup: 110.0000
Límites IQR -> lim_inf: -294.5100 | lim_sup: 490.8500
Límites IQR -> lim_inf: -141.9762 | lim_sup: 2034.2137
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -41.5688 | lim_sup: 69.2812
Límites IQR -> lim_inf: -308.4750 | lim_sup: 514.1250
Límites IQR -> lim_inf: 110.0000 | lim_sup: 110.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -265.6250 | lim_sup: 2869.3550
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -270.0000 | lim_sup: 450.0000
Límites IQR -> lim_inf: -586.7700 | lim_sup: 3611.7300
Límites IQR -> lim_inf: -2543.1400 | lim_sup: 8837.3200
Límites IQR -> lim_inf: -2654.4275 | lim_sup: 6108.5525
Límites IQR -> lim_inf: 8309.1950 | lim_sup: 12900.8750
Límites IQR -> lim_inf: -1251.5250 | lim_sup: 2085.8750
Límites IQR -> lim_inf: -3925.5437 | lim_sup: 52528.3262
Límites IQR -> lim_inf: 213.9045 | lim_sup: 590.0864
Límites IQR -> lim_inf: 49.0909 | lim_sup: 100.0000
Límites IQR -> lim_inf: -830.0489 | lim_sup: 3770.4693
Límites IQR -> lim_inf: 3764.1182 | lim_sup: 5488.4091
Límites IQR -> lim_inf: -153.4091 | lim_sup: 255.6818
Límites IQR -> lim_inf: 11944.0000 | lim_sup: 15333.2727
Límites IQR -> lim_inf: 97.2500 | lim_sup: 523.3955
Límites IQR -> lim_inf: 49.0909 | lim_sup: 100.0000
Límites IQR -> lim_inf: -1300.9727 | lim_sup: 3527.3909
Límites IQR -> lim_inf: 4439.9682 | lim_sup: 6987.2045
Límites IQR -> lim_inf: -547.1591 | lim_sup: 911.9318
Límites IQR -> lim_inf: 10153.0705 | lim_sup: 12210.8886
Límites IQR -> lim_inf: 73.9182 | lim_sup: 510.0636
Límites IQR -> lim_inf: 66.3636 | lim_sup: 131.8182
Límites IQR -> lim_inf: -1900.2636 | lim_sup: 4142.8636
Límites IQR -> lim_inf: 3890.3591 | lim_sup: 7365.0864
Límites IQR -> lim_inf: -545.5091 | lim_sup: 909.1818
Límites IQR -> lim_inf: 8762.0307 | lim_sup: 10998.3852
Límites IQR -> lim_inf: 48.8500 | lim_sup: 595.6500
Límites IQR -> lim_inf: -504.5455 | lim_sup: 1022.7273
Límites IQR -> lim_inf: -2705.6250 | lim_sup: 8006.0295
Límites IQR -> lim_inf: 6490.1591 | lim_sup: 8466.3409
Límites IQR -> lim_inf: -1153.9000 | lim_sup: 1965.5909
Límites IQR -> lim_inf: 15128.7080 | lim_sup: 19879.3352
Límites IQR -> lim_inf: -34.4364 | lim_sup: 266.3636
Límites IQR -> lim_inf: 66.3636 | lim_sup: 131.8182
Límites IQR -> lim_inf: -798.7295 | lim_sup: 1904.6341
Límites IQR -> lim_inf: 3494.4545 | lim_sup: 3730.8182
Límites IQR -> lim_inf: -252.2727 | lim_sup: 420.4545
Límites IQR -> lim_inf: 1417.3125 | lim_sup: 9850.0852
Límites IQR -> lim_inf: 165.9182 | lim_sup: 360.7545
Límites IQR -> lim_inf: 66.3636 | lim_sup: 131.8182
Límites IQR -> lim_inf: -1142.4648 | lim_sup: 2595.0170
Límites IQR -> lim_inf: 3570.0455 | lim_sup: 3851.8636
Límites IQR -> lim_inf: -86.9318 | lim_sup: 144.8864
Límites IQR -> lim_inf: 2413.8864 | lim_sup: 12284.2500
Límites IQR -> lim_inf: 136.1318 | lim_sup: 636.3136
Límites IQR -> lim_inf: 88.6364 | lim_sup: 143.1818
Límites IQR -> lim_inf: -1357.7761 | lim_sup: 4135.5239
Límites IQR -> lim_inf: 4212.2955 | lim_sup: 6126.5864
Límites IQR -> lim_inf: -340.9091 | lim_sup: 568.1818
Límites IQR -> lim_inf: 8293.5261 | lim_sup: 12404.0807
Límites IQR -> lim_inf: 128.3227 | lim_sup: 308.4318
Límites IQR -> lim_inf: 88.6364 | lim_sup: 143.1818
Límites IQR -> lim_inf: -1121.2955 | lim_sup: 3449.3591
Límites IQR -> lim_inf: 4709.3068 | lim_sup: 5475.9795
Límites IQR -> lim_inf: -352.6091 | lim_sup: 587.6818
Límites IQR -> lim_inf: 4979.1568 | lim_sup: 12827.1205
Límites IQR -> lim_inf: 103.2682 | lim_sup: 594.3591
Límites IQR -> lim_inf: 1142.2727 | lim_sup: 1193.1818
Límites IQR -> lim_inf: -2395.6045 | lim_sup: 7782.8318
Límites IQR -> lim_inf: 4677.7636 | lim_sup: 6861.5455
Límites IQR -> lim_inf: -490.9091 | lim_sup: 818.1818
Límites IQR -> lim_inf: 11931.5955 | lim_sup: 25971.0136
Límites IQR -> lim_inf: 95.3273 | lim_sup: 735.4727
Límites IQR -> lim_inf: -136.3636 | lim_sup: 409.0909
Límites IQR -> lim_inf: -1020.3398 | lim_sup: 3532.9420
Límites IQR -> lim_inf: 6706.9091 | lim_sup: 7841.4545
Límites IQR -> lim_inf: -409.2375 | lim_sup: 682.0625
Límites IQR -> lim_inf: 15369.4330 | lim_sup: 18380.4966
Límites IQR -> lim_inf: 51.4000 | lim_sup: 479.1818
Límites IQR -> lim_inf: 49.0909 | lim_sup: 100.0000
Límites IQR -> lim_inf: -2160.1034 | lim_sup: 5507.9602
Límites IQR -> lim_inf: 5386.5909 | lim_sup: 8283.6818
Límites IQR -> lim_inf: -81.8182 | lim_sup: 136.3636
Límites IQR -> lim_inf: 8010.1000 | lim_sup: 15238.2818
Límites IQR -> lim_inf: 103.2682 | lim_sup: 594.3591
Límites IQR -> lim_inf: 49.0909 | lim_sup: 100.0000
Límites IQR -> lim_inf: -842.7182 | lim_sup: 3866.0091
Límites IQR -> lim_inf: 5439.6591 | lim_sup: 8479.8409
Límites IQR -> lim_inf: -230.1136 | lim_sup: 383.5227
Límites IQR -> lim_inf: 15417.5034 | lim_sup: 18399.2216
Límites IQR -> lim_inf: 105.0045 | lim_sup: 495.4409
Límites IQR -> lim_inf: -250.0000 | lim_sup: 659.0909
Límites IQR -> lim_inf: -809.3216 | lim_sup: 2374.7239
Límites IQR -> lim_inf: 3879.0523 | lim_sup: 7361.5795
Límites IQR -> lim_inf: -761.7648 | lim_sup: 1269.6080
Límites IQR -> lim_inf: 7503.6693 | lim_sup: 10433.5148
Límites IQR -> lim_inf: 93.1955 | lim_sup: 678.3955
Límites IQR -> lim_inf: -284.0909 | lim_sup: 715.9091
Límites IQR -> lim_inf: -1896.1977 | lim_sup: 4910.0386
Límites IQR -> lim_inf: 5278.6045 | lim_sup: 7358.1318
Límites IQR -> lim_inf: -322.3841 | lim_sup: 537.3068
Límites IQR -> lim_inf: 11828.9159 | lim_sup: 14922.3159
Límites IQR -> lim_inf: 257.7955 | lim_sup: 552.3045
Límites IQR -> lim_inf: 49.0909 | lim_sup: 100.0000
Límites IQR -> lim_inf: -2222.7455 | lim_sup: 5458.1273
Límites IQR -> lim_inf: 5535.6864 | lim_sup: 8363.1409
Límites IQR -> lim_inf: -1055.5682 | lim_sup: 1801.7045
Límites IQR -> lim_inf: 12337.7330 | lim_sup: 15916.2693
Límites IQR -> lim_inf: 121.8455 | lim_sup: 206.4273
Límites IQR -> lim_inf: 31.8182 | lim_sup: 68.1818
Límites IQR -> lim_inf: -840.0852 | lim_sup: 1771.4148
Límites IQR -> lim_inf: 3586.6364 | lim_sup: 3729.1818
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 2360.0750 | lim_sup: 13028.1114
Límites IQR -> lim_inf: 493.5364 | lim_sup: 493.5364
Límites IQR -> lim_inf: 49.0909 | lim_sup: 100.0000
Límites IQR -> lim_inf: -1293.0739 | lim_sup: 5463.0625
Límites IQR -> lim_inf: 63.6364 | lim_sup: 63.6364
Límites IQR -> lim_inf: 7163.3636 | lim_sup: 7327.7273
Límites IQR -> lim_inf: -498.4091 | lim_sup: 830.6818
Límites IQR -> lim_inf: 16545.4534 | lim_sup: 21226.8261
Límites IQR -> lim_inf: 183.9591 | lim_sup: 675.3409
Límites IQR -> lim_inf: 155.0273 | lim_sup: 155.0273
Límites IQR -> lim_inf: -995.8795 | lim_sup: 2348.8659
Límites IQR -> lim_inf: 3550.1364 | lim_sup: 3624.6818
Límites IQR -> lim_inf: -51.1364 | lim_sup: 85.2273
Límites IQR -> lim_inf: 6334.9148 | lim_sup: 8596.7239
Límites IQR -> lim_inf: 59.0455 | lim_sup: 587.0455
Límites IQR -> lim_inf: 224.7545 | lim_sup: 417.4091
Límites IQR -> lim_inf: -1937.6614 | lim_sup: 5341.7932
Límites IQR -> lim_inf: 5641.0364 | lim_sup: 6585.0364
Límites IQR -> lim_inf: -323.8636 | lim_sup: 539.7727
Límites IQR -> lim_inf: 19353.7273 | lim_sup: 25012.9455
Límites IQR -> lim_inf: 95.7636 | lim_sup: 510.8182
Límites IQR -> lim_inf: -3179.7545 | lim_sup: 5299.5909
Límites IQR -> lim_inf: -2224.3182 | lim_sup: 10137.2636
Límites IQR -> lim_inf: 5464.0727 | lim_sup: 8442.1091
Límites IQR -> lim_inf: -771.8045 | lim_sup: 1286.3409
Límites IQR -> lim_inf: 14454.5398 | lim_sup: 18920.6398
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -303.6300 | lim_sup: 506.0500
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -77.3200 | lim_sup: 1286.1200
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -306.0000 | lim_sup: 510.0000
Límites IQR -> lim_inf: 110.0000 | lim_sup: 110.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -388.7600 | lim_sup: 1513.6800
Límites IQR -> lim_inf: -135.0000 | lim_sup: 225.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -332.2275 | lim_sup: 553.7125
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -477.2600 | lim_sup: 2411.6400
Límites IQR -> lim_inf: -219.3750 | lim_sup: 365.6250
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -374.7000 | lim_sup: 624.5000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -621.9137 | lim_sup: 2793.1762
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -1735.5000 | lim_sup: 2892.5000
Límites IQR -> lim_inf: -11.5312 | lim_sup: 263.5387
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -201.3750 | lim_sup: 335.6250
Límites IQR -> lim_inf: 55.5350 | lim_sup: 200.7750
Límites IQR -> lim_inf: 254.0000 | lim_sup: 254.0000
Límites IQR -> lim_inf: -171.8425 | lim_sup: 2639.1375
Límites IQR -> lim_inf: -135.0000 | lim_sup: 225.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -204.8625 | lim_sup: 341.4375
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -253.7463 | lim_sup: 2791.2638
Límites IQR -> lim_inf: 170.5591 | lim_sup: 665.9409
Límites IQR -> lim_inf: 49.0909 | lim_sup: 100.0000
Límites IQR -> lim_inf: -1953.5830 | lim_sup: 4507.0989
Límites IQR -> lim_inf: 4839.2909 | lim_sup: 5052.7455
Límites IQR -> lim_inf: -204.5455 | lim_sup: 340.9091
Límites IQR -> lim_inf: 12031.5943 | lim_sup: 14234.2216
Límites IQR -> lim_inf: 154.3591 | lim_sup: 159.7045
Límites IQR -> lim_inf: 454.5455 | lim_sup: 454.5455
Límites IQR -> lim_inf: -1013.1409 | lim_sup: 1688.5682
Límites IQR -> lim_inf: 614.3818 | lim_sup: 1653.8727
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 1113.0852 | lim_sup: 6315.4943
Límites IQR -> lim_inf: 101.2864 | lim_sup: 520.8136
Límites IQR -> lim_inf: 500.0000 | lim_sup: 500.0000
Límites IQR -> lim_inf: -1497.9432 | lim_sup: 3980.6386
Límites IQR -> lim_inf: 4222.0682 | lim_sup: 7528.0682
Límites IQR -> lim_inf: -409.0841 | lim_sup: 681.8068
Límites IQR -> lim_inf: 23276.9955 | lim_sup: 30792.1591
Límites IQR -> lim_inf: -623.6260 | lim_sup: 1545.2744
Límites IQR -> lim_inf: 600.0000 | lim_sup: 600.0000
Límites IQR -> lim_inf: -152.2667 | lim_sup: 290.2419
Límites IQR -> lim_inf: 1701.4328 | lim_sup: 6724.7450
Límites IQR -> lim_inf: 109.4848 | lim_sup: 3461.5042
Límites IQR -> lim_inf: 600.0000 | lim_sup: 600.0000
Límites IQR -> lim_inf: -549.4245 | lim_sup: 1021.3075
Límites IQR -> lim_inf: 1899.1721 | lim_sup: 13528.7486
Límites IQR -> lim_inf: -500.8709 | lim_sup: 1175.7180
Límites IQR -> lim_inf: -97.8944 | lim_sup: 215.8870
Límites IQR -> lim_inf: 1289.4087 | lim_sup: 1830.9220
Límites IQR -> lim_inf: -516.2367 | lim_sup: 1678.9204
Límites IQR -> lim_inf: -121.5787 | lim_sup: 239.0312
Límites IQR -> lim_inf: 850.5422 | lim_sup: 3638.5524
Límites IQR -> lim_inf: -474.5459 | lim_sup: 1028.4315
Límites IQR -> lim_inf: -22.3700 | lim_sup: 77.9500
Límites IQR -> lim_inf: 754.2430 | lim_sup: 2688.8861
Límites IQR -> lim_inf: -411.3802 | lim_sup: 970.8361
Límites IQR -> lim_inf: -66.2400 | lim_sup: 110.4000
Límites IQR -> lim_inf: 988.1715 | lim_sup: 3304.7007
Límites IQR -> lim_inf: -607.1558 | lim_sup: 1335.7595
Límites IQR -> lim_inf: -38.0100 | lim_sup: 95.8300
Límites IQR -> lim_inf: 284.1209 | lim_sup: 4268.9192
Límites IQR -> lim_inf: 68.8800 | lim_sup: 109.2000
Límites IQR -> lim_inf: -1299.1851 | lim_sup: 3619.9331
Límites IQR -> lim_inf: -246.9456 | lim_sup: 455.0827
Límites IQR -> lim_inf: 1253.2759 | lim_sup: 6658.6674
Límites IQR -> lim_inf: -816.6443 | lim_sup: 1859.8154
Límites IQR -> lim_inf: 4441.6963 | lim_sup: 5309.9263
Límites IQR -> lim_inf: -148.6665 | lim_sup: 412.1055
Límites IQR -> lim_inf: 406.5040 | lim_sup: 586.5283
Límites IQR -> lim_inf: -485.4447 | lim_sup: 948.2313
Límites IQR -> lim_inf: -70.5659 | lim_sup: 117.6099
Límites IQR -> lim_inf: 976.6573 | lim_sup: 1087.5349
Límites IQR -> lim_inf: -1164.9207 | lim_sup: 5764.3863
Límites IQR -> lim_inf: -803.4491 | lim_sup: 1409.4819
Límites IQR -> lim_inf: 2809.9389 | lim_sup: 13371.0307
Límites IQR -> lim_inf: -245.2192 | lim_sup: 482.6183
Límites IQR -> lim_inf: 600.0000 | lim_sup: 600.0000
Límites IQR -> lim_inf: -28.7400 | lim_sup: 77.4680
Límites IQR -> lim_inf: -5.3611 | lim_sup: 1056.8327
Límites IQR -> lim_inf: -407.3199 | lim_sup: 814.6060
Límites IQR -> lim_inf: 600.0000 | lim_sup: 600.0000
Límites IQR -> lim_inf: -19.6015 | lim_sup: 46.7961
Límites IQR -> lim_inf: 212.7136 | lim_sup: 1144.8378
Límites IQR -> lim_inf: -1396.5583 | lim_sup: 6345.7265
Límites IQR -> lim_inf: -504.4228 | lim_sup: 1126.9580
Límites IQR -> lim_inf: 5604.3070 | lim_sup: 26104.8960
Límites IQR -> lim_inf: -728.2979 | lim_sup: 4505.6372
Límites IQR -> lim_inf: -467.8102 | lim_sup: 990.2170
Límites IQR -> lim_inf: 3378.1440 | lim_sup: 17894.7560
Límites IQR -> lim_inf: -476.0184 | lim_sup: 1119.2622
Límites IQR -> lim_inf: -95.7560 | lim_sup: 202.2600
Límites IQR -> lim_inf: 471.7773 | lim_sup: 3940.4348
Límites IQR -> lim_inf: -779.4589 | lim_sup: 1949.3563
Límites IQR -> lim_inf: -120.1125 | lim_sup: 264.4275
Límites IQR -> lim_inf: 757.8717 | lim_sup: 5858.3394
Límites IQR -> lim_inf: -1000.7213 | lim_sup: 3068.7344
Límites IQR -> lim_inf: -139.2126 | lim_sup: 301.6210
Límites IQR -> lim_inf: 1013.3390 | lim_sup: 7557.5202
Límites IQR -> lim_inf: -726.2362 | lim_sup: 1557.2446
Límites IQR -> lim_inf: -57.9150 | lim_sup: 128.5250
Límites IQR -> lim_inf: 251.3045 | lim_sup: 2576.9444
Límites IQR -> lim_inf: -424.4872 | lim_sup: 827.9092
Límites IQR -> lim_inf: -22.9280 | lim_sup: 56.6112
Límites IQR -> lim_inf: 282.2997 | lim_sup: 1197.3144
Límites IQR -> lim_inf: -277.1423 | lim_sup: 920.3685
Límites IQR -> lim_inf: -63.9528 | lim_sup: 138.5880
Límites IQR -> lim_inf: -5.0062 | lim_sup: 3854.5030
Límites IQR -> lim_inf: -286.3220 | lim_sup: 837.9436
Límites IQR -> lim_inf: -69.0159 | lim_sup: 139.0265
Límites IQR -> lim_inf: 390.2628 | lim_sup: 3921.7104
Límites IQR -> lim_inf: -771.5521 | lim_sup: 1577.7935
Límites IQR -> lim_inf: -49.7587 | lim_sup: 121.4592
Límites IQR -> lim_inf: 830.6945 | lim_sup: 2232.3452
Límites IQR -> lim_inf: -323.2201 | lim_sup: 749.3978
Límites IQR -> lim_inf: -62.6400 | lim_sup: 126.5760
Límites IQR -> lim_inf: 117.3858 | lim_sup: 1464.7287
Límites IQR -> lim_inf: -390.6372 | lim_sup: 902.5856
Límites IQR -> lim_inf: -110.3500 | lim_sup: 224.0500
Límites IQR -> lim_inf: 37.3177 | lim_sup: 2789.6522
Límites IQR -> lim_inf: -180.8982 | lim_sup: 553.8232
Límites IQR -> lim_inf: -63.0037 | lim_sup: 156.6062
Límites IQR -> lim_inf: 489.3038 | lim_sup: 3347.7999
Límites IQR -> lim_inf: -212.9365 | lim_sup: 498.2173
Límites IQR -> lim_inf: -78.1125 | lim_sup: 178.1875
Límites IQR -> lim_inf: 179.5700 | lim_sup: 2064.4936
Límites IQR -> lim_inf: -518.5243 | lim_sup: 2379.5109
Límites IQR -> lim_inf: -24.4659 | lim_sup: 104.7765
Límites IQR -> lim_inf: 1325.1181 | lim_sup: 10134.3692
Límites IQR -> lim_inf: -318.1363 | lim_sup: 751.9305
Límites IQR -> lim_inf: -36.8400 | lim_sup: 93.4000
Límites IQR -> lim_inf: 623.9692 | lim_sup: 3273.3901
Límites IQR -> lim_inf: -843.2727 | lim_sup: 1951.0892
Límites IQR -> lim_inf: 600.0000 | lim_sup: 600.0000
Límites IQR -> lim_inf: -50.1750 | lim_sup: 115.6250
Límites IQR -> lim_inf: -841.6530 | lim_sup: 4948.6968
Límites IQR -> lim_inf: -24.5967 | lim_sup: 1563.0198
Límites IQR -> lim_inf: -82.4895 | lim_sup: 185.4825
Límites IQR -> lim_inf: 1187.2826 | lim_sup: 6902.0096
Límites IQR -> lim_inf: -648.4745 | lim_sup: 1460.7215
Límites IQR -> lim_inf: -21.8613 | lim_sup: 84.9688
Límites IQR -> lim_inf: 1445.1106 | lim_sup: 6278.4522
Límites IQR -> lim_inf: -436.4840 | lim_sup: 1001.5000
Límites IQR -> lim_inf: -38.0225 | lim_sup: 112.7575
Límites IQR -> lim_inf: -681.0997 | lim_sup: 8205.9061
Límites IQR -> lim_inf: -168.1700 | lim_sup: 1050.9500
Límites IQR -> lim_inf: -225.4025 | lim_sup: 438.3647
Límites IQR -> lim_inf: 306.7065 | lim_sup: 4020.6481
Límites IQR -> lim_inf: -621.9629 | lim_sup: 1766.7953
Límites IQR -> lim_inf: -117.5500 | lim_sup: 265.2500
Límites IQR -> lim_inf: 806.7714 | lim_sup: 5335.3213
Límites IQR -> lim_inf: -349.3534 | lim_sup: 985.9064
Límites IQR -> lim_inf: -99.8500 | lim_sup: 187.7500
Límites IQR -> lim_inf: 860.3250 | lim_sup: 2533.0286
Límites IQR -> lim_inf: -726.1325 | lim_sup: 2031.8875
Límites IQR -> lim_inf: -66.9825 | lim_sup: 225.7895
Límites IQR -> lim_inf: 500.2857 | lim_sup: 3392.2669
Límites IQR -> lim_inf: -343.7205 | lim_sup: 840.7019
Límites IQR -> lim_inf: -61.0722 | lim_sup: 101.7870
Límites IQR -> lim_inf: 330.0806 | lim_sup: 1942.4234
Límites IQR -> lim_inf: -443.6007 | lim_sup: 1093.2595
Límites IQR -> lim_inf: 600.0000 | lim_sup: 600.0000
Límites IQR -> lim_inf: -131.5500 | lim_sup: 219.2500
Límites IQR -> lim_inf: 576.6447 | lim_sup: 2360.0537
Límites IQR -> lim_inf: -506.7464 | lim_sup: 2228.4747
Límites IQR -> lim_inf: 600.0000 | lim_sup: 600.0000
Límites IQR -> lim_inf: -186.1087 | lim_sup: 358.1812
Límites IQR -> lim_inf: 2571.1106 | lim_sup: 6995.0231
Límites IQR -> lim_inf: -467.1099 | lim_sup: 1254.6571
Límites IQR -> lim_inf: -39.3670 | lim_sup: 86.9450
Límites IQR -> lim_inf: 874.9282 | lim_sup: 3130.9005
Límites IQR -> lim_inf: -329.7701 | lim_sup: 696.9361
Límites IQR -> lim_inf: -75.6000 | lim_sup: 126.0000
Límites IQR -> lim_inf: 238.3319 | lim_sup: 1001.9058
Límites IQR -> lim_inf: -1039.4335 | lim_sup: 2089.5939
Límites IQR -> lim_inf: 600.0000 | lim_sup: 600.0000
Límites IQR -> lim_inf: -215.5565 | lim_sup: 417.9275
Límites IQR -> lim_inf: 888.0866 | lim_sup: 4367.7675
Límites IQR -> lim_inf: -413.1050 | lim_sup: 1307.0684
Límites IQR -> lim_inf: -80.9810 | lim_sup: 179.7683
Límites IQR -> lim_inf: 1506.0915 | lim_sup: 2995.9759
Límites IQR -> lim_inf: -895.9638 | lim_sup: 3381.3652
Límites IQR -> lim_inf: -496.8982 | lim_sup: 1190.4964
Límites IQR -> lim_inf: 688.6497 | lim_sup: 10713.9367
Límites IQR -> lim_inf: -391.4056 | lim_sup: 1129.2597
Límites IQR -> lim_inf: -74.4600 | lim_sup: 156.1000
Límites IQR -> lim_inf: 542.7695 | lim_sup: 2470.9024
Límites IQR -> lim_inf: -441.5856 | lim_sup: 1297.9680
Límites IQR -> lim_inf: -85.5000 | lim_sup: 174.5000
Límites IQR -> lim_inf: 482.2110 | lim_sup: 4172.5382
Límites IQR -> lim_inf: -589.2091 | lim_sup: 2333.1211
Límites IQR -> lim_inf: -28.5175 | lim_sup: 96.8625
Límites IQR -> lim_inf: 791.1437 | lim_sup: 5771.3149
Límites IQR -> lim_inf: -362.0118 | lim_sup: 974.4926
Límites IQR -> lim_inf: 3570.8524 | lim_sup: 3570.8524
Límites IQR -> lim_inf: -94.0470 | lim_sup: 211.6250
Límites IQR -> lim_inf: 1080.7015 | lim_sup: 4500.4045
Límites IQR -> lim_inf: -525.0076 | lim_sup: 1557.6906
Límites IQR -> lim_inf: -72.5500 | lim_sup: 158.2500
Límites IQR -> lim_inf: 489.0133 | lim_sup: 4057.7613
Límites IQR -> lim_inf: -452.3453 | lim_sup: 963.1967
Límites IQR -> lim_inf: -32.0925 | lim_sup: 85.4875
Límites IQR -> lim_inf: 254.8974 | lim_sup: 2431.0275
Límites IQR -> lim_inf: -451.3292 | lim_sup: 1125.5487
Límites IQR -> lim_inf: -138.8022 | lim_sup: 231.3370
Límites IQR -> lim_inf: -70.4224 | lim_sup: 3136.9844
Límites IQR -> lim_inf: -414.0118 | lim_sup: 986.2142
Límites IQR -> lim_inf: -60.6750 | lim_sup: 140.7250
Límites IQR -> lim_inf: 30.2744 | lim_sup: 2189.7177
Límites IQR -> lim_inf: -1657.4863 | lim_sup: 6542.8351
Límites IQR -> lim_inf: 3651.8672 | lim_sup: 3651.8672
Límites IQR -> lim_inf: -260.3801 | lim_sup: 1689.9668
Límites IQR -> lim_inf: -3966.2396 | lim_sup: 12756.0354
Límites IQR -> lim_inf: -6.9000 | lim_sup: 11.5000
Límites IQR -> lim_inf: -1264.2180 | lim_sup: 4518.2186
Límites IQR -> lim_inf: -300.4339 | lim_sup: 579.8699
Límites IQR -> lim_inf: 2016.9301 | lim_sup: 7919.3361
Límites IQR -> lim_inf: -633.5679 | lim_sup: 4596.7742
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -399.0367 | lim_sup: 880.3091
Límites IQR -> lim_inf: -2164.9744 | lim_sup: 15651.7986
Límites IQR -> lim_inf: -90.0674 | lim_sup: 3059.1790
Límites IQR -> lim_inf: -482.8068 | lim_sup: 881.6513
Límites IQR -> lim_inf: 2377.8845 | lim_sup: 11525.4022
Límites IQR -> lim_inf: -566.2650 | lim_sup: 943.7750
Límites IQR -> lim_inf: -58.6800 | lim_sup: 97.8000
Límites IQR -> lim_inf: 77.5526 | lim_sup: 445.2898
Límites IQR -> lim_inf: -652.8481 | lim_sup: 1400.3639
Límites IQR -> lim_inf: 600.0000 | lim_sup: 600.0000
Límites IQR -> lim_inf: -108.6527 | lim_sup: 234.4211
Límites IQR -> lim_inf: 605.9152 | lim_sup: 4473.7203
Límites IQR -> lim_inf: -857.3446 | lim_sup: 4676.7670
Límites IQR -> lim_inf: 600.0000 | lim_sup: 600.0000
Límites IQR -> lim_inf: -363.9162 | lim_sup: 786.5270
Límites IQR -> lim_inf: 867.6791 | lim_sup: 11532.2800
Límites IQR -> lim_inf: -848.7227 | lim_sup: 5065.4940
Límites IQR -> lim_inf: -458.5846 | lim_sup: 833.6410
Límites IQR -> lim_inf: 1726.3470 | lim_sup: 11770.7725
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -535.1025 | lim_sup: 891.8375
Límites IQR -> lim_inf: 145.6300 | lim_sup: 163.9500
Límites IQR -> lim_inf: -141.2250 | lim_sup: 235.3750
Límites IQR -> lim_inf: -255.7425 | lim_sup: 1266.4375
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -277.9538 | lim_sup: 463.2563
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 263.2000 | lim_sup: 263.2000
Límites IQR -> lim_inf: -102.9500 | lim_sup: 1693.2900
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 122.4100 | lim_sup: 122.4100
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 753.6300 | lim_sup: 1536.8300
Límites IQR -> lim_inf: -2505.0950 | lim_sup: 9183.6050
Límites IQR -> lim_inf: 924.0000 | lim_sup: 924.0000
Límites IQR -> lim_inf: -34.6150 | lim_sup: 351.0250
Límites IQR -> lim_inf: -183.7500 | lim_sup: 306.2500
Límites IQR -> lim_inf: -5653.6362 | lim_sup: 39793.1137
Límites IQR -> lim_inf: 580.8900 | lim_sup: 580.8900
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -130.1250 | lim_sup: 216.8750
Límites IQR -> lim_inf: 182.6500 | lim_sup: 214.4100
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -84.7362 | lim_sup: 426.3338
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -2552.2125 | lim_sup: 4683.3275
Límites IQR -> lim_inf: -1805.6138 | lim_sup: 4576.7163
Límites IQR -> lim_inf: 3958.7600 | lim_sup: 5286.7600
Límites IQR -> lim_inf: -714.4725 | lim_sup: 1190.7875
Límites IQR -> lim_inf: -3050.8938 | lim_sup: 24105.5163
Límites IQR -> lim_inf: 190.5636 | lim_sup: 576.7455
Límites IQR -> lim_inf: 49.0909 | lim_sup: 100.0000
Límites IQR -> lim_inf: -982.5852 | lim_sup: 2795.3693
Límites IQR -> lim_inf: 4716.2864 | lim_sup: 6954.5773
Límites IQR -> lim_inf: -136.3636 | lim_sup: 227.2727
Límites IQR -> lim_inf: 6756.1420 | lim_sup: 10778.6057
Límites IQR -> lim_inf: -263.2500 | lim_sup: 438.7500
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -166.5337 | lim_sup: 277.5562
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 217.1100 | lim_sup: 217.1100
Límites IQR -> lim_inf: -105.0000 | lim_sup: 175.0000
Límites IQR -> lim_inf: -391.4700 | lim_sup: 652.4500
Límites IQR -> lim_inf: -192.0000 | lim_sup: 320.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -90.0000 | lim_sup: 150.0000
Límites IQR -> lim_inf: -180.0000 | lim_sup: 300.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -413.7900 | lim_sup: 689.6500
Límites IQR -> lim_inf: -194.7000 | lim_sup: 324.5000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -393.0263 | lim_sup: 655.0438
Límites IQR -> lim_inf: 110.0000 | lim_sup: 110.0000
Límites IQR -> lim_inf: -25.5000 | lim_sup: 42.5000
Límites IQR -> lim_inf: -308.3038 | lim_sup: 1407.8863
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -141.5400 | lim_sup: 235.9000
Límites IQR -> lim_inf: 110.0000 | lim_sup: 110.0000
Límites IQR -> lim_inf: -225.3850 | lim_sup: 1409.0350
Límites IQR -> lim_inf: -135.0000 | lim_sup: 405.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -323.4675 | lim_sup: 539.1125
Límites IQR -> lim_inf: 110.0000 | lim_sup: 110.0000
Límites IQR -> lim_inf: -249.8713 | lim_sup: 1418.6588
Límites IQR -> lim_inf: -90.0000 | lim_sup: 150.0000
Límites IQR -> lim_inf: -248.7750 | lim_sup: 414.6250
Límites IQR -> lim_inf: -214.9663 | lim_sup: 763.3637
Límites IQR -> lim_inf: -1313.5095 | lim_sup: 4700.3105
Límites IQR -> lim_inf: -479.6917 | lim_sup: 932.8195
Límites IQR -> lim_inf: 3316.1329 | lim_sup: 20122.5447
Límites IQR -> lim_inf: -97.6313 | lim_sup: 162.7188
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -402.9000 | lim_sup: 671.5000
Límites IQR -> lim_inf: -906.2200 | lim_sup: 4603.7000
Límites IQR -> lim_inf: -152.0362 | lim_sup: 568.8937
Límites IQR -> lim_inf: -8.5925 | lim_sup: 225.7275
Límites IQR -> lim_inf: -616.3688 | lim_sup: 1922.7812
Límites IQR -> lim_inf: 80.0000 | lim_sup: 80.0000
Límites IQR -> lim_inf: -475.3650 | lim_sup: 792.2750
Límites IQR -> lim_inf: 100.0000 | lim_sup: 100.0000
Límites IQR -> lim_inf: 100.0000 | lim_sup: 100.0000
Límites IQR -> lim_inf: -308.8950 | lim_sup: 1928.4050
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -1310.2500 | lim_sup: 2183.7500
Límites IQR -> lim_inf: -924.1500 | lim_sup: 2220.5700
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 3285.1500 | lim_sup: 5700.7500
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -833.5887 | lim_sup: 9669.3812
Límites IQR -> lim_inf: -34.8825 | lim_sup: 58.1375
Límites IQR -> lim_inf: 27.2312 | lim_sup: 748.6413
Límites IQR -> lim_inf: -898.8250 | lim_sup: 2412.1950
Límites IQR -> lim_inf: 2806.0300 | lim_sup: 10753.8700
Límites IQR -> lim_inf: -102.0000 | lim_sup: 170.0000
Límites IQR -> lim_inf: -2869.2900 | lim_sup: 14231.3300
Límites IQR -> lim_inf: 406.1800 | lim_sup: 406.1800
Límites IQR -> lim_inf: 68.0000 | lim_sup: 68.0000
Límites IQR -> lim_inf: -70.4213 | lim_sup: 493.2488
Límites IQR -> lim_inf: -189.0325 | lim_sup: 671.9875
Límites IQR -> lim_inf: -412.5000 | lim_sup: 687.5000
Límites IQR -> lim_inf: -852.3025 | lim_sup: 1732.7775
Límites IQR -> lim_inf: -1057.1637 | lim_sup: 2929.5063
Límites IQR -> lim_inf: -641.6237 | lim_sup: 2934.3663
Límites IQR -> lim_inf: -22.5000 | lim_sup: 37.5000
Límites IQR -> lim_inf: -115.3350 | lim_sup: 192.2250
Límites IQR -> lim_inf: -194.9287 | lim_sup: 1826.4212
Límites IQR -> lim_inf: 133.8500 | lim_sup: 133.8500
Límites IQR -> lim_inf: -1052.5712 | lim_sup: 1769.6188
Límites IQR -> lim_inf: 155.2400 | lim_sup: 160.5200
Límites IQR -> lim_inf: -75.0000 | lim_sup: 125.0000
Límites IQR -> lim_inf: 42.0500 | lim_sup: 290.2700
Límites IQR -> lim_inf: -49.1437 | lim_sup: 173.8462
Límites IQR -> lim_inf: -724.6413 | lim_sup: 3002.7888
Límites IQR -> lim_inf: -487.5000 | lim_sup: 812.5000
Límites IQR -> lim_inf: -57.7125 | lim_sup: 96.1875
Límites IQR -> lim_inf: -189.0750 | lim_sup: 315.1250
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -68.2000 | lim_sup: 1268.2400
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -184.8750 | lim_sup: 308.1250
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -297.9613 | lim_sup: 1889.5087
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -64.8375 | lim_sup: 108.0625
Límites IQR -> lim_inf: -596.8688 | lim_sup: 994.7813
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -105.6500 | lim_sup: 2221.7900
Límites IQR -> lim_inf: -479.0550 | lim_sup: 1828.1850
Límites IQR -> lim_inf: 80.0000 | lim_sup: 80.0000
Límites IQR -> lim_inf: -185.2650 | lim_sup: 921.7950
Límites IQR -> lim_inf: -68.9100 | lim_sup: 756.4500
Límites IQR -> lim_inf: -376.6900 | lim_sup: 2466.7300
Límites IQR -> lim_inf: -314.3425 | lim_sup: 1616.7775
Límites IQR -> lim_inf: -104.5455 | lim_sup: 440.9091
Límites IQR -> lim_inf: -393.0832 | lim_sup: 1147.8306
Límites IQR -> lim_inf: -598.6154 | lim_sup: 997.6923
Límites IQR -> lim_inf: 81.4122 | lim_sup: 661.1743
Límites IQR -> lim_inf: -639.8491 | lim_sup: 2977.1544
Límites IQR -> lim_inf: -1594.0909 | lim_sup: 2656.8182
Límites IQR -> lim_inf: -1087.7578 | lim_sup: 3615.3343
Límites IQR -> lim_inf: 476.8801 | lim_sup: 1155.4634
Límites IQR -> lim_inf: 111.9452 | lim_sup: 2891.9159
Límites IQR -> lim_inf: -16228.6563 | lim_sup: 37479.6847
Límites IQR -> lim_inf: -32.6250 | lim_sup: 54.3750
Límites IQR -> lim_inf: 3879.7243 | lim_sup: 13888.4569
Límites IQR -> lim_inf: -4121.9771 | lim_sup: 22738.5584
Límites IQR -> lim_inf: -3244.7424 | lim_sup: 38409.6151
Límites IQR -> lim_inf: -3113.5801 | lim_sup: 9154.2309
Límites IQR -> lim_inf: 15312.1719 | lim_sup: 39048.4449
Límites IQR -> lim_inf: -54.1238 | lim_sup: 90.2063
Límites IQR -> lim_inf: 1382.1845 | lim_sup: 2670.2765
Límites IQR -> lim_inf: -1303.2129 | lim_sup: 5015.0321
Límites IQR -> lim_inf: 1432.0975 | lim_sup: 11059.4335
Límites IQR -> lim_inf: -1223.9190 | lim_sup: 2039.8650
Límites IQR -> lim_inf: 4944.2276 | lim_sup: 17072.2451
Límites IQR -> lim_inf: -75.0000 | lim_sup: 125.0000
Límites IQR -> lim_inf: 1929.4137 | lim_sup: 3594.9397
Límites IQR -> lim_inf: -2060.9314 | lim_sup: 5833.4931
Límites IQR -> lim_inf: 1231.8040 | lim_sup: 9188.9240
Límites IQR -> lim_inf: -277.9665 | lim_sup: 463.2775
Límites IQR -> lim_inf: 901.7534 | lim_sup: 17453.0479
Límites IQR -> lim_inf: -490.7500 | lim_sup: 1472.2500
Límites IQR -> lim_inf: 847.3685 | lim_sup: 2637.3205
Límites IQR -> lim_inf: -270.0967 | lim_sup: 3109.2552
Límites IQR -> lim_inf: -379.6005 | lim_sup: 11192.7595
Límites IQR -> lim_inf: -612.2562 | lim_sup: 1321.7537
Límites IQR -> lim_inf: 3293.9313 | lim_sup: 10782.7048
Límites IQR -> lim_inf: 87.0000 | lim_sup: 87.0000
Límites IQR -> lim_inf: 818.3010 | lim_sup: 2556.0770
Límites IQR -> lim_inf: -983.8793 | lim_sup: 2760.0008
Límites IQR -> lim_inf: 1032.6935 | lim_sup: 5953.2295
Límites IQR -> lim_inf: -420.6432 | lim_sup: 1227.0743
Límites IQR -> lim_inf: 1165.8639 | lim_sup: 7647.2464
Límites IQR -> lim_inf: -675.0000 | lim_sup: 1125.0000
Límites IQR -> lim_inf: 1388.4615 | lim_sup: 2944.9095
Límites IQR -> lim_inf: -43.7501 | lim_sup: 2832.9554
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -2922.6872 | lim_sup: 13339.3723
Límites IQR -> lim_inf: -1029.3187 | lim_sup: 2523.5312
Límites IQR -> lim_inf: 1978.1789 | lim_sup: 8143.0599
Límites IQR -> lim_inf: -29.9337 | lim_sup: 89.8012
Límites IQR -> lim_inf: 1687.4331 | lim_sup: 2450.4117
Límites IQR -> lim_inf: -45.1928 | lim_sup: 1352.6253
Límites IQR -> lim_inf: 144.9515 | lim_sup: 6555.9715
Límites IQR -> lim_inf: -253.5000 | lim_sup: 422.5000
Límites IQR -> lim_inf: 3520.6992 | lim_sup: 10776.9617
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 782.8240 | lim_sup: 2318.0400
Límites IQR -> lim_inf: -807.5751 | lim_sup: 2795.8769
Límites IQR -> lim_inf: 891.0318 | lim_sup: 7273.5058
Límites IQR -> lim_inf: -133.7625 | lim_sup: 222.9375
Límites IQR -> lim_inf: 5540.3007 | lim_sup: 13419.5598
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 3355.4182 | lim_sup: 6557.5868
Límites IQR -> lim_inf: -411.1585 | lim_sup: 4957.0715
Límites IQR -> lim_inf: 3912.2582 | lim_sup: 14169.1523
Límites IQR -> lim_inf: 126.5000 | lim_sup: 762.5000
Límites IQR -> lim_inf: 13759.2056 | lim_sup: 26749.0381
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 1990.6608 | lim_sup: 3966.6286
Límites IQR -> lim_inf: 112.4172 | lim_sup: 4552.1334
Límites IQR -> lim_inf: 1844.2677 | lim_sup: 9768.6538
Límites IQR -> lim_inf: -849.9307 | lim_sup: 1416.5512
Límites IQR -> lim_inf: 8319.5026 | lim_sup: 18270.2086
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -531.7321 | lim_sup: 886.2202
Límites IQR -> lim_inf: 46.3595 | lim_sup: 81.4643
Límites IQR -> lim_inf: -268.4196 | lim_sup: 447.3661
Límites IQR -> lim_inf: -473.2402 | lim_sup: 3731.0479
Límites IQR -> lim_inf: -234.8214 | lim_sup: 391.3690
Límites IQR -> lim_inf: -480.9929 | lim_sup: 1445.0690
Límites IQR -> lim_inf: -71.4286 | lim_sup: 119.0476
Límites IQR -> lim_inf: -206.6551 | lim_sup: 428.5521
Límites IQR -> lim_inf: -1436.2149 | lim_sup: 9793.1232
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -350.0000 | lim_sup: 583.3333
Límites IQR -> lim_inf: -1001.5086 | lim_sup: 2499.3747
Límites IQR -> lim_inf: -58.6857 | lim_sup: 222.1333
Límites IQR -> lim_inf: -301.2963 | lim_sup: 502.1604
Límites IQR -> lim_inf: -1330.0568 | lim_sup: 11094.4217
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 1097.0873 | lim_sup: 2911.3892
Límites IQR -> lim_inf: 207.1973 | lim_sup: 1997.0438
Límites IQR -> lim_inf: 1047.1213 | lim_sup: 6255.8032
Límites IQR -> lim_inf: -502.0800 | lim_sup: 836.8000
Límites IQR -> lim_inf: 2018.7014 | lim_sup: 10241.7134
Límites IQR -> lim_inf: -102.6786 | lim_sup: 171.1310
Límites IQR -> lim_inf: -578.1107 | lim_sup: 963.5179
Límites IQR -> lim_inf: -11.6488 | lim_sup: 165.4464
Límites IQR -> lim_inf: -151.3107 | lim_sup: 252.1845
Límites IQR -> lim_inf: -938.5533 | lim_sup: 6303.7348
Límites IQR -> lim_inf: -128.5714 | lim_sup: 214.2857
Límites IQR -> lim_inf: -222.6497 | lim_sup: 529.1003
Límites IQR -> lim_inf: -25.6500 | lim_sup: 42.7500
Límites IQR -> lim_inf: -152.6250 | lim_sup: 254.3750
Límites IQR -> lim_inf: -507.9711 | lim_sup: 3976.7313
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -575.8676 | lim_sup: 1107.2824
Límites IQR -> lim_inf: -43.1464 | lim_sup: 71.9107
Límites IQR -> lim_inf: -147.5786 | lim_sup: 245.9643
Límites IQR -> lim_inf: 931.8313 | lim_sup: 4156.5860
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 1115.5605 | lim_sup: 2443.4965
Límites IQR -> lim_inf: -379.1206 | lim_sup: 2578.1874
Límites IQR -> lim_inf: 1026.6622 | lim_sup: 7566.9043
Límites IQR -> lim_inf: -400.6733 | lim_sup: 1084.4388
Límites IQR -> lim_inf: 4099.4649 | lim_sup: 8884.7644
Límites IQR -> lim_inf: -250.0000 | lim_sup: 416.6667
Límites IQR -> lim_inf: -508.8571 | lim_sup: 1362.3810
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 99.0357 | lim_sup: 99.0357
Límites IQR -> lim_inf: -278.3741 | lim_sup: 463.9568
Límites IQR -> lim_inf: 749.4583 | lim_sup: 9738.4060
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 1653.0877 | lim_sup: 3409.1707
Límites IQR -> lim_inf: -984.4246 | lim_sup: 4521.2084
Límites IQR -> lim_inf: -330.2837 | lim_sup: 9853.6902
Límites IQR -> lim_inf: -191.5043 | lim_sup: 319.1738
Límites IQR -> lim_inf: 7220.7356 | lim_sup: 12469.6081
Límites IQR -> lim_inf: -625.0000 | lim_sup: 1041.6667
Límites IQR -> lim_inf: -1686.2208 | lim_sup: 4703.4268
Límites IQR -> lim_inf: -22.7702 | lim_sup: 244.0583
Límites IQR -> lim_inf: -572.0863 | lim_sup: 1042.3661
Límites IQR -> lim_inf: -684.8548 | lim_sup: 16325.5167
Límites IQR -> lim_inf: -594.6429 | lim_sup: 991.0714
Límites IQR -> lim_inf: -787.6161 | lim_sup: 3693.7458
Límites IQR -> lim_inf: -5.6095 | lim_sup: 233.7619
Límites IQR -> lim_inf: -625.7500 | lim_sup: 1282.0500
Límites IQR -> lim_inf: -2817.5985 | lim_sup: 10504.7372
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 1577.0925 | lim_sup: 2911.3965
Límites IQR -> lim_inf: -1530.5037 | lim_sup: 3993.5688
Límites IQR -> lim_inf: 745.0825 | lim_sup: 4168.5825
Límites IQR -> lim_inf: -219.0000 | lim_sup: 365.0000
Límites IQR -> lim_inf: 537.0208 | lim_sup: 14198.2033
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 1856.5632 | lim_sup: 3610.6813
Límites IQR -> lim_inf: -1651.7155 | lim_sup: 5665.5675
Límites IQR -> lim_inf: 1017.6397 | lim_sup: 6245.5258
Límites IQR -> lim_inf: -647.4187 | lim_sup: 1225.0312
Límites IQR -> lim_inf: 2979.3320 | lim_sup: 13280.6250
Límites IQR -> lim_inf: 311.2200 | lim_sup: 311.2200
Límites IQR -> lim_inf: 4430.5024 | lim_sup: 8673.5562
Límites IQR -> lim_inf: -1141.8087 | lim_sup: 10358.5752
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 4042.1347 | lim_sup: 17551.8927
Límites IQR -> lim_inf: -1245.8925 | lim_sup: 2076.4875
Límites IQR -> lim_inf: 16082.0059 | lim_sup: 34867.6879
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 1362.8518 | lim_sup: 2833.0137
Límites IQR -> lim_inf: -1319.2276 | lim_sup: 3994.2034
Límites IQR -> lim_inf: 67.2367 | lim_sup: 10421.3732
Límites IQR -> lim_inf: -734.6250 | lim_sup: 1224.3750
Límites IQR -> lim_inf: 3558.5551 | lim_sup: 16020.8221
Límites IQR -> lim_inf: 1450.0000 | lim_sup: 1450.0000
Límites IQR -> lim_inf: 45.2088 | lim_sup: 64.4748
Límites IQR -> lim_inf: -747.5625 | lim_sup: 1245.9375
Límites IQR -> lim_inf: 327.9425 | lim_sup: 1143.0745
Límites IQR -> lim_inf: 172.0000 | lim_sup: 172.0000
Límites IQR -> lim_inf: 313.3101 | lim_sup: 759.1261
Límites IQR -> lim_inf: -441.9130 | lim_sup: 736.5217
Límites IQR -> lim_inf: -271.8258 | lim_sup: 853.2880
Límites IQR -> lim_inf: 156.7969 | lim_sup: 395.0295
Límites IQR -> lim_inf: -148.2348 | lim_sup: 950.8838
Límites IQR -> lim_inf: 1348.4384 | lim_sup: 3719.8261
Límites IQR -> lim_inf: -240.1442 | lim_sup: 477.1635
Límites IQR -> lim_inf: -645.8165 | lim_sup: 2048.6971
Límites IQR -> lim_inf: 496.8015 | lim_sup: 496.8015
Límites IQR -> lim_inf: 203.0400 | lim_sup: 570.1145
Límites IQR -> lim_inf: 235.1444 | lim_sup: 1527.0684
Límites IQR -> lim_inf: 436.0446 | lim_sup: 13090.3946
Límites IQR -> lim_inf: -150.0000 | lim_sup: 250.0000
Límites IQR -> lim_inf: -361.3470 | lim_sup: 602.2449
Límites IQR -> lim_inf: -156.2308 | lim_sup: 260.3846
Límites IQR -> lim_inf: -232.6886 | lim_sup: 387.8144
Límites IQR -> lim_inf: 81.7881 | lim_sup: 1995.4976
Límites IQR -> lim_inf: -392.8500 | lim_sup: 654.7500
Límites IQR -> lim_inf: -456.6198 | lim_sup: 966.0293
Límites IQR -> lim_inf: 147.0738 | lim_sup: 324.1618
Límites IQR -> lim_inf: -240.3984 | lim_sup: 638.2680
Límites IQR -> lim_inf: -1293.8924 | lim_sup: 8289.4049
Límites IQR -> lim_inf: -51.7308 | lim_sup: 660.5769
Límites IQR -> lim_inf: -975.2411 | lim_sup: 2237.2565
Límites IQR -> lim_inf: 266.5169 | lim_sup: 720.8283
Límites IQR -> lim_inf: 218.2691 | lim_sup: 2226.7191
Límites IQR -> lim_inf: -2600.6975 | lim_sup: 22225.3569
Límites IQR -> lim_inf: -112.5000 | lim_sup: 187.5000
Límites IQR -> lim_inf: -317.3857 | lim_sup: 898.2101
Límites IQR -> lim_inf: 184.9662 | lim_sup: 493.8203
Límites IQR -> lim_inf: -32.6466 | lim_sup: 1096.7546
Límites IQR -> lim_inf: 1767.9101 | lim_sup: 6562.0651
Límites IQR -> lim_inf: -165.4196 | lim_sup: 521.8531
Límites IQR -> lim_inf: -2271.0004 | lim_sup: 16503.1591
Límites IQR -> lim_inf: 4688.2396 | lim_sup: 16561.8569
Límites IQR -> lim_inf: -663.3813 | lim_sup: 1105.6356
Límites IQR -> lim_inf: 14548.3093 | lim_sup: 45908.7788
Límites IQR -> lim_inf: -180.0000 | lim_sup: 300.0000
Límites IQR -> lim_inf: 1386.7423 | lim_sup: 3518.0122
Límites IQR -> lim_inf: -2207.6681 | lim_sup: 7514.8671
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -174.4425 | lim_sup: 9726.8175
Límites IQR -> lim_inf: -1046.3391 | lim_sup: 1743.8984
Límites IQR -> lim_inf: 7647.9629 | lim_sup: 16895.0409
Límites IQR -> lim_inf: -101.1189 | lim_sup: 291.6084
Límites IQR -> lim_inf: -643.4734 | lim_sup: 3075.6351
Límites IQR -> lim_inf: 2769.2167 | lim_sup: 3698.5007
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 1379.7027 | lim_sup: 12190.9488
Límites IQR -> lim_inf: -101.1189 | lim_sup: 291.6084
Límites IQR -> lim_inf: -1781.7348 | lim_sup: 5796.4597
Límites IQR -> lim_inf: 3076.7087 | lim_sup: 12175.9071
Límites IQR -> lim_inf: -32.1450 | lim_sup: 53.5750
Límites IQR -> lim_inf: 2762.2295 | lim_sup: 34240.1895
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -315.4250 | lim_sup: 706.3750
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 110.0000 | lim_sup: 110.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -1288.6137 | lim_sup: 4197.9762
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -93.2625 | lim_sup: 155.4375
Límites IQR -> lim_inf: -327.7500 | lim_sup: 546.2500
Límites IQR -> lim_inf: 110.0000 | lim_sup: 110.0000
Límites IQR -> lim_inf: 29.2800 | lim_sup: 29.2800
Límites IQR -> lim_inf: -1683.8350 | lim_sup: 4127.2050
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -45.4950 | lim_sup: 75.8250
Límites IQR -> lim_inf: -208.1025 | lim_sup: 346.8375
Límites IQR -> lim_inf: 110.0000 | lim_sup: 110.0000
Límites IQR -> lim_inf: 65.0000 | lim_sup: 65.0000
Límites IQR -> lim_inf: -1451.5075 | lim_sup: 3840.3725
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -366.2437 | lim_sup: 610.4062
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 209.5000 | lim_sup: 209.5000
Límites IQR -> lim_inf: -301.8200 | lim_sup: 1556.5800
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -57.7125 | lim_sup: 96.1875
Límites IQR -> lim_inf: -356.5987 | lim_sup: 594.3312
Límites IQR -> lim_inf: 110.0000 | lim_sup: 110.0000
Límites IQR -> lim_inf: -1144.0300 | lim_sup: 2920.1100
Límites IQR -> lim_inf: 46.1538 | lim_sup: 46.1538
Límites IQR -> lim_inf: -404.1346 | lim_sup: 673.5577
Límites IQR -> lim_inf: 357.6923 | lim_sup: 388.4615
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -665.4929 | lim_sup: 1638.1685
Límites IQR -> lim_inf: -101.1189 | lim_sup: 291.6084
Límites IQR -> lim_inf: -790.2415 | lim_sup: 1979.5531
Límites IQR -> lim_inf: 2813.5907 | lim_sup: 7864.9377
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 1425.6927 | lim_sup: 10370.5187
Límites IQR -> lim_inf: -101.1189 | lim_sup: 291.6084
Límites IQR -> lim_inf: -1150.8423 | lim_sup: 3991.4038
Límites IQR -> lim_inf: 3046.6839 | lim_sup: 4229.6936
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 3199.6594 | lim_sup: 9481.1946
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -682.5000 | lim_sup: 1137.5000
Límites IQR -> lim_inf: -431.9400 | lim_sup: 857.1000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 110.0000 | lim_sup: 110.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -191.9837 | lim_sup: 2286.0462
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -395.1975 | lim_sup: 658.6625
Límites IQR -> lim_inf: -192.0000 | lim_sup: 320.0000
Límites IQR -> lim_inf: 68.0000 | lim_sup: 68.0000
Límites IQR -> lim_inf: -489.2762 | lim_sup: 2002.3937
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -368.3663 | lim_sup: 613.9438
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -665.3762 | lim_sup: 2208.2337
Límites IQR -> lim_inf: -112.6125 | lim_sup: 517.6875
Límites IQR -> lim_inf: -78.0000 | lim_sup: 130.0000
Límites IQR -> lim_inf: -277.4775 | lim_sup: 462.4625
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -84.0713 | lim_sup: 1791.5788
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -303.3000 | lim_sup: 505.5000
Límites IQR -> lim_inf: -368.0250 | lim_sup: 613.3750
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -816.5388 | lim_sup: 3299.5113
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -422.1696 | lim_sup: 703.6161
Límites IQR -> lim_inf: -619.2940 | lim_sup: 1431.0440
Límites IQR -> lim_inf: -35.7143 | lim_sup: 59.5238
Límites IQR -> lim_inf: -244.5455 | lim_sup: 407.5759
Límites IQR -> lim_inf: 244.6000 | lim_sup: 11217.7238
Límites IQR -> lim_inf: -605.2524 | lim_sup: 1667.8146
Límites IQR -> lim_inf: -117.8481 | lim_sup: 271.7469
Límites IQR -> lim_inf: 1192.9898 | lim_sup: 5292.4215
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -461.6662 | lim_sup: 913.4438
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -388.0500 | lim_sup: 2427.6300
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -315.0000 | lim_sup: 685.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -53.6850 | lim_sup: 2800.2750
Límites IQR -> lim_inf: -509.8509 | lim_sup: 1832.3797
Límites IQR -> lim_inf: -251.6449 | lim_sup: 450.7551
Límites IQR -> lim_inf: -225.5266 | lim_sup: 7897.3500
Límites IQR -> lim_inf: 290.0000 | lim_sup: 290.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -538.3800 | lim_sup: 897.3000
Límites IQR -> lim_inf: 110.0000 | lim_sup: 110.0000
Límites IQR -> lim_inf: -408.5000 | lim_sup: 1225.5000
Límites IQR -> lim_inf: -653.2400 | lim_sup: 2886.3200
Límites IQR -> lim_inf: -90.0000 | lim_sup: 150.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -269.6250 | lim_sup: 449.3750
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -343.6137 | lim_sup: 2363.8162
Límites IQR -> lim_inf: -84.0000 | lim_sup: 140.0000
Límites IQR -> lim_inf: -1558.5000 | lim_sup: 2597.5000
Límites IQR -> lim_inf: -323.1012 | lim_sup: 3566.3487
Límites IQR -> lim_inf: 598.5100 | lim_sup: 1878.5100
Límites IQR -> lim_inf: -931.6350 | lim_sup: 1552.7250
Límites IQR -> lim_inf: -1600.5275 | lim_sup: 9320.2325
Límites IQR -> lim_inf: -711.7949 | lim_sup: 2079.8117
Límites IQR -> lim_inf: -205.4610 | lim_sup: 342.4350
Límites IQR -> lim_inf: 1187.5395 | lim_sup: 6233.8782
Límites IQR -> lim_inf: -5.6787 | lim_sup: 106.9112
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -34.1250 | lim_sup: 56.8750
Límites IQR -> lim_inf: 0.2263 | lim_sup: 54.5963
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -87.7028 | lim_sup: 831.6353
Límites IQR -> lim_inf: -429.3573 | lim_sup: 3822.7142
Límites IQR -> lim_inf: 3066.5295 | lim_sup: 8719.1015
Límites IQR -> lim_inf: -500.4563 | lim_sup: 834.0938
Límites IQR -> lim_inf: 9475.8656 | lim_sup: 21867.5636
Límites IQR -> lim_inf: 106.6455 | lim_sup: 504.7545
Límites IQR -> lim_inf: -373.6364 | lim_sup: 804.5455
Límites IQR -> lim_inf: -1026.2261 | lim_sup: 4771.2011
Límites IQR -> lim_inf: 4351.8182 | lim_sup: 5655.8182
Límites IQR -> lim_inf: -607.8886 | lim_sup: 1013.1477
Límites IQR -> lim_inf: 9577.6420 | lim_sup: 12565.8330
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -864.4050 | lim_sup: 1440.6750
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -328.9688 | lim_sup: 2422.0813
Límites IQR -> lim_inf: 149.2545 | lim_sup: 415.7273
Límites IQR -> lim_inf: 181.8182 | lim_sup: 181.8182
Límites IQR -> lim_inf: -1274.4170 | lim_sup: 5399.3193
Límites IQR -> lim_inf: -9007.0509 | lim_sup: 17250.2000
Límites IQR -> lim_inf: -463.6364 | lim_sup: 772.7273
Límites IQR -> lim_inf: 4782.1682 | lim_sup: 6759.3136
Límites IQR -> lim_inf: -13.6818 | lim_sup: 463.4091
Límites IQR -> lim_inf: 49.0909 | lim_sup: 100.0000
Límites IQR -> lim_inf: -2219.1852 | lim_sup: 5584.1330
Límites IQR -> lim_inf: 5639.3091 | lim_sup: 6364.4000
Límites IQR -> lim_inf: -284.6455 | lim_sup: 474.4091
Límites IQR -> lim_inf: 6571.5250 | lim_sup: 11372.3250
Límites IQR -> lim_inf: 280.0000 | lim_sup: 280.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -336.2212 | lim_sup: 560.3687
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -100.5237 | lim_sup: 2385.1062
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -473.3550 | lim_sup: 788.9250
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -470.2400 | lim_sup: 2392.1800
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 1703.7155 | lim_sup: 3537.8795
Límites IQR -> lim_inf: -1140.2244 | lim_sup: 3258.7001
Límites IQR -> lim_inf: 1983.8699 | lim_sup: 10250.1124
Límites IQR -> lim_inf: -47.2500 | lim_sup: 78.7500
Límites IQR -> lim_inf: 5984.0782 | lim_sup: 18494.1223
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -339.6045 | lim_sup: 566.0074
Límites IQR -> lim_inf: -899.0937 | lim_sup: 2008.9753
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -107.1429 | lim_sup: 178.5714
Límites IQR -> lim_inf: -513.8571 | lim_sup: 856.4286
Límites IQR -> lim_inf: 581.4357 | lim_sup: 1279.1167
Límites IQR -> lim_inf: 127.5500 | lim_sup: 332.9682
Límites IQR -> lim_inf: -250.0000 | lim_sup: 659.0909
Límites IQR -> lim_inf: -2077.6591 | lim_sup: 4261.4318
Límites IQR -> lim_inf: 4194.8614 | lim_sup: 5732.8432
Límites IQR -> lim_inf: -421.0023 | lim_sup: 701.6705
Límites IQR -> lim_inf: 7644.1318 | lim_sup: 9967.9318
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 868.6400 | lim_sup: 868.6400
Límites IQR -> lim_inf: -57.6562 | lim_sup: 338.5938
Límites IQR -> lim_inf: 280.0000 | lim_sup: 280.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -353.4600 | lim_sup: 589.1000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -324.4350 | lim_sup: 2036.9450
Límites IQR -> lim_inf: -2146.3950 | lim_sup: 3577.3250
Límites IQR -> lim_inf: -958.3000 | lim_sup: 2962.3400
Límites IQR -> lim_inf: -1097.2050 | lim_sup: 4614.6750
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 124.7625 | lim_sup: 1772.0625
Límites IQR -> lim_inf: -325.3312 | lim_sup: 1772.1388
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -150.0412 | lim_sup: 250.0687
Límites IQR -> lim_inf: -5404.5975 | lim_sup: 14145.0825
Límites IQR -> lim_inf: -212.2612 | lim_sup: 353.7687
Límites IQR -> lim_inf: -713.9350 | lim_sup: 2174.4250
Límites IQR -> lim_inf: -616.8500 | lim_sup: 2039.0700
Límites IQR -> lim_inf: -297.7950 | lim_sup: 2040.4850
Límites IQR -> lim_inf: -191.4850 | lim_sup: 2138.1550
Límites IQR -> lim_inf: -497.2350 | lim_sup: 1678.2450
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -392.8571 | lim_sup: 654.7619
Límites IQR -> lim_inf: -1275.0485 | lim_sup: 3353.7967
Límites IQR -> lim_inf: -28.8571 | lim_sup: 175.7714
Límites IQR -> lim_inf: -400.1149 | lim_sup: 702.4137
Límites IQR -> lim_inf: -20.8557 | lim_sup: 11415.9801
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -524.7597 | lim_sup: 2040.0463
Límites IQR -> lim_inf: 47.8786 | lim_sup: 47.8786
Límites IQR -> lim_inf: -220.4131 | lim_sup: 577.5012
Límites IQR -> lim_inf: 417.5991 | lim_sup: 9809.0539
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -630.3571 | lim_sup: 1050.5952
Límites IQR -> lim_inf: -1038.3318 | lim_sup: 3367.4610
Límites IQR -> lim_inf: -96.3000 | lim_sup: 288.1762
Límites IQR -> lim_inf: -590.0068 | lim_sup: 1224.6146
Límites IQR -> lim_inf: 795.4863 | lim_sup: 9403.5720
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -494.0476 | lim_sup: 1077.3810
Límites IQR -> lim_inf: -1485.9518 | lim_sup: 4418.5196
Límites IQR -> lim_inf: -55.5429 | lim_sup: 220.2476
Límites IQR -> lim_inf: -560.8155 | lim_sup: 1210.0417
Límites IQR -> lim_inf: 1419.3619 | lim_sup: 10711.2667
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -141.5812 | lim_sup: 235.9687
Límites IQR -> lim_inf: -598.8155 | lim_sup: 1631.1036
Límites IQR -> lim_inf: -11.9696 | lim_sup: 83.7875
Límites IQR -> lim_inf: -156.2554 | lim_sup: 374.7113
Límites IQR -> lim_inf: -588.0509 | lim_sup: 6137.9182
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -207.1429 | lim_sup: 345.2381
Límites IQR -> lim_inf: -535.4149 | lim_sup: 1337.6232
Límites IQR -> lim_inf: -71.8179 | lim_sup: 119.6964
Límites IQR -> lim_inf: -373.8920 | lim_sup: 623.1533
Límites IQR -> lim_inf: -857.1613 | lim_sup: 7253.6673
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -178.5714 | lim_sup: 297.6190
Límites IQR -> lim_inf: -538.2012 | lim_sup: 1361.4750
Límites IQR -> lim_inf: -71.8179 | lim_sup: 119.6964
Límites IQR -> lim_inf: -229.6482 | lim_sup: 382.7470
Límites IQR -> lim_inf: 47.4565 | lim_sup: 5431.7470
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -564.2857 | lim_sup: 940.4762
Límites IQR -> lim_inf: -862.8833 | lim_sup: 2392.1024
Límites IQR -> lim_inf: -11.9696 | lim_sup: 83.7875
Límites IQR -> lim_inf: -196.2146 | lim_sup: 476.9116
Límites IQR -> lim_inf: 192.5571 | lim_sup: 6978.8714
Límites IQR -> lim_inf: -107.7804 | lim_sup: 179.6339
Límites IQR -> lim_inf: -850.9670 | lim_sup: 1418.2783
Límites IQR -> lim_inf: -75.6643 | lim_sup: 126.1071
Límites IQR -> lim_inf: -165.2688 | lim_sup: 275.4479
Límites IQR -> lim_inf: 876.6982 | lim_sup: 4624.7792
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -464.2050 | lim_sup: 2241.4550
Límites IQR -> lim_inf: -583.4787 | lim_sup: 2036.5313
Límites IQR -> lim_inf: -705.8113 | lim_sup: 2488.8988
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -423.6300 | lim_sup: 706.0500
Límites IQR -> lim_inf: -307.2250 | lim_sup: 1965.2950
Límites IQR -> lim_inf: -28.1250 | lim_sup: 46.8750
Límites IQR -> lim_inf: -126.0000 | lim_sup: 210.0000
Límites IQR -> lim_inf: -160.0275 | lim_sup: 2094.8725
Límites IQR -> lim_inf: -427.0512 | lim_sup: 1996.3787
Límites IQR -> lim_inf: -1481.5750 | lim_sup: 4965.1650
Límites IQR -> lim_inf: -1598.7588 | lim_sup: 3280.1713
Límites IQR -> lim_inf: -405.3575 | lim_sup: 921.0425
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -142.1250 | lim_sup: 236.8750
Límites IQR -> lim_inf: 68.0000 | lim_sup: 68.0000
Límites IQR -> lim_inf: -351.4988 | lim_sup: 1963.8713
Límites IQR -> lim_inf: 50.0000 | lim_sup: 50.0000
Límites IQR -> lim_inf: -146.0250 | lim_sup: 243.3750
Límites IQR -> lim_inf: -423.6638 | lim_sup: 2430.0463
Límites IQR -> lim_inf: -1046.1800 | lim_sup: 3225.5600
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -156.1800 | lim_sup: 260.3000
Límites IQR -> lim_inf: -296.8500 | lim_sup: 645.3100
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -1066.3387 | lim_sup: 3901.0312
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -66.8850 | lim_sup: 111.4750
Límites IQR -> lim_inf: -157.0500 | lim_sup: 261.7500
Límites IQR -> lim_inf: 110.0000 | lim_sup: 110.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -771.4463 | lim_sup: 4061.9038
Límites IQR -> lim_inf: -66.1000 | lim_sup: 306.2800
Límites IQR -> lim_inf: -538.9900 | lim_sup: 2348.1300
Límites IQR -> lim_inf: -1561.0213 | lim_sup: 3873.1687
Límites IQR -> lim_inf: -157.1000 | lim_sup: 614.3200
Límites IQR -> lim_inf: -954.5750 | lim_sup: 3043.7850
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -396.3750 | lim_sup: 660.6250
Límites IQR -> lim_inf: 55.7600 | lim_sup: 761.6000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -127.3862 | lim_sup: 688.2037
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -208.5000 | lim_sup: 347.5000
Límites IQR -> lim_inf: -926.7375 | lim_sup: 1544.5625
Límites IQR -> lim_inf: 624.2000 | lim_sup: 624.2000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -353.8575 | lim_sup: 1618.6625
Límites IQR -> lim_inf: 214.0000 | lim_sup: 214.0000
Límites IQR -> lim_inf: -916.7250 | lim_sup: 1527.8750
Límites IQR -> lim_inf: -101.7763 | lim_sup: 587.3138
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -439.1063 | lim_sup: 731.8438
Límites IQR -> lim_inf: -164.4675 | lim_sup: 1161.5525
Límites IQR -> lim_inf: -336.8363 | lim_sup: 561.3937
Límites IQR -> lim_inf: -528.9150 | lim_sup: 881.5250
Límites IQR -> lim_inf: -302.8825 | lim_sup: 1880.3375
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -67.6500 | lim_sup: 112.7500
Límites IQR -> lim_inf: -298.1250 | lim_sup: 535.6750
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -846.1987 | lim_sup: 1540.3312
Límites IQR -> lim_inf: -685.6825 | lim_sup: 2415.1375
Límites IQR -> lim_inf: 127.5500 | lim_sup: 332.9682
Límites IQR -> lim_inf: -373.6364 | lim_sup: 804.5455
Límites IQR -> lim_inf: -2168.5216 | lim_sup: 5656.4693
Límites IQR -> lim_inf: -7635.6364 | lim_sup: 13498.9091
Límites IQR -> lim_inf: -255.6818 | lim_sup: 426.1364
Límites IQR -> lim_inf: 11468.9705 | lim_sup: 18900.6977
Límites IQR -> lim_inf: -487.5000 | lim_sup: 812.5000
Límites IQR -> lim_inf: -54.1200 | lim_sup: 90.2000
Límites IQR -> lim_inf: -100.3988 | lim_sup: 167.3313
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -455.3863 | lim_sup: 2777.1638
Límites IQR -> lim_inf: -496.5375 | lim_sup: 827.5625
Límites IQR -> lim_inf: -57.7125 | lim_sup: 96.1875
Límites IQR -> lim_inf: -248.6812 | lim_sup: 414.4688
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -313.6313 | lim_sup: 522.7188
Límites IQR -> lim_inf: -934.6000 | lim_sup: 3049.3200
Límites IQR -> lim_inf: -902.5950 | lim_sup: 2723.6850
Límites IQR -> lim_inf: -1483.8500 | lim_sup: 3841.2700
Límites IQR -> lim_inf: -1515.0700 | lim_sup: 4175.8900
Límites IQR -> lim_inf: -1436.9800 | lim_sup: 4222.6200
Límites IQR -> lim_inf: 7264.5945 | lim_sup: 7264.5945
Límites IQR -> lim_inf: 1700.0000 | lim_sup: 1700.0000
Límites IQR -> lim_inf: 1266.1182 | lim_sup: 3819.3364
Límites IQR -> lim_inf: -1070.6317 | lim_sup: 8336.3698
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 3066.6022 | lim_sup: 18301.5043
Límites IQR -> lim_inf: 255.2345 | lim_sup: 3605.7465
Límites IQR -> lim_inf: 9254.0573 | lim_sup: 32625.3933
Límites IQR -> lim_inf: 233.1300 | lim_sup: 973.1300
Límites IQR -> lim_inf: 182.6450 | lim_sup: 1553.4050
Límites IQR -> lim_inf: 1120.8275 | lim_sup: 1778.8075
Límites IQR -> lim_inf: -2363.7150 | lim_sup: 3939.5250
Límites IQR -> lim_inf: 3.3863 | lim_sup: 997.0162
Límites IQR -> lim_inf: 1072.6850 | lim_sup: 1499.9650
Límites IQR -> lim_inf: -544.6738 | lim_sup: 1126.6562
Límites IQR -> lim_inf: -1904.4250 | lim_sup: 5115.2950
Límites IQR -> lim_inf: -907.8900 | lim_sup: 2639.9500
Límites IQR -> lim_inf: -101.1400 | lim_sup: 3397.5200
Límites IQR -> lim_inf: -144.2308 | lim_sup: 240.3846
Límites IQR -> lim_inf: -334.5577 | lim_sup: 557.5962
Límites IQR -> lim_inf: -1143.0158 | lim_sup: 2184.3879
Límites IQR -> lim_inf: -75.5466 | lim_sup: 547.9264
Límites IQR -> lim_inf: 113.3632 | lim_sup: 2064.0101
Límites IQR -> lim_inf: -650.3625 | lim_sup: 2186.4975
Límites IQR -> lim_inf: -671.0800 | lim_sup: 2072.8400
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -283.8475 | lim_sup: 2488.6125
Límites IQR -> lim_inf: -565.2500 | lim_sup: 2918.4500
Límites IQR -> lim_inf: -128.6014 | lim_sup: 460.4895
Límites IQR -> lim_inf: -4645.5143 | lim_sup: 13933.5710
Límites IQR -> lim_inf: 342.3077 | lim_sup: 373.0769
Límites IQR -> lim_inf: -150.0000 | lim_sup: 250.0000
Límites IQR -> lim_inf: 560.6108 | lim_sup: 48201.5054
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -341.2500 | lim_sup: 568.7500
Límites IQR -> lim_inf: -473.3625 | lim_sup: 788.9375
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -130.1400 | lim_sup: 216.9000
Límites IQR -> lim_inf: -359.1137 | lim_sup: 1623.0963
Límites IQR -> lim_inf: -0.3465 | lim_sup: 0.5775
Límites IQR -> lim_inf: -22.5000 | lim_sup: 37.5000
Límites IQR -> lim_inf: -198.7500 | lim_sup: 331.2500
Límites IQR -> lim_inf: -412.2525 | lim_sup: 2248.2875
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -261.7950 | lim_sup: 436.3250
Límites IQR -> lim_inf: 315.8900 | lim_sup: 315.8900
Límites IQR -> lim_inf: -773.8700 | lim_sup: 3828.4100
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -251.3850 | lim_sup: 418.9750
Límites IQR -> lim_inf: -698.8500 | lim_sup: 1164.7500
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -192.0000 | lim_sup: 320.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -1349.4500 | lim_sup: 3472.7500
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -266.1750 | lim_sup: 443.6250
Límites IQR -> lim_inf: -196.5525 | lim_sup: 327.5875
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -268.6800 | lim_sup: 447.8000
Límites IQR -> lim_inf: -332.1337 | lim_sup: 553.5562
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -1019.4962 | lim_sup: 4522.5737
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -361.9950 | lim_sup: 603.3250
Límites IQR -> lim_inf: -89.0437 | lim_sup: 148.4062
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -599.5962 | lim_sup: 2474.6937
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -177.4350 | lim_sup: 295.7250
Límites IQR -> lim_inf: -140.8200 | lim_sup: 234.7000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 121.7300 | lim_sup: 121.7300
Límites IQR -> lim_inf: -348.5975 | lim_sup: 1588.0625
Límites IQR -> lim_inf: -1281.0000 | lim_sup: 2135.0000
Límites IQR -> lim_inf: -177.4350 | lim_sup: 295.7250
Límites IQR -> lim_inf: -253.5863 | lim_sup: 422.6438
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -948.0725 | lim_sup: 2625.9875
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -641.5200 | lim_sup: 1069.2000
Límites IQR -> lim_inf: -172.3425 | lim_sup: 287.2375
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 43.8187 | lim_sup: 47.8688
Límites IQR -> lim_inf: 2749.4400 | lim_sup: 2749.4400
Límites IQR -> lim_inf: -511.0100 | lim_sup: 3636.1700
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -375.0000 | lim_sup: 625.0000
Límites IQR -> lim_inf: -438.4575 | lim_sup: 730.7625
Límites IQR -> lim_inf: -175.0000 | lim_sup: 525.0000
Límites IQR -> lim_inf: 103.0000 | lim_sup: 103.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -1492.8162 | lim_sup: 4426.9337
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -221.7900 | lim_sup: 369.6500
Límites IQR -> lim_inf: -148.5000 | lim_sup: 247.5000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -18.7500 | lim_sup: 31.2500
Límites IQR -> lim_inf: -952.2013 | lim_sup: 2655.9688
Límites IQR -> lim_inf: 650.0000 | lim_sup: 650.0000
Límites IQR -> lim_inf: -341.9400 | lim_sup: 569.9000
Límites IQR -> lim_inf: -148.5975 | lim_sup: 247.6625
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -41.2500 | lim_sup: 68.7500
Límites IQR -> lim_inf: -223.3500 | lim_sup: 372.2500
Límites IQR -> lim_inf: 103.0000 | lim_sup: 103.0000
Límites IQR -> lim_inf: -25.5000 | lim_sup: 42.5000
Límites IQR -> lim_inf: -687.1450 | lim_sup: 2693.1550
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -646.4175 | lim_sup: 1077.3625
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 103.0000 | lim_sup: 103.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 387.1100 | lim_sup: 1533.6700
Límites IQR -> lim_inf: -515.8900 | lim_sup: 1704.7100
Límites IQR -> lim_inf: 2067.0100 | lim_sup: 2067.0100
Límites IQR -> lim_inf: 1913.0400 | lim_sup: 2069.0400
Límites IQR -> lim_inf: -109.7850 | lim_sup: 690.1950
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -479.4000 | lim_sup: 799.0000
Límites IQR -> lim_inf: -417.1350 | lim_sup: 921.2250
Límites IQR -> lim_inf: -908.5050 | lim_sup: 1514.1750
Límites IQR -> lim_inf: -45.0000 | lim_sup: 75.0000
Límites IQR -> lim_inf: -1559.3125 | lim_sup: 6381.0875
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -180.0000 | lim_sup: 300.0000
Límites IQR -> lim_inf: -105.9375 | lim_sup: 176.5625
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -434.6600 | lim_sup: 2085.3800
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -153.3375 | lim_sup: 255.5625
Límites IQR -> lim_inf: -104.5875 | lim_sup: 174.3125
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -205.0813 | lim_sup: 1555.0888
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -202.5000 | lim_sup: 337.5000
Límites IQR -> lim_inf: -278.5425 | lim_sup: 464.2375
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -455.3037 | lim_sup: 1980.2662
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -202.7062 | lim_sup: 337.8438
Límites IQR -> lim_inf: -473.0325 | lim_sup: 788.3875
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 33.9500 | lim_sup: 33.9500
Límites IQR -> lim_inf: -344.3487 | lim_sup: 1912.4212
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -197.7113 | lim_sup: 329.5188
Límites IQR -> lim_inf: -157.3575 | lim_sup: 262.2625
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -353.8538 | lim_sup: 1832.1562
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -202.7062 | lim_sup: 337.8438
Límites IQR -> lim_inf: -549.2250 | lim_sup: 915.3750
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -333.4500 | lim_sup: 555.7500
Límites IQR -> lim_inf: -433.2150 | lim_sup: 722.0250
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -96.0000 | lim_sup: 160.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -557.2738 | lim_sup: 2493.0363
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -153.7500 | lim_sup: 256.2500
Límites IQR -> lim_inf: -431.0325 | lim_sup: 718.3875
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -412.5000 | lim_sup: 687.5000
Límites IQR -> lim_inf: 208.0000 | lim_sup: 208.0000
Límites IQR -> lim_inf: -314.8125 | lim_sup: 524.6875
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 155.6300 | lim_sup: 155.6300
Límites IQR -> lim_inf: 25.5000 | lim_sup: 93.5000
Límites IQR -> lim_inf: 359.4700 | lim_sup: 1414.8300
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -286.5000 | lim_sup: 477.5000
Límites IQR -> lim_inf: -251.1825 | lim_sup: 418.6375
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -63.6938 | lim_sup: 106.1563
Límites IQR -> lim_inf: -78.3675 | lim_sup: 1276.4925
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -203.3363 | lim_sup: 338.8938
Límites IQR -> lim_inf: -5.2500 | lim_sup: 8.7500
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -291.6900 | lim_sup: 2202.5100
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -187.0312 | lim_sup: 311.7188
Límites IQR -> lim_inf: -118.2900 | lim_sup: 197.1500
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -1802.8787 | lim_sup: 4659.1912
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -293.0775 | lim_sup: 488.4625
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -437.0887 | lim_sup: 2100.4012
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -161.2500 | lim_sup: 268.7500
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -223.1362 | lim_sup: 371.8937
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 154.0000 | lim_sup: 154.0000
Límites IQR -> lim_inf: -128.7325 | lim_sup: 1536.0275
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -153.7500 | lim_sup: 256.2500
Límites IQR -> lim_inf: -247.5000 | lim_sup: 412.5000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -558.8213 | lim_sup: 2041.3088
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -104.0625 | lim_sup: 173.4375
Límites IQR -> lim_inf: -152.0362 | lim_sup: 253.3938
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -267.0438 | lim_sup: 1597.8063
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -198.7500 | lim_sup: 331.2500
Límites IQR -> lim_inf: -103.6875 | lim_sup: 172.8125
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -404.8725 | lim_sup: 1929.5475
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -217.6687 | lim_sup: 362.7812
Límites IQR -> lim_inf: -220.1850 | lim_sup: 366.9750
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -367.3188 | lim_sup: 2443.8913
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -195.0000 | lim_sup: 325.0000
Límites IQR -> lim_inf: -376.8750 | lim_sup: 628.1250
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -224.5050 | lim_sup: 1894.0750
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -204.6112 | lim_sup: 341.0187
Límites IQR -> lim_inf: -182.3850 | lim_sup: 303.9750
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 160.0000 | lim_sup: 160.0000
Límites IQR -> lim_inf: -241.3288 | lim_sup: 1984.9013
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -186.7500 | lim_sup: 311.2500
Límites IQR -> lim_inf: -135.0000 | lim_sup: 225.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -226.3862 | lim_sup: 1393.2437
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -112.5000 | lim_sup: 187.5000
Límites IQR -> lim_inf: -234.6825 | lim_sup: 391.1375
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -83.3563 | lim_sup: 1421.8338
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -110.1750 | lim_sup: 183.6250
Límites IQR -> lim_inf: -423.0000 | lim_sup: 705.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -298.5188 | lim_sup: 2406.2712
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -108.3375 | lim_sup: 180.5625
Límites IQR -> lim_inf: -533.4000 | lim_sup: 889.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -76.4625 | lim_sup: 127.4375
Límites IQR -> lim_inf: -415.7838 | lim_sup: 2276.6063
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -732.3900 | lim_sup: 1220.6500
Límites IQR -> lim_inf: -550.5337 | lim_sup: 917.5562
Límites IQR -> lim_inf: 1128.7200 | lim_sup: 1590.7200
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -1344.7813 | lim_sup: 5110.3888
Límites IQR -> lim_inf: -537.7875 | lim_sup: 896.3125
Límites IQR -> lim_inf: -767.5800 | lim_sup: 1279.3000
Límites IQR -> lim_inf: -318.0375 | lim_sup: 530.0625
Límites IQR -> lim_inf: 1428.1150 | lim_sup: 2012.6350
Límites IQR -> lim_inf: -39.9000 | lim_sup: 66.5000
Límites IQR -> lim_inf: -1561.3338 | lim_sup: 6483.5763
Límites IQR -> lim_inf: -151.5000 | lim_sup: 252.5000
Límites IQR -> lim_inf: -1250.8200 | lim_sup: 2084.7000
Límites IQR -> lim_inf: -981.1550 | lim_sup: 1859.7050
Límites IQR -> lim_inf: 749.3550 | lim_sup: 5917.6350
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -2851.6837 | lim_sup: 9269.4862
Límites IQR -> lim_inf: -54.1687 | lim_sup: 90.2812
Límites IQR -> lim_inf: -1140.0600 | lim_sup: 1900.1000
Límites IQR -> lim_inf: -462.3525 | lim_sup: 1376.0675
Límites IQR -> lim_inf: 775.2200 | lim_sup: 5114.2600
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -4367.9950 | lim_sup: 11207.0450
Límites IQR -> lim_inf: -817.1625 | lim_sup: 1361.9375
Límites IQR -> lim_inf: -720.0000 | lim_sup: 1200.0000
Límites IQR -> lim_inf: -442.0500 | lim_sup: 747.5100
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -766.5525 | lim_sup: 1277.5875
Límites IQR -> lim_inf: 79.8000 | lim_sup: 407.0800
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -253.0250 | lim_sup: 3455.1750
Límites IQR -> lim_inf: -1038.3250 | lim_sup: 3622.1150
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 177.0000 | lim_sup: 273.0000
Límites IQR -> lim_inf: -731.6250 | lim_sup: 1219.3750
Límites IQR -> lim_inf: -9415.5788 | lim_sup: 47745.1113
Límites IQR -> lim_inf: -1258.1175 | lim_sup: 2096.8625
Límites IQR -> lim_inf: -1207.0850 | lim_sup: 3717.3550
Límites IQR -> lim_inf: -339.6100 | lim_sup: 932.2300
Límites IQR -> lim_inf: 180.0000 | lim_sup: 180.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -392.8575 | lim_sup: 654.7625
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -1009.8200 | lim_sup: 3879.5400
Límites IQR -> lim_inf: 30.0000 | lim_sup: 30.0000
Límites IQR -> lim_inf: -154.4925 | lim_sup: 257.4875
Límites IQR -> lim_inf: -387.6212 | lim_sup: 1770.7887
Límites IQR -> lim_inf: -88.1250 | lim_sup: 146.8750
Límites IQR -> lim_inf: -243.4200 | lim_sup: 405.7000
Límites IQR -> lim_inf: -348.0950 | lim_sup: 1723.2050
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -351.1837 | lim_sup: 671.6063
Límites IQR -> lim_inf: -112.0000 | lim_sup: 272.0000
Límites IQR -> lim_inf: -117.5550 | lim_sup: 195.9250
Límites IQR -> lim_inf: 20.1591 | lim_sup: 113.2864
Límites IQR -> lim_inf: 31.8182 | lim_sup: 68.1818
Límites IQR -> lim_inf: -1894.9864 | lim_sup: 4263.3591
Límites IQR -> lim_inf: 4731.7000 | lim_sup: 4942.6091
Límites IQR -> lim_inf: -201.1364 | lim_sup: 335.2273
Límites IQR -> lim_inf: 3994.5727 | lim_sup: 11748.4091
Límites IQR -> lim_inf: -95.6250 | lim_sup: 159.3750
Límites IQR -> lim_inf: -278.3375 | lim_sup: 1060.0025
Límites IQR -> lim_inf: -727.4625 | lim_sup: 1212.4375
Límites IQR -> lim_inf: -1203.9637 | lim_sup: 2006.6062
Límites IQR -> lim_inf: 1673.9550 | lim_sup: 3540.3150
Límites IQR -> lim_inf: -22.5000 | lim_sup: 37.5000
Límites IQR -> lim_inf: -1759.5525 | lim_sup: 6617.2275
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -19.5375 | lim_sup: 32.5625
Límites IQR -> lim_inf: -731.2050 | lim_sup: 1218.6750
Límites IQR -> lim_inf: -739.4250 | lim_sup: 1232.3750
Límites IQR -> lim_inf: 254.3400 | lim_sup: 1053.1400
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -247.1025 | lim_sup: 1699.6775
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -294.6175 | lim_sup: 1063.3625
Límites IQR -> lim_inf: 49.5979 | lim_sup: 2298.3334
Límites IQR -> lim_inf: 806.1457 | lim_sup: 4406.9862
Límites IQR -> lim_inf: -839.1525 | lim_sup: 1943.4915
Límites IQR -> lim_inf: 3198.3758 | lim_sup: 11638.6387
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 245.1885 | lim_sup: 342.6125
Límites IQR -> lim_inf: -394.3286 | lim_sup: 2369.5069
Límites IQR -> lim_inf: 28.1225 | lim_sup: 8729.0545
Límites IQR -> lim_inf: -446.6644 | lim_sup: 1506.3446
Límites IQR -> lim_inf: 1560.2642 | lim_sup: 13782.2897
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -243.6288 | lim_sup: 978.3813
Límites IQR -> lim_inf: -769.7693 | lim_sup: 2202.3112
Límites IQR -> lim_inf: 968.0995 | lim_sup: 5928.3675
Límites IQR -> lim_inf: -113.4454 | lim_sup: 1326.0541
Límites IQR -> lim_inf: 6952.6608 | lim_sup: 21268.0068
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 191.5165 | lim_sup: 191.5165
Límites IQR -> lim_inf: -1655.7702 | lim_sup: 4049.4837
Límites IQR -> lim_inf: -5340.0633 | lim_sup: 9292.5322
Límites IQR -> lim_inf: -837.1111 | lim_sup: 1589.8519
Límites IQR -> lim_inf: 6282.4243 | lim_sup: 12222.8558
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -259.5067 | lim_sup: 1291.0673
Límites IQR -> lim_inf: -1085.5642 | lim_sup: 5107.2143
Límites IQR -> lim_inf: 2817.1608 | lim_sup: 8787.6133
Límites IQR -> lim_inf: -630.5403 | lim_sup: 2703.5802
Límites IQR -> lim_inf: 4748.7657 | lim_sup: 27103.5512
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 67.6988 | lim_sup: 1145.0648
Límites IQR -> lim_inf: -1317.1004 | lim_sup: 3455.9006
Límites IQR -> lim_inf: 2325.0285 | lim_sup: 6969.1445
Límites IQR -> lim_inf: -882.9803 | lim_sup: 2069.3672
Límites IQR -> lim_inf: 750.5172 | lim_sup: 28192.7587
Límites IQR -> lim_inf: 3541.9845 | lim_sup: 3541.9845
Límites IQR -> lim_inf: -273.7500 | lim_sup: 456.2500
Límites IQR -> lim_inf: 1339.4060 | lim_sup: 2542.1949
Límites IQR -> lim_inf: -1306.3912 | lim_sup: 5928.1453
Límites IQR -> lim_inf: -1063.0008 | lim_sup: 12333.2293
Límites IQR -> lim_inf: -2059.1860 | lim_sup: 4511.1440
Límites IQR -> lim_inf: -24788.1357 | lim_sup: 50125.0538
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -270.0000 | lim_sup: 450.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -39.9925 | lim_sup: 1496.5475
Límites IQR -> lim_inf: -163.4850 | lim_sup: 272.4750
Límites IQR -> lim_inf: 24.4300 | lim_sup: 49.1500
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -67.4525 | lim_sup: 203.8075
Límites IQR -> lim_inf: -227.7650 | lim_sup: 349.7350
Límites IQR -> lim_inf: -10.1250 | lim_sup: 16.8750
Límites IQR -> lim_inf: 2842.9500 | lim_sup: 2842.9500
Límites IQR -> lim_inf: 20.0750 | lim_sup: 29.7950
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -50.4950 | lim_sup: 327.1050
Límites IQR -> lim_inf: -202.5000 | lim_sup: 337.5000
Límites IQR -> lim_inf: 155.6300 | lim_sup: 155.6300
Límites IQR -> lim_inf: 309.5700 | lim_sup: 1716.7700
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 117.4000 | lim_sup: 117.4000
Límites IQR -> lim_inf: -844.4962 | lim_sup: 1407.4937
Límites IQR -> lim_inf: -974.7413 | lim_sup: 1981.6888
Límites IQR -> lim_inf: -4427.5950 | lim_sup: 10416.7850
Límites IQR -> lim_inf: 1875.2250 | lim_sup: 2657.2650
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -419.7350 | lim_sup: 2474.2450
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -869.1488 | lim_sup: 1448.5813
Límites IQR -> lim_inf: -96.0000 | lim_sup: 160.0000
Límites IQR -> lim_inf: -579.4000 | lim_sup: 4899.8000
Límites IQR -> lim_inf: -629.2975 | lim_sup: 2906.8225
Límites IQR -> lim_inf: -0.9350 | lim_sup: 49.2050
Límites IQR -> lim_inf: 276.9231 | lim_sup: 276.9231
Límites IQR -> lim_inf: -1194.7328 | lim_sup: 1991.2213
Límites IQR -> lim_inf: 136.4850 | lim_sup: 235.5850
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -594.3262 | lim_sup: 990.5437
Límites IQR -> lim_inf: -149.9000 | lim_sup: 785.7000
Límites IQR -> lim_inf: -76.5000 | lim_sup: 127.5000
Límites IQR -> lim_inf: 61.4818 | lim_sup: 233.6273
Límites IQR -> lim_inf: -2347.5500 | lim_sup: 4128.3409
Límites IQR -> lim_inf: -1012.3636 | lim_sup: 2353.6000
Límites IQR -> lim_inf: 4093.9023 | lim_sup: 5677.7750
Límites IQR -> lim_inf: -774.9955 | lim_sup: 1291.6591
Límites IQR -> lim_inf: 8030.4977 | lim_sup: 14582.1523
Límites IQR -> lim_inf: 80.9091 | lim_sup: 80.9091
Límites IQR -> lim_inf: 645.9909 | lim_sup: 645.9909
Límites IQR -> lim_inf: 3551.1545 | lim_sup: 3551.1545
Límites IQR -> lim_inf: 2336.5273 | lim_sup: 2336.5273
Límites IQR -> lim_inf: 54.5455 | lim_sup: 54.5455
Límites IQR -> lim_inf: -1035.2614 | lim_sup: 2257.0477
Límites IQR -> lim_inf: 4607.5182 | lim_sup: 5062.0636
Límites IQR -> lim_inf: 40.4773 | lim_sup: 566.1136
Límites IQR -> lim_inf: 3242.8591 | lim_sup: 23211.7682
Límites IQR -> lim_inf: -103.5675 | lim_sup: 505.0325
Límites IQR -> lim_inf: -322.4608 | lim_sup: 919.0714
Límites IQR -> lim_inf: -31.8250 | lim_sup: 115.3350
Límites IQR -> lim_inf: 314.9256 | lim_sup: 7002.8304
Límites IQR -> lim_inf: -1226.8387 | lim_sup: 2781.5289
Límites IQR -> lim_inf: -105.4750 | lim_sup: 240.3250
Límites IQR -> lim_inf: 770.0666 | lim_sup: 2905.6368
Límites IQR -> lim_inf: 89.5664 | lim_sup: 122.1360
Límites IQR -> lim_inf: -989.5391 | lim_sup: 2677.7303
Límites IQR -> lim_inf: -104.7200 | lim_sup: 252.5600
Límites IQR -> lim_inf: -108.4932 | lim_sup: 6666.5634
Límites IQR -> lim_inf: -922.0667 | lim_sup: 2402.3205
Límites IQR -> lim_inf: -112.5000 | lim_sup: 187.5000
Límites IQR -> lim_inf: 205.1568 | lim_sup: 3991.8556
Límites IQR -> lim_inf: -1314.5285 | lim_sup: 4530.3547
Límites IQR -> lim_inf: -271.7750 | lim_sup: 511.6250
Límites IQR -> lim_inf: 2447.0511 | lim_sup: 9790.5464
Límites IQR -> lim_inf: -371.0077 | lim_sup: 1013.7315
Límites IQR -> lim_inf: -38.0200 | lim_sup: 116.7000
Límites IQR -> lim_inf: 180.8981 | lim_sup: 4803.1481
Límites IQR -> lim_inf: 61.4818 | lim_sup: 233.6273
Límites IQR -> lim_inf: -159.8045 | lim_sup: 266.3409
Límites IQR -> lim_inf: -392.5466 | lim_sup: 1139.8443
Límites IQR -> lim_inf: -1101.9091 | lim_sup: 9059.1091
Límites IQR -> lim_inf: -483.5045 | lim_sup: 805.8409
Límites IQR -> lim_inf: 5753.7636 | lim_sup: 9234.5636
Límites IQR -> lim_inf: 128.8909 | lim_sup: 551.6545
Límites IQR -> lim_inf: -1313.3455 | lim_sup: 2325.2727
Límites IQR -> lim_inf: -2858.1977 | lim_sup: 6160.4750
Límites IQR -> lim_inf: -4779.6750 | lim_sup: 13595.0341
Límites IQR -> lim_inf: -347.7273 | lim_sup: 579.5455
Límites IQR -> lim_inf: 14611.0420 | lim_sup: 19798.5784
Límites IQR -> lim_inf: -96.0858 | lim_sup: 160.1430
Límites IQR -> lim_inf: -123.6117 | lim_sup: 206.0194
Límites IQR -> lim_inf: -7.8254 | lim_sup: 423.1962
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -798.0000 | lim_sup: 1330.0000
Límites IQR -> lim_inf: 381.5000 | lim_sup: 617.5000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -73.7663 | lim_sup: 476.1438
Límites IQR -> lim_inf: -2504.1400 | lim_sup: 6870.9000
Límites IQR -> lim_inf: -837.7000 | lim_sup: 2293.9800
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 299.0200 | lim_sup: 299.0200
Límites IQR -> lim_inf: -1144.6063 | lim_sup: 2436.7238
Límites IQR -> lim_inf: 144.2250 | lim_sup: 526.2650
Límites IQR -> lim_inf: -191.7300 | lim_sup: 319.5500
Límites IQR -> lim_inf: -1412.8550 | lim_sup: 10568.3850
Límites IQR -> lim_inf: 472.7762 | lim_sup: 1045.4462
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 144.9500 | lim_sup: 144.9500
Límites IQR -> lim_inf: -456.6112 | lim_sup: 761.0187
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 26.8175 | lim_sup: 104.6375
Límites IQR -> lim_inf: 80.9091 | lim_sup: 80.9091
Límites IQR -> lim_inf: -1200.2375 | lim_sup: 2835.4261
Límites IQR -> lim_inf: 4087.6364 | lim_sup: 4451.2727
Límites IQR -> lim_inf: -229.8409 | lim_sup: 1020.8864
Límites IQR -> lim_inf: -761.2000 | lim_sup: 29313.3455
Límites IQR -> lim_inf: 64.0900 | lim_sup: 94.2500
Límites IQR -> lim_inf: -226.9590 | lim_sup: 1750.9450
Límites IQR -> lim_inf: -47.9500 | lim_sup: 133.2500
Límites IQR -> lim_inf: 3575.2589 | lim_sup: 8589.7510
Límites IQR -> lim_inf: 103.5100 | lim_sup: 389.1100
Límites IQR -> lim_inf: 136.3200 | lim_sup: 465.2800
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -182.9775 | lim_sup: 304.9625
Límites IQR -> lim_inf: -338.0625 | lim_sup: 673.0375
Límites IQR -> lim_inf: 64.0000 | lim_sup: 64.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 44.3275 | lim_sup: 2751.4275
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -270.7200 | lim_sup: 451.2000
Límites IQR -> lim_inf: -569.6213 | lim_sup: 949.3688
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 269.0000 | lim_sup: 269.0000
Límites IQR -> lim_inf: -120.2925 | lim_sup: 200.4875
Límites IQR -> lim_inf: -412.5000 | lim_sup: 687.5000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 68.0000 | lim_sup: 68.0000
Límites IQR -> lim_inf: -137.2500 | lim_sup: 228.7500
Límites IQR -> lim_inf: -230.9625 | lim_sup: 432.9375
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -901.5000 | lim_sup: 1502.5000
Límites IQR -> lim_inf: 4551.4000 | lim_sup: 4551.4000
Límites IQR -> lim_inf: 368.0100 | lim_sup: 540.7300
Límites IQR -> lim_inf: -1048.7250 | lim_sup: 3278.8750
Límites IQR -> lim_inf: -60.0000 | lim_sup: 100.0000
Límites IQR -> lim_inf: -819.1950 | lim_sup: 1365.3250
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -117.7500 | lim_sup: 196.2500
Límites IQR -> lim_inf: -60.0000 | lim_sup: 100.0000
Límites IQR -> lim_inf: -522.9900 | lim_sup: 871.6500
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 284.0000 | lim_sup: 284.0000
Límites IQR -> lim_inf: -92.2500 | lim_sup: 153.7500
Límites IQR -> lim_inf: -339.5938 | lim_sup: 619.7563
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 68.0000 | lim_sup: 68.0000
Límites IQR -> lim_inf: -33.0000 | lim_sup: 55.0000
Límites IQR -> lim_inf: -492.5000 | lim_sup: 907.5000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -102.0000 | lim_sup: 170.0000
Límites IQR -> lim_inf: -1613.9350 | lim_sup: 6726.6650
Límites IQR -> lim_inf: 461.2455 | lim_sup: 461.2455
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -434.6591 | lim_sup: 724.4318
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 203.4000 | lim_sup: 203.4000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -2517.3375 | lim_sup: 8093.0825
Límites IQR -> lim_inf: -777.0200 | lim_sup: 3817.2200
Límites IQR -> lim_inf: -538.5825 | lim_sup: 897.6375
Límites IQR -> lim_inf: -116.3475 | lim_sup: 193.9125
Límites IQR -> lim_inf: -37.5250 | lim_sup: 228.1950
Límites IQR -> lim_inf: -1172.5000 | lim_sup: 2347.5000
Límites IQR -> lim_inf: 172.0625 | lim_sup: 423.3625
Límites IQR -> lim_inf: 512.0300 | lim_sup: 1227.2300
Límites IQR -> lim_inf: -1016.1000 | lim_sup: 3300.0600
Límites IQR -> lim_inf: 1602.4800 | lim_sup: 1602.4800
Límites IQR -> lim_inf: 231.3231 | lim_sup: 4555.9951
Límites IQR -> lim_inf: 7911.7000 | lim_sup: 7911.7000
Límites IQR -> lim_inf: 45.6697 | lim_sup: 4819.3982
Límites IQR -> lim_inf: 9232.2632 | lim_sup: 15899.6192
Límites IQR -> lim_inf: 4382.6700 | lim_sup: 4382.6700
Límites IQR -> lim_inf: 146.3375 | lim_sup: 184.4775
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 8360.7770 | lim_sup: 8360.7770
Límites IQR -> lim_inf: 310.5100 | lim_sup: 310.5100

print(df_train_enr.shape)

print(df_train_enr.head())

(65466, 9)
   ID_BUILDING           FM_COST_TYPE      FECHA  YEAR  cost_float_mod  \
0            2  Eficiencia Energética 2021-12-01  2021            0.00
1            2              Licencias 2021-01-01  2021         1145.46
2            2              Licencias 2021-02-01  2021           55.95
3            2              Licencias 2021-03-01  2021            0.00
4            2              Licencias 2021-04-01  2021            0.00

   is_outlier  cost_float_mod_2  cost_float_mean_y  cost_float_mod_3
0           0          0.000000                NaN            0.0000
1           1        255.576875           219.4425          219.4425
2           0        241.589375           219.4425           55.9500
3           0        185.639375           219.4425            0.0000
4           0        185.639375           219.4425            0.0000

# =====================================================
# 1) Clasificación de estabilidad (test siempre usa cost_float_mod)
# =====================================================
def evaluar_estabilidad(s_train, s_test, umbral_media=0.3, umbral_wape=0.5, umbral_std=2.0):
    mean_train, mean_test = s_train.mean(), s_test.mean()
    std_train, std_test   = s_train.std(ddof=0), s_test.std(ddof=0)
    ratio_media = np.inf if mean_train == 0 else abs(mean_test - mean_train) / abs(mean_train)
    ratio_std   = np.inf if std_train == 0 else std_test / std_train
    if len(s_train.unique()) > 1 and len(s_test.unique()) > 1:
        _, p_value = mannwhitneyu(s_train, s_test, alternative="two-sided")
    else:
        p_value = 1.0
    wape = np.sum(np.abs(s_test - s_train.iloc[-1])) / np.sum(np.abs(s_test)) if s_test.sum() != 0 else np.inf
    if (ratio_media > umbral_media) or (wape > umbral_wape) or (p_value < 0.05) or (ratio_std > umbral_std) or (ratio_std < 1/umbral_std):
        clasificacion = "CAMBIO"
    else:
        clasificacion = "ESTABLE"
    return ratio_media, ratio_std, p_value, wape, clasificacion

resultados = []
vars_coste = ["cost_float_mod", "cost_float_mod_2", "cost_float_mod_3"]

for (bid, ctype), g_train in df_train_enr.groupby(["ID_BUILDING","FM_COST_TYPE"]):
    g_test = df_fd1_v5_ITE1_test[(df_fd1_v5_ITE1_test["ID_BUILDING"]==bid) & (df_fd1_v5_ITE1_test["FM_COST_TYPE"]==ctype)]
    if g_test.empty:
        continue
    s_test = g_test.set_index("FECHA")["cost_float_mod"]  # SIEMPRE real 2024 sin transformar
    for var in vars_coste:
        s_train = g_train.set_index("FECHA")[var]
        # Aseguramos alineación mensual (por si faltaran meses en test)
        s_train = s_train.sort_index()
        s_test  = s_test.sort_index()
        ratio_media, ratio_std, p_value, wape, clasificacion = evaluar_estabilidad(s_train, s_test)
        resultados.append({
            "ID_BUILDING": bid,
            "FM_COST_TYPE": ctype,
            "variable": var,
            "ratio_media": ratio_media,
            "ratio_std": ratio_std,
            "p_value": p_value,
            "WAPE": wape,
            "clasificacion": clasificacion
        })

df_estabilidad = pd.DataFrame(resultados)

# =====================================================
# 2) Resumen porcentual por variable
# =====================================================
resumen = (
    df_estabilidad.groupby(["variable","clasificacion"])
    .size().groupby(level=0).apply(lambda x: 100*x/x.sum())
    .unstack(fill_value=0)
).sort_index()

print("=== RESUMEN % ESTABLES POR VARIABLE (test usa cost_float_mod real) ===")
print(resumen)

print("\n=== MUESTRA DEL DETALLE ===")
print(df_estabilidad.head())def evaluar_estabilidad(s_train, s_test, umbral_media=0.3, umbral_wape=0.5, umbral_std=2.0):
    # Medias y desviaciones
    mean_train, mean_test = s_train.mean(), s_test.mean()
    std_train, std_test   = s_train.std(ddof=0), s_test.std(ddof=0)

    # Ratio de medias
    ratio_media = np.inf if mean_train == 0 else abs(mean_test - mean_train) / abs(mean_train)
    # Ratio de desviaciones
    ratio_std   = np.inf if std_train == 0 else std_test / std_train

    # Test de distribuciones (si hay varianza en ambas partes)
    if len(s_train.unique()) > 1 and len(s_test.unique()) > 1:
        stat, p_value = mannwhitneyu(s_train, s_test, alternative="two-sided")
    else:
        p_value = 1.0

    # WAPE naïve: predicción = último valor de train
    naive_pred = pd.Series([s_train.iloc[-1]] * len(s_test), index=s_test.index)
    wape = np.sum(np.abs(s_test - naive_pred)) / np.sum(np.abs(s_test)) if s_test.sum() != 0 else np.inf

    # Clasificación
    if (ratio_media > umbral_media) or (wape > umbral_wape) or (p_value < 0.05) or (ratio_std > umbral_std) or (ratio_std < 1/umbral_std):
        clasificacion = "CAMBIO"
    else:
        clasificacion = "ESTABLE"

    return {
        "ratio_media": ratio_media,
        "ratio_std": ratio_std,
        "p_value": p_value,
        "WAPE": wape,
        "clasificacion": clasificacion
    }

# Ejemplo aplicado a UNA serie (mod)
resultado = evaluar_estabilidad(
    serie_train_enriquecida.set_index("FECHA")["cost_float_mod"],
    s_real.set_index("FECHA")["valor"]
)

print(resultado)

{'ratio_media': np.float64(0.6075727799407163), 'ratio_std': 0.330877232154499, 'p_value': np.float64(0.03942269395503449), 'WAPE': np.float64(2.996209566379454), 'clasificacion': 'CAMBIO'}

# Nosotros usamos esta evaluación desglosada
def evaluar_estabilidad_desglosado(
    s_train: pd.Series,
    s_test: pd.Series,
    umbral_media: float = 0.3,
    umbral_wape: float = 0.5,
    umbral_std: float = 2.0,
    alpha_pvalue: float = 0.05,
):
    # Aseguramos tipos y orden
    s_train = s_train.astype(float).sort_index()
    s_test  = s_test.astype(float).sort_index()

    # Métricas base
    mean_train, mean_test = s_train.mean(), s_test.mean()
    std_train,  std_test  = s_train.std(ddof=0), s_test.std(ddof=0)

    ratio_media = np.inf if mean_train == 0 else abs(mean_test - mean_train) / abs(mean_train)
    ratio_std   = np.inf if std_train == 0 else (std_test / std_train)

    if s_train.nunique() > 1 and s_test.nunique() > 1:
        _, p_value = mannwhitneyu(s_train.values, s_test.values, alternative="two-sided")
    else:
        p_value = 1.0  # sin variabilidad suficiente, no señalamos diferencia

    if s_test.abs().sum() == 0:
        wape = np.inf
    else:
        naive_pred = pd.Series([s_train.iloc[-1]] * len(s_test), index=s_test.index)
        wape = (s_test - naive_pred).abs().sum() / s_test.abs().sum()

    # Reglas por métrica (flags)
    flag_media     = ratio_media > umbral_media
    flag_ratio_std = (ratio_std > umbral_std) or (ratio_std < 1.0/umbral_std)
    flag_pvalue    = p_value < alpha_pvalue
    flag_wape      = wape > umbral_wape

    # Clasificaciones por métrica
    clasificacion_media      = "CAMBIO" if flag_media     else "ESTABLE"
    clasificacion_ratio_std  = "CAMBIO" if flag_ratio_std else "ESTABLE"
    clasificacion_pvalue     = "CAMBIO" if flag_pvalue    else "ESTABLE"
    clasificacion_wape       = "CAMBIO" if flag_wape      else "ESTABLE"

    # Clasificación global (OR de las reglas)
    flag_global = flag_media or flag_ratio_std or flag_pvalue or flag_wape
    clasificacion_global = "CAMBIO" if flag_global else "ESTABLE"

    return {
        # Métricas numéricas
        "ratio_media": ratio_media,
        "ratio_std": ratio_std,
        "p_value": p_value,
        "WAPE": wape,
        # Flags (0/1) por si queremos auditar fácil
        "flag_media": int(flag_media),
        "flag_ratio_std": int(flag_ratio_std),
        "flag_pvalue": int(flag_pvalue),
        "flag_wape": int(flag_wape),
        "flag_global": int(flag_global),
        # Clasificaciones por métrica
        "clasificacion_media": clasificacion_media,
        "clasificacion_ratio_std": clasificacion_ratio_std,
        "clasificacion_pvalue": clasificacion_pvalue,
        "clasificacion_wape": clasificacion_wape,
        # Clasificación global
        "clasificacion_estad_global_train_vs_test": clasificacion_global,
    }

# Ejemplo de integración dentro del bucle que construye df_estabilidad
# (asumimos que ya tenemos df_train_enr y df_fd1_v5_ITE1_test, y la función _to_monthly_series definida)
def _to_monthly_series(g: pd.DataFrame, col: str):
    s = g.set_index("FECHA")[col].sort_index()
    return s.resample("MS").sum().fillna(0.0)

resultados = []
vars_coste = ["cost_float_mod", "cost_float_mod_2", "cost_float_mod_3"]

pairs_train = df_train_enr[["ID_BUILDING","FM_COST_TYPE"]].drop_duplicates()
pairs_test  = df_fd1_v5_ITE1_test[["ID_BUILDING","FM_COST_TYPE"]].drop_duplicates()
pairs_both  = pairs_train.merge(pairs_test, on=["ID_BUILDING","FM_COST_TYPE"])

for _, row in pairs_both.iterrows():
    bid, ctype = row["ID_BUILDING"], row["FM_COST_TYPE"]

    g_train = df_train_enr[(df_train_enr["ID_BUILDING"]==bid) & (df_train_enr["FM_COST_TYPE"]==ctype)]
    g_test  = df_fd1_v5_ITE1_test[(df_fd1_v5_ITE1_test["ID_BUILDING"]==bid) & (df_fd1_v5_ITE1_test["FM_COST_TYPE"]==ctype)]
    if g_train.empty or g_test.empty:
        continue

    # Serie real 2024 sin transformar
    s_test = _to_monthly_series(g_test, "cost_float_mod")
    s_test = s_test[(s_test.index.year == 2024)]

    for var in vars_coste:
        s_train = _to_monthly_series(g_train, var)
        s_train = s_train[(s_train.index.year >= 2021) & (s_train.index.year <= 2023)]
        if len(s_train)==0 or len(s_test)==0:
            continue

        res = evaluar_estabilidad_desglosado(s_train, s_test)

        resultados.append({
            "ID_BUILDING": bid,
            "FM_COST_TYPE": ctype,
            "variable": var,
            **res  # expandimos el diccionario con métricas, flags y clasificaciones
        })

df_estabilidad = pd.DataFrame(resultados)

# Resumen: % ESTABLE/CAMBIO usando la clasificación global
resumen_global = (
    df_estabilidad.groupby(["variable","clasificacion_estad_global_train_vs_test"])
    .size()
    .groupby(level=0)
    .apply(lambda x: 100*x/x.sum())
    .unstack(fill_value=0)
).round(2)

print("=== RESUMEN % (clasificación global) ===")
print(resumen_global)

print("\n=== Muestra del detalle con desgloses ===")
print(df_estabilidad.head())

=== RESUMEN % (clasificación global) ===
clasificacion_estad_global_train_vs_test  CAMBIO  ESTABLE
variable         variable
cost_float_mod   cost_float_mod            97.40     2.60
cost_float_mod_2 cost_float_mod_2          96.95     3.05
cost_float_mod_3 cost_float_mod_3          97.40     2.60

=== Muestra del detalle con desgloses ===
   ID_BUILDING     FM_COST_TYPE          variable  ratio_media  ratio_std  \
0            2        Licencias    cost_float_mod     0.622975   0.000000
1            2        Licencias  cost_float_mod_2     0.622975   0.000000
2            2        Licencias  cost_float_mod_3     3.469217   0.000000
3            2  Mtto. Contratos    cost_float_mod     0.750333   0.614012
4            2  Mtto. Contratos  cost_float_mod_2     0.750333   0.653764

   p_value      WAPE  flag_media  flag_ratio_std  flag_pvalue  flag_wape  \
0  1.00000  1.000000           1               1            0          1
1  1.00000  1.000000           1               1            0          1
2  1.00000  1.000000           1               1            0          1
3  0.00138  3.078200           1               0            1          1
4  0.00138  3.226715           1               0            1          1

   flag_global clasificacion_media clasificacion_ratio_std  \
0            1              CAMBIO                  CAMBIO
1            1              CAMBIO                  CAMBIO
2            1              CAMBIO                  CAMBIO
3            1              CAMBIO                 ESTABLE
4            1              CAMBIO                 ESTABLE

  clasificacion_pvalue clasificacion_wape  \
0              ESTABLE             CAMBIO
1              ESTABLE             CAMBIO
2              ESTABLE             CAMBIO
3               CAMBIO             CAMBIO
4               CAMBIO             CAMBIO

  clasificacion_estad_global_train_vs_test
0                                   CAMBIO
1                                   CAMBIO
2                                   CAMBIO
3                                   CAMBIO
4                                   CAMBIO

# 1) Auxiliares
def _to_ms(g, col):
    s = g.set_index("FECHA")[col].sort_index()
    return s.resample("MS").sum().fillna(0.0)

def _mad(x):
    # MAD robusta (normalizada a sigma ~ 1.4826 * MAD si quisiéramos)
    x = np.asarray(x, dtype=float)
    return median_abs_deviation(x, scale=1.0, nan_policy="omit")

def _prop_zeros(x):
    x = np.asarray(x, dtype=float)
    return np.mean(x == 0.0)

# 2) Evaluación robusta por pareja y variable
def evaluar_estabilidad_robusta(
    s_train, s_test,
    alpha_pvalue=0.05,
    umbral_mediana=0.30,         # |mediana_24 - mediana_train| / |mediana_train|
    umbral_mad_hi=1.75,          # ratio MAD fuera [1/umbral_mad_hi, umbral_mad_hi]
    umbral_wape=0.50,            # WAPE vs naive anual
    umbral_anual=0.30,           # |suma_24 - media_anual_train| / |media_anual_train|
    umbral_delta_zeros=0.30,     # cambio en proporción de ceros > 30 p.p.
    regla_mayoria_k=2            # nº mínimo de flags para CAMBIO
):
    s_train = s_train.astype(float)
    s_test  = s_test.astype(float)

    # Ventanas
    train_yrs = s_train.index.year
    y_max = train_yrs.max() if len(train_yrs) else 2023
    s_train_last_year = s_train[train_yrs == y_max]  # último año del train (2023)
    # Si por alguna razón no hay 2023, usamos todo el train como fallback
    if s_train_last_year.empty:
        s_train_last_year = s_train

    # Señales robustas
    med_tr, med_te = np.median(s_train), np.median(s_test)
    ratio_mediana = np.inf if med_tr == 0 else abs(med_te - med_tr) / abs(med_tr)

    mad_tr, mad_te = _mad(s_train), _mad(s_test)
    if mad_tr == 0:
        ratio_mad = np.inf if mad_te > 0 else 1.0
    else:
        ratio_mad = mad_te / mad_tr

    # Test de distribuciones (Mann-Whitney) si hay variabilidad
    if s_train.nunique() > 1 and s_test.nunique() > 1:
        _, p_value = mannwhitneyu(s_train.values, s_test.values, alternative="two-sided")
    else:
        p_value = 1.0

    # WAPE vs naïve anual (predecimos todo 2024 con la mediana del último año de train)
    naive_level = float(np.median(s_train_last_year))
    denom = s_test.abs().sum()
    if denom == 0:
        wape = np.inf
    else:
        wape = (s_test - naive_level).abs().sum() / denom

    # Chequeo anual (comparativa de masa)
    # media_anual_train = promedio de sumas anuales en 2021-2023
    train_by_year = s_train.groupby(s_train.index.year).sum()
    if len(train_by_year) == 0 or train_by_year.mean() == 0:
        ratio_anual = np.inf
    else:
        ratio_anual = abs(s_test.sum() - train_by_year.mean()) / abs(train_by_year.mean())

    # Sparsidad: cambio en % de ceros
    dz = abs(_prop_zeros(s_test) - _prop_zeros(s_train))

    # Flags
    flag_mediana   = ratio_mediana > umbral_mediana
    flag_mad       = (ratio_mad > umbral_mad_hi) or (ratio_mad < 1.0/umbral_mad_hi)
    flag_pvalue    = p_value < alpha_pvalue
    flag_wape      = wape > umbral_wape
    flag_anual     = ratio_anual > umbral_anual
    flag_sparsidad = dz > umbral_delta_zeros

    flags = dict(
        flag_mediana=int(flag_mediana),
        flag_mad=int(flag_mad),
        flag_pvalue=int(flag_pvalue),
        flag_wape=int(flag_wape),
        flag_anual=int(flag_anual),
        flag_sparsidad=int(flag_sparsidad),
    )
    n_flags = sum(flags.values())

    clasif_global = "CAMBIO" if n_flags >= regla_mayoria_k else "ESTABLE"

    return {
        # Métricas
        "ratio_mediana": ratio_mediana,
        "ratio_mad": ratio_mad,
        "p_value": p_value,
        "WAPE_naive_anual": wape,
        "ratio_anual": ratio_anual,
        "delta_prop_zeros": dz,
        # Flags
        **flags,
        "n_flags": n_flags,
        # Clasificación
        "clasificacion_estad_global_train_vs_test": clasif_global
    }

# Aplicamos al panel (train enriquecido vs test real)
vars_coste = ["cost_float_mod", "cost_float_mod_2", "cost_float_mod_3"]
resultados = []

pairs_train = df_train_enr[["ID_BUILDING","FM_COST_TYPE"]].drop_duplicates()
pairs_test  = df_fd1_v5_ITE1_test[["ID_BUILDING","FM_COST_TYPE"]].drop_duplicates()
pairs_both  = pairs_train.merge(pairs_test, on=["ID_BUILDING","FM_COST_TYPE"])

for _, row in pairs_both.iterrows():
    bid, ctype = row["ID_BUILDING"], row["FM_COST_TYPE"]
    g_train = df_train_enr[(df_train_enr["ID_BUILDING"]==bid) & (df_train_enr["FM_COST_TYPE"]==ctype)]
    g_test  = df_fd1_v5_ITE1_test[(df_fd1_v5_ITE1_test["ID_BUILDING"]==bid) & (df_fd1_v5_ITE1_test["FM_COST_TYPE"]==ctype)]
    if g_train.empty or g_test.empty:
        continue

    s_test = _to_ms(g_test, "cost_float_mod")
    s_test = s_test[(s_test.index.year == 2024)]

    for var in vars_coste:
        s_train = _to_ms(g_train, var)
        s_train = s_train[(s_train.index.year >= 2021) & (s_train.index.year <= 2023)]
        if len(s_train)==0 or len(s_test)==0:
            continue

        res = evaluar_estabilidad_robusta(s_train, s_test,
                                          alpha_pvalue=0.05,
                                          umbral_mediana=0.30,
                                          umbral_mad_hi=1.75,
                                          umbral_wape=0.50,
                                          umbral_anual=0.30,
                                          umbral_delta_zeros=0.30,
                                          regla_mayoria_k=2)   # aquí fijamos mayoría (2 señales)

        resultados.append({
            "ID_BUILDING": bid,
            "FM_COST_TYPE": ctype,
            "variable": var,
            **res
        })

df_estabilidad_rb = pd.DataFrame(resultados)

# 4) Resumen por variable con la nueva clasificación global (mayoría)
resumen_rb = (
    df_estabilidad_rb.groupby(["variable","clasificacion_estad_global_train_vs_test"])
    .size()
    .groupby(level=0)
    .apply(lambda x: 100*x/x.sum())
    .unstack(fill_value=0)
).round(2)

print("=== RESUMEN % (regla de mayoría, señales robustas) ===")
print(resumen_rb)

print("\n=== Muestra detalle ===")
print(df_estabilidad_rb.head())

=== RESUMEN % (regla de mayoría, señales robustas) ===
clasificacion_estad_global_train_vs_test  CAMBIO  ESTABLE
variable         variable
cost_float_mod   cost_float_mod            84.76    15.24
cost_float_mod_2 cost_float_mod_2          86.02    13.98
cost_float_mod_3 cost_float_mod_3          84.89    15.11

=== Muestra detalle ===
   ID_BUILDING     FM_COST_TYPE          variable  ratio_mediana  ratio_mad  \
0            2        Licencias    cost_float_mod            inf        1.0
1            2        Licencias  cost_float_mod_2       0.590248        0.0
2            2        Licencias  cost_float_mod_3            inf        1.0
3            2  Mtto. Contratos    cost_float_mod       0.830177        0.0
4            2  Mtto. Contratos  cost_float_mod_2       0.838205        0.0

   p_value  WAPE_naive_anual  ratio_anual  delta_prop_zeros  flag_mediana  \
0  1.00000          1.000000     0.860888          0.742857             1
1  1.00000          1.000000     0.860888          0.314286             1
2  1.00000          1.000000     0.616924          0.742857             1
3  0.00138          3.435270     0.750333          0.027778             1
4  0.00138          3.583785     0.750333          0.027778             1

   flag_mad  flag_pvalue  flag_wape  flag_anual  flag_sparsidad  n_flags  \
0         0            0          1           1               1        4
1         1            0          1           1               1        5
2         0            0          1           1               1        4
3         1            1          1           1               0        5
4         1            1          1           1               0        5

  clasificacion_estad_global_train_vs_test
0                                   CAMBIO
1                                   CAMBIO
2                                   CAMBIO
3                                   CAMBIO
4                                   CAMBIO

# =====================================================
# Clasificación de estabilidad usando sólo 2022–2023
# =====================================================

vars_coste = ["cost_float_mod", "cost_float_mod_2", "cost_float_mod_3"]
resultados = []

pairs_train = df_train_enr[["ID_BUILDING","FM_COST_TYPE"]].drop_duplicates()
pairs_test  = df_fd1_v5_ITE1_test[["ID_BUILDING","FM_COST_TYPE"]].drop_duplicates()
pairs_both  = pairs_train.merge(pairs_test, on=["ID_BUILDING","FM_COST_TYPE"])

for _, row in pairs_both.iterrows():
    bid, ctype = row["ID_BUILDING"], row["FM_COST_TYPE"]

    g_train = df_train_enr[
        (df_train_enr["ID_BUILDING"]==bid) &
        (df_train_enr["FM_COST_TYPE"]==ctype) &
        (df_train_enr["YEAR"].between(2022, 2023))
    ]
    g_test  = df_fd1_v5_ITE1_test[
        (df_fd1_v5_ITE1_test["ID_BUILDING"]==bid) &
        (df_fd1_v5_ITE1_test["FM_COST_TYPE"]==ctype) &
        (df_fd1_v5_ITE1_test["YEAR"]==2024)
    ]
    if g_train.empty or g_test.empty:
        continue

    s_test = _to_ms(g_test, "cost_float_mod")
    s_test = s_test[(s_test.index.year == 2024)]

    for var in vars_coste:
        s_train = _to_ms(g_train, var)
        s_train = s_train[(s_train.index.year >= 2022) & (s_train.index.year <= 2023)]
        if len(s_train)==0 or len(s_test)==0:
            continue

        res = evaluar_estabilidad_robusta(
            s_train, s_test,
            alpha_pvalue=0.05,
            umbral_mediana=0.30,
            umbral_mad_hi=1.75,
            umbral_wape=0.50,
            umbral_anual=0.30,
            umbral_delta_zeros=0.30,
            regla_mayoria_k=2
        )

        resultados.append({
            "ID_BUILDING": bid,
            "FM_COST_TYPE": ctype,
            "variable": var,
            **res
        })

df_estabilidad_rb_22_23 = pd.DataFrame(resultados)

# Resumen % por variable
resumen_rb_22_23 = (
    df_estabilidad_rb_22_23.groupby(["variable","clasificacion_estad_global_train_vs_test"])
    .size()
    .groupby(level=0)
    .apply(lambda x: 100*x/x.sum())
    .unstack(fill_value=0)
).round(2)

print("=== RESUMEN % (2022–2023 vs 2024) ===")
print(resumen_rb_22_23)

print("\n=== Muestra detalle ===")
print(df_estabilidad_rb_22_23.head())

=== RESUMEN % (2022–2023 vs 2024) ===
clasificacion_estad_global_train_vs_test  CAMBIO  ESTABLE
variable         variable
cost_float_mod   cost_float_mod            84.27    15.73
cost_float_mod_2 cost_float_mod_2          83.95    16.05
cost_float_mod_3 cost_float_mod_3          84.77    15.23

=== Muestra detalle ===
   ID_BUILDING     FM_COST_TYPE          variable  ratio_mediana  ratio_mad  \
0            2        Licencias    cost_float_mod            inf        1.0
1            2        Licencias  cost_float_mod_2       0.590248        0.0
2            2        Licencias  cost_float_mod_3            inf        1.0
3            2  Mtto. Contratos    cost_float_mod       0.859309        0.0
4            2  Mtto. Contratos  cost_float_mod_2       0.863133        0.0

    p_value  WAPE_naive_anual  ratio_anual  delta_prop_zeros  flag_mediana  \
0  1.000000          1.000000     0.766915          0.869565             1
1  1.000000          1.000000     0.766915          0.478261             1
2  1.000000          1.000000     0.067661          0.869565             1
3  0.000028          3.435270     0.810539          0.166667             1
4  0.000028          3.583785     0.810539          0.166667             1

   flag_mad  flag_pvalue  flag_wape  flag_anual  flag_sparsidad  n_flags  \
0         0            0          1           1               1        4
1         1            0          1           1               1        5
2         0            0          1           0               1        3
3         1            1          1           1               0        5
4         1            1          1           1               0        5

  clasificacion_estad_global_train_vs_test
0                                   CAMBIO
1                                   CAMBIO
2                                   CAMBIO
3                                   CAMBIO
4                                   CAMBIO

# Ruta base donde guardamos los outputs
ruta_base = "/content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/"

# Dataset de entrenamiento (Iteración 2: 2022-2024)
df_fd1_v5_ITE2 = pd.read_csv(ruta_base + "df_fd1_v5_ITE2.csv", sep=";")

# Dataset de reales (Iteración 2: datos del 2025 hasta agosto)
df_fd1_v5_Real_ITE2 = pd.read_csv(ruta_base + "df_fd1_v5_Real_ITE2_2025_hasta_agosto.csv", sep=";")

# Confirmamos tamaños y columnas
print("ITE2 shape:", df_fd1_v5_ITE2.shape)
print("Real_ITE2 shape:", df_fd1_v5_Real_ITE2.shape)
print("\nColumnas ITE2:", df_fd1_v5_ITE2.columns.tolist())
print("Columnas Real_ITE2:", df_fd1_v5_Real_ITE2.columns.tolist())

ITE2 shape: (243916, 12)
Real_ITE2 shape: (57932, 12)

Columnas ITE2: ['ID_ORDER', 'ID_BUILDING', 'FM_COST_TYPE', 'MONTH', 'YEAR', 'cost_float_mod', 'SUPPLIER_TYPE_MOD_2', 'FM_RESPONSIBLE_MOD', 'TIPO_USO', 'COUNTRY_CATALOGO', 'ID_REGION_GRUPO', 'COUNTRY_DEF']
Columnas Real_ITE2: ['ID_ORDER', 'ID_BUILDING', 'FM_COST_TYPE', 'MONTH', 'YEAR', 'cost_float_mod', 'SUPPLIER_TYPE_MOD_2', 'FM_RESPONSIBLE_MOD', 'TIPO_USO', 'COUNTRY_CATALOGO', 'ID_REGION_GRUPO', 'COUNTRY_DEF']

# ===============================================================
# Secuencia ITE2) Generamos df_fd1_v5_ITE2_train y df_fd1_v5_ITE2_test
# Partimos de:
#   - df_fd1_v5_ITE2            (detalle 2022–2024)
#   - df_fd1_v5_Real_ITE2       (detalle 2025 hasta agosto)
# ===============================================================

# 1) Creamos FECHA en ambos orígenes
df_ite2_detalle_train = crear_fecha(df_fd1_v5_ITE2)
df_ite2_detalle_test  = crear_fecha(df_fd1_v5_Real_ITE2)

# 2) Filtramos por YEAR explícitamente
df_ite2_detalle_train = df_ite2_detalle_train[(df_ite2_detalle_train["YEAR"]>=2022) & (df_ite2_detalle_train["YEAR"]<=2024)]
df_ite2_detalle_test  = df_ite2_detalle_test[(df_ite2_detalle_test["YEAR"]==2025)]

# 3) Agregamos a nivel mensual por pareja (suma de cost_float_mod)
agg_train_ite2 = serie_mensual_con_huecos(df_ite2_detalle_train, 2022, 2024)
agg_test_ite2  = serie_mensual_con_huecos(df_ite2_detalle_test,  2025, 2025)

# 4) Rellenamos meses faltantes entre el primer y último mes observado de cada pareja
serie_train_completa_ite2 = completar_meses_por_pareja(agg_train_ite2)
serie_test_completa_ite2  = completar_meses_por_pareja(agg_test_ite2)

# 5) Calculamos contexto por pareja usando SOLO 2022–2024 (moda)
ctx_2022_2024 = contexto_por_pareja(df_ite2_detalle_train)

# 6) Ensamblamos finales con contexto y añadimos YEAR/MONTH
df_fd1_v5_ITE2_train = ensamblar_final(serie_train_completa_ite2, ctx_2022_2024)
df_fd1_v5_ITE2_test  = ensamblar_final(serie_test_completa_ite2,  ctx_2022_2024)

# 7) Comprobaciones rápidas
chequeo_rapido("df_fd1_v5_ITE2_train (2022–2024)", df_fd1_v5_ITE2_train, 2022, 2024)
chequeo_rapido("df_fd1_v5_ITE2_test (2025)",       df_fd1_v5_ITE2_test,  2025, 2025)

=== df_fd1_v5_ITE2_train (2022–2024) ===
Shape: (72492, 11)
FECHA min: 2022-01-01 00:00:00 | FECHA max: 2024-12-01 00:00:00
Duplicados por pareja/FECHA: 0

Muestra de filas:
   ID_BUILDING FM_COST_TYPE      FECHA  YEAR  MONTH  cost_float_mod  \
0            2    Licencias 2022-02-01  2022      2          610.95
1            2    Licencias 2022-03-01  2022      3          927.09
2            2    Licencias 2022-04-01  2022      4            0.00
3            2    Licencias 2022-05-01  2022      5            0.00
4            2    Licencias 2022-06-01  2022      6            0.00

  COUNTRY_DEF ID_REGION_GRUPO  TIPO_USO SUPPLIER_TYPE_MOD_2 FM_RESPONSIBLE_MOD
0      España               2  Oficinas             INTERNO          Licencias
1      España               2  Oficinas             INTERNO          Licencias
2      España               2  Oficinas             INTERNO          Licencias
3      España               2  Oficinas             INTERNO          Licencias
4      España               2  Oficinas             INTERNO          Licencias
============================================================

=== df_fd1_v5_ITE2_test (2025) ===
Shape: (14836, 11)
FECHA min: 2025-01-01 00:00:00 | FECHA max: 2025-08-01 00:00:00
Duplicados por pareja/FECHA: 0

Muestra de filas:
   ID_BUILDING     FM_COST_TYPE      FECHA  YEAR  MONTH  cost_float_mod  \
0            2        Licencias 2025-02-01  2025      2           20.00
1            2  Mtto. Contratos 2025-01-01  2025      1         1126.99
2            2  Mtto. Contratos 2025-02-01  2025      2            0.00
3            2  Mtto. Contratos 2025-03-01  2025      3            0.00
4            2  Mtto. Contratos 2025-04-01  2025      4            0.00

  COUNTRY_DEF ID_REGION_GRUPO  TIPO_USO SUPPLIER_TYPE_MOD_2 FM_RESPONSIBLE_MOD
0      España               2  Oficinas             INTERNO          Licencias
1      España               2  Oficinas             EXTERNO      Mantenimiento
2      España               2  Oficinas             EXTERNO      Mantenimiento
3      España               2  Oficinas             EXTERNO      Mantenimiento
4      España               2  Oficinas             EXTERNO      Mantenimiento
============================================================

# =======================
# Funciones Auxiliares reutilizados en un bloque
# =======================

def _ensure_monthly(df, date_col="FECHA"):
    df = df.copy()
    df[date_col] = pd.to_datetime(df[date_col])
    df = df.sort_values(date_col)
    fmin, fmax = df[date_col].min(), df[date_col].max()
    idx = pd.date_range(fmin, fmax, freq="MS")
    df = df.set_index(date_col).reindex(idx)
    df.index.name = date_col
    df = df.reset_index()
    return df

def _iqr_limits(x: pd.Series):
    Q1 = x.quantile(0.25); Q3 = x.quantile(0.75); IQR = Q3 - Q1
    return float(Q1 - 1.5*IQR), float(Q3 + 1.5*IQR)

def _redistribuir_deficit_sin_negativos(vals: np.ndarray, deficit: float) -> np.ndarray:
    s = vals.astype(float).copy(); rem = float(deficit)
    if rem <= 0: return s
    while rem > 1e-12:
        pos_mask = s > 0; total_pos = s[pos_mask].sum()
        if total_pos <= 0: break
        reduccion = np.zeros_like(s); reduccion[pos_mask] = rem * (s[pos_mask]/total_pos)
        cap = s.copy(); exceso_cap = np.maximum(reduccion - cap, 0.0)
        reduccion_efectiva = reduccion - exceso_cap
        s = s - reduccion_efectiva; s[s < 0] = 0.0
        rem = float(exceso_cap.sum());
        if rem <= 1e-12: rem = 0.0; break
    return s

def transformar_serie_outliers(df_serie, date_col="FECHA", value_col="cost_float_mod", asegurar_mensual=True):
    st = df_serie[[date_col, value_col]].copy()
    st[date_col] = pd.to_datetime(st[date_col]); st = st.sort_values(date_col)
    if asegurar_mensual:
        st = _ensure_monthly(st, date_col=date_col)
        if value_col not in st.columns: st[value_col] = 0.0
        st[value_col] = st[value_col].fillna(0.0)
    st["YEAR"] = st[date_col].dt.year
    st[value_col] = st[value_col].astype(float)

    lim_inf, lim_sup = _iqr_limits(st[value_col])
    st["is_outlier"] = ((st[value_col] < lim_inf) | (st[value_col] > lim_sup)).astype(int)
    st["cost_float_mod"]  = st[value_col]
    st["cost_float_mod_2"] = st["cost_float_mod"].astype(float)

    # Outliers positivos: aplanado + redistribución anual del exceso
    for y in sorted(st["YEAR"].unique()):
        m = st["YEAR"] == y
        vals = st.loc[m, "cost_float_mod_2"].values
        exceso = np.maximum(vals - lim_sup, 0.0); exceso_total = float(exceso.sum())
        if exceso_total > 0:
            st.loc[m, "cost_float_mod_2"] = np.minimum(vals, lim_sup)
            st.loc[m, "cost_float_mod_2"] += exceso_total / int(m.sum())

    # Outliers negativos: reglas
    for y in sorted(st["YEAR"].unique()):
        m = st["YEAR"] == y
        vals_orig = st.loc[m, "cost_float_mod"].values
        vals_adj  = st.loc[m, "cost_float_mod_2"].values
        lim_inf_y, _ = _iqr_limits(st["cost_float_mod"])  # lim_inf global de la serie
        out_neg = vals_orig < lim_inf_y
        if not np.any(out_neg):
            continue
        suma_anual = float(vals_orig.sum())
        if suma_anual < 0:
            vals_adj[out_neg] = 0.0
        else:
            deficit = float(np.abs(vals_adj[out_neg]).sum())
            vals_adj[out_neg] = 0.0
            if deficit > 0: vals_adj = _redistribuir_deficit_sin_negativos(vals_adj, deficit)
        st.loc[m, "cost_float_mod_2"] = vals_adj

    # cost_float_mean_y: media anual original en años con outliers
    st["cost_float_mean_y"] = np.nan
    years_out = st.groupby("YEAR")["is_outlier"].max()
    years_out = years_out[years_out == 1].index.tolist()
    if years_out:
        media_anual = st.groupby("YEAR")["cost_float_mod"].mean()
        for y in years_out:
            st.loc[st["YEAR"] == y, "cost_float_mean_y"] = media_anual.loc[y]

    # cost_float_mod_3: original con outliers sustituidos por la media anual
    st["cost_float_mod_3"] = st["cost_float_mod"]
    m_out = st["is_outlier"] == 1
    st.loc[m_out, "cost_float_mod_3"] = st.loc[m_out, "cost_float_mean_y"]

    return st[[date_col, "YEAR", "cost_float_mod","cost_float_mod_2","cost_float_mean_y","cost_float_mod_3"]].sort_values(date_col).reset_index(drop=True)

def transformar_panel_outliers(df_panel, id_cols=("ID_BUILDING","FM_COST_TYPE"), date_col="FECHA", value_col="cost_float_mod", asegurar_mensual=True):
    cols_min = list(id_cols) + [date_col, value_col]
    missing = set(cols_min) - set(df_panel.columns)
    if missing: raise ValueError(f"Faltan columnas: {missing}")
    df_panel = df_panel[cols_min].copy()
    df_panel[date_col] = pd.to_datetime(df_panel[date_col])
    outs = []
    for keys, g in df_panel.groupby(list(id_cols), sort=False):
        t = transformar_serie_outliers(g[[date_col, value_col]], date_col=date_col, value_col=value_col, asegurar_mensual=asegurar_mensual)
        for i, col in enumerate(id_cols):
            t[col] = keys[i] if isinstance(keys, tuple) else keys
        outs.append(t)
    out = pd.concat(outs, ignore_index=True)
    out = out[list(id_cols) + [date_col, "YEAR", "cost_float_mod","cost_float_mod_2","cost_float_mean_y","cost_float_mod_3"]]
    return out.sort_values(list(id_cols) + [date_col]).reset_index(drop=True)

def _to_ms(g, col):
    s = g.set_index("FECHA")[col].sort_index()
    return s.resample("MS").sum().fillna(0.0)

def _mad(x):
    x = np.asarray(x, dtype=float)
    return median_abs_deviation(x, scale=1.0, nan_policy="omit")

def _prop_zeros(x):
    x = np.asarray(x, dtype=float)
    return np.mean(x == 0.0)

def evaluar_estabilidad_robusta(
    s_train, s_test,
    alpha_pvalue=0.05,
    umbral_mediana=0.30,
    umbral_mad_hi=1.75,
    umbral_wape=0.50,
    umbral_anual=0.30,
    umbral_delta_zeros=0.30,
    regla_mayoria_k=2
):
    s_train = s_train.astype(float); s_test = s_test.astype(float)
    # nivel naïve = mediana del último año del train
    train_yrs = s_train.index.year
    y_max = train_yrs.max() if len(train_yrs) else 2024
    s_train_last_year = s_train[train_yrs == y_max] if (train_yrs == y_max).any() else s_train

    med_tr, med_te = np.median(s_train), np.median(s_test)
    ratio_mediana = np.inf if med_tr == 0 else abs(med_te - med_tr) / abs(med_tr)

    mad_tr, mad_te = _mad(s_train), _mad(s_test)
    ratio_mad = (np.inf if mad_tr == 0 and mad_te > 0 else (1.0 if mad_tr == 0 else mad_te / mad_tr))

    if s_train.nunique() > 1 and s_test.nunique() > 1:
        _, p_value = mannwhitneyu(s_train.values, s_test.values, alternative="two-sided")
    else:
        p_value = 1.0

    denom = s_test.abs().sum()
    naive_level = float(np.median(s_train_last_year))
    wape = np.inf if denom == 0 else (s_test - naive_level).abs().sum() / denom

    train_by_year = s_train.groupby(s_train.index.year).sum()
    ratio_anual = (np.inf if len(train_by_year)==0 or train_by_year.mean()==0
                   else abs(s_test.sum() - train_by_year.mean()) / abs(train_by_year.mean()))

    dz = abs(_prop_zeros(s_test) - _prop_zeros(s_train))

    flags = dict(
        flag_mediana = ratio_mediana > umbral_mediana,
        flag_mad     = (ratio_mad > umbral_mad_hi) or (ratio_mad < 1.0/umbral_mad_hi),
        flag_pvalue  = p_value < alpha_pvalue,
        flag_wape    = wape > umbral_wape,
        flag_anual   = ratio_anual > umbral_anual,
        flag_sparsidad = dz > umbral_delta_zeros
    )
    n_flags = sum(int(v) for v in flags.values())
    clasif = "CAMBIO" if n_flags >= regla_mayoria_k else "ESTABLE"

    return {
        "ratio_mediana": ratio_mediana,
        "ratio_mad": ratio_mad,
        "p_value": p_value,
        "WAPE_naive_anual": wape,
        "ratio_anual": ratio_anual,
        "delta_prop_zeros": dz,
        **{k:int(v) for k,v in flags.items()},
        "n_flags": n_flags,
        "clasificacion_estad_global_train_vs_test": clasif
    }

# =========================================
# 1) Enriquecemos SOLO el TRAIN ITE2
#     (partimos de df_fd1_v5_ITE2_train ya creado por tu pipeline)
# =========================================
# Si no existen mod_2/mod_3, los generamos
if not {"cost_float_mod_2","cost_float_mod_3"}.issubset(df_fd1_v5_ITE2_train.columns):
    df_ite2_train_enr = transformar_panel_outliers(
        df_fd1_v5_ITE2_train,
        id_cols=("ID_BUILDING","FM_COST_TYPE"),
        date_col="FECHA",
        value_col="cost_float_mod",
        asegurar_mensual=True
    )
else:
    df_ite2_train_enr = df_fd1_v5_ITE2_train.copy()

print("ITE2 train enriquecido:", df_ite2_train_enr.shape)

Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -2078.2750 | lim_sup: 3932.0050
Límites IQR -> lim_inf: -856.7375 | lim_sup: 1472.5025
Límites IQR -> lim_inf: -93.0712 | lim_sup: 155.1187
Límites IQR -> lim_inf: 5981.2850 | lim_sup: 6439.7250
Límites IQR -> lim_inf: -282.4875 | lim_sup: 470.8125
Límites IQR -> lim_inf: -10163.2637 | lim_sup: 28464.8662
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -423.7500 | lim_sup: 706.2500
Límites IQR -> lim_inf: -45.1412 | lim_sup: 3500.4288
Límites IQR -> lim_inf: -1224.6862 | lim_sup: 3603.7238
Límites IQR -> lim_inf: -907.7400 | lim_sup: 1512.9000
Límites IQR -> lim_inf: 18521.7138 | lim_sup: 36733.5438
Límites IQR -> lim_inf: -345.3050 | lim_sup: 736.8150
Límites IQR -> lim_inf: -3318.7075 | lim_sup: 19606.9725
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -16.8750 | lim_sup: 28.1250
Límites IQR -> lim_inf: 248.0512 | lim_sup: 3330.1013
Límites IQR -> lim_inf: -1157.6275 | lim_sup: 2280.3325
Límites IQR -> lim_inf: -7058.7888 | lim_sup: 11928.6813
Límites IQR -> lim_inf: -9593.2913 | lim_sup: 30422.8388
Límites IQR -> lim_inf: -555.3937 | lim_sup: 1117.8762
Límites IQR -> lim_inf: -1214.0600 | lim_sup: 11667.9400
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -428.9100 | lim_sup: 714.8500
Límites IQR -> lim_inf: 200.8400 | lim_sup: 200.8400
Límites IQR -> lim_inf: -451.5113 | lim_sup: 752.5187
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -278.7675 | lim_sup: 464.6125
Límites IQR -> lim_inf: -1232.8050 | lim_sup: 2054.6750
Límites IQR -> lim_inf: -929.8013 | lim_sup: 1549.6688
Límites IQR -> lim_inf: -2944.4562 | lim_sup: 5382.5537
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -832.0950 | lim_sup: 1386.8250
Límites IQR -> lim_inf: 4350.0000 | lim_sup: 4350.0000
Límites IQR -> lim_inf: 168.6800 | lim_sup: 2061.8000
Límites IQR -> lim_inf: -387.9562 | lim_sup: 646.5938
Límites IQR -> lim_inf: -611.7463 | lim_sup: 3207.1438
Límites IQR -> lim_inf: 165.0500 | lim_sup: 165.0500
Límites IQR -> lim_inf: -86.8650 | lim_sup: 144.7750
Límites IQR -> lim_inf: -159.5413 | lim_sup: 804.8088
Límites IQR -> lim_inf: -284.9562 | lim_sup: 1651.4937
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 3448.1400 | lim_sup: 4846.3800
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -383.0037 | lim_sup: 6872.2262
Límites IQR -> lim_inf: 248.2500 | lim_sup: 248.2500
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 165.3000 | lim_sup: 224.1000
Límites IQR -> lim_inf: -1735.0063 | lim_sup: 3688.7038
Límites IQR -> lim_inf: 5201.3038 | lim_sup: 6853.6937
Límites IQR -> lim_inf: -259.5000 | lim_sup: 432.5000
Límites IQR -> lim_inf: -1824.3400 | lim_sup: 11960.9400
Límites IQR -> lim_inf: 158.7200 | lim_sup: 158.7200
Límites IQR -> lim_inf: 250.0000 | lim_sup: 250.0000
Límites IQR -> lim_inf: 162.1700 | lim_sup: 260.1700
Límites IQR -> lim_inf: -814.4937 | lim_sup: 2332.9562
Límites IQR -> lim_inf: 2305.1862 | lim_sup: 3028.5563
Límites IQR -> lim_inf: -338.8050 | lim_sup: 564.6750
Límites IQR -> lim_inf: -1872.0812 | lim_sup: 8788.0887
Límites IQR -> lim_inf: 102.1100 | lim_sup: 102.1100
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -132.7600 | lim_sup: 723.9600
Límites IQR -> lim_inf: -1192.3962 | lim_sup: 2618.0937
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 1686.5000 | lim_sup: 2354.9800
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -500.8763 | lim_sup: 5382.0938
Límites IQR -> lim_inf: 249.5200 | lim_sup: 249.5200
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -430.8150 | lim_sup: 1272.9050
Límites IQR -> lim_inf: -1033.0050 | lim_sup: 3077.0950
Límites IQR -> lim_inf: 3320.5762 | lim_sup: 4910.1863
Límites IQR -> lim_inf: -532.6350 | lim_sup: 887.7250
Límites IQR -> lim_inf: -451.3275 | lim_sup: 10185.5725
Límites IQR -> lim_inf: 240.0800 | lim_sup: 240.0800
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 227.9550 | lim_sup: 365.7550
Límites IQR -> lim_inf: -920.8600 | lim_sup: 2384.7200
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 3041.5800 | lim_sup: 5319.7400
Límites IQR -> lim_inf: -604.5900 | lim_sup: 1007.6500
Límites IQR -> lim_inf: -1116.9200 | lim_sup: 9515.6800
Límites IQR -> lim_inf: 157.1100 | lim_sup: 157.1100
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -619.4175 | lim_sup: 1032.3625
Límites IQR -> lim_inf: -1092.4675 | lim_sup: 2673.1125
Límites IQR -> lim_inf: 250.0000 | lim_sup: 250.0000
Límites IQR -> lim_inf: 1509.0650 | lim_sup: 3217.6650
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -367.1188 | lim_sup: 6661.8513
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -412.6313 | lim_sup: 1071.0188
Límites IQR -> lim_inf: -920.9638 | lim_sup: 1664.6863
Límites IQR -> lim_inf: 250.0000 | lim_sup: 250.0000
Límites IQR -> lim_inf: 1799.2750 | lim_sup: 2845.4750
Límites IQR -> lim_inf: -183.2138 | lim_sup: 305.3562
Límites IQR -> lim_inf: -244.1562 | lim_sup: 3445.5137
Límites IQR -> lim_inf: -10193.4000 | lim_sup: 16989.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -530.0625 | lim_sup: 1112.4375
Límites IQR -> lim_inf: -788.8500 | lim_sup: 2349.2500
Límites IQR -> lim_inf: 250.0000 | lim_sup: 250.0000
Límites IQR -> lim_inf: 3661.9475 | lim_sup: 5418.8875
Límites IQR -> lim_inf: -322.0312 | lim_sup: 536.7188
Límites IQR -> lim_inf: -2259.7500 | lim_sup: 9549.9700
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -822.1950 | lim_sup: 1370.3250
Límites IQR -> lim_inf: -575.3100 | lim_sup: 1285.0100
Límites IQR -> lim_inf: 250.0000 | lim_sup: 250.0000
Límites IQR -> lim_inf: 1805.3187 | lim_sup: 2704.3887
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -264.7775 | lim_sup: 4324.1025
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 541.2500 | lim_sup: 803.2500
Límites IQR -> lim_inf: -1445.3438 | lim_sup: 2508.7063
Límites IQR -> lim_inf: -139.6050 | lim_sup: 232.6750
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -1730.0738 | lim_sup: 9740.2763
Límites IQR -> lim_inf: 211.9400 | lim_sup: 211.9400
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 257.7450 | lim_sup: 290.1050
Límites IQR -> lim_inf: -1230.2300 | lim_sup: 2944.6100
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -8.2500 | lim_sup: 453.7500
Límites IQR -> lim_inf: -828.8400 | lim_sup: 1381.4000
Límites IQR -> lim_inf: -594.0137 | lim_sup: 8409.8562
Límites IQR -> lim_inf: 43.3300 | lim_sup: 43.3300
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -194.2500 | lim_sup: 323.7500
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 218.5700 | lim_sup: 218.5700
Límites IQR -> lim_inf: -259.0875 | lim_sup: 1915.7725
Límites IQR -> lim_inf: 262.9000 | lim_sup: 262.9000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 282.5625 | lim_sup: 4404.2225
Límites IQR -> lim_inf: -747.5300 | lim_sup: 1762.7500
Límites IQR -> lim_inf: -100.8450 | lim_sup: 168.0750
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -2519.2125 | lim_sup: 10976.8675
Límites IQR -> lim_inf: 368.3900 | lim_sup: 368.3900
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -211.7050 | lim_sup: 1577.8550
Límites IQR -> lim_inf: -624.2313 | lim_sup: 3273.1388
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 4309.5550 | lim_sup: 5658.2350
Límites IQR -> lim_inf: -318.4275 | lim_sup: 530.7125
Límites IQR -> lim_inf: -1090.1950 | lim_sup: 14285.0650
Límites IQR -> lim_inf: -12960.0000 | lim_sup: 21600.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -458.6938 | lim_sup: 1544.6363
Límites IQR -> lim_inf: -1327.7613 | lim_sup: 3527.9887
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 1338.3100 | lim_sup: 6080.8700
Límites IQR -> lim_inf: -200.8125 | lim_sup: 334.6875
Límites IQR -> lim_inf: -3950.8625 | lim_sup: 15999.2775
Límites IQR -> lim_inf: -4666.5000 | lim_sup: 7777.5000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -965.7312 | lim_sup: 2373.0387
Límites IQR -> lim_inf: -1615.8112 | lim_sup: 3935.8387
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 3140.1525 | lim_sup: 4298.4125
Límites IQR -> lim_inf: -222.0000 | lim_sup: 370.0000
Límites IQR -> lim_inf: -2095.3200 | lim_sup: 13647.3400
Límites IQR -> lim_inf: 225.6900 | lim_sup: 225.6900
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 102.2750 | lim_sup: 338.6350
Límites IQR -> lim_inf: -1028.1650 | lim_sup: 2116.2750
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 3220.1950 | lim_sup: 4473.5150
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -890.0150 | lim_sup: 8804.3250
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 239.0000 | lim_sup: 239.0000
Límites IQR -> lim_inf: -720.6262 | lim_sup: 7274.2237
Límites IQR -> lim_inf: 235.3300 | lim_sup: 235.3300
Límites IQR -> lim_inf: -104.9625 | lim_sup: 174.9375
Límites IQR -> lim_inf: 384.0000 | lim_sup: 384.0000
Límites IQR -> lim_inf: -632.3662 | lim_sup: 1102.5437
Límites IQR -> lim_inf: 753.7000 | lim_sup: 2541.4600
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -1097.3763 | lim_sup: 9410.2938
Límites IQR -> lim_inf: 192.1700 | lim_sup: 192.1700
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 384.0000 | lim_sup: 384.0000
Límites IQR -> lim_inf: -267.3900 | lim_sup: 445.6500
Límites IQR -> lim_inf: 806.8925 | lim_sup: 2442.8325
Límites IQR -> lim_inf: -327.8550 | lim_sup: 546.4250
Límites IQR -> lim_inf: -2131.1337 | lim_sup: 9103.7762
Límites IQR -> lim_inf: 190.0700 | lim_sup: 190.0700
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 40.8750 | lim_sup: 251.8750
Límites IQR -> lim_inf: -1174.3775 | lim_sup: 2861.7225
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 1944.6000 | lim_sup: 2995.3200
Límites IQR -> lim_inf: -458.1000 | lim_sup: 763.5000
Límites IQR -> lim_inf: 19.5388 | lim_sup: 7440.2287
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 120.0000 | lim_sup: 120.0000
Límites IQR -> lim_inf: -443.8913 | lim_sup: 739.8188
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 384.5037 | lim_sup: 2247.5338
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -122.0475 | lim_sup: 3762.4725
Límites IQR -> lim_inf: 186.3800 | lim_sup: 186.3800
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -542.1875 | lim_sup: 1895.4325
Límites IQR -> lim_inf: -780.3188 | lim_sup: 1840.6113
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 2935.4550 | lim_sup: 3815.4950
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -1129.5788 | lim_sup: 9287.6513
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -428.4700 | lim_sup: 1993.0500
Límites IQR -> lim_inf: -1133.9900 | lim_sup: 3390.8700
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 4496.3988 | lim_sup: 5515.3288
Límites IQR -> lim_inf: -75.7500 | lim_sup: 126.2500
Límites IQR -> lim_inf: 177.1025 | lim_sup: 10289.9625
Límites IQR -> lim_inf: 169.7500 | lim_sup: 169.7500
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -473.3700 | lim_sup: 1188.9500
Límites IQR -> lim_inf: -1373.8187 | lim_sup: 2766.9512
Límites IQR -> lim_inf: 1054.1950 | lim_sup: 4632.1550
Límites IQR -> lim_inf: -340.9500 | lim_sup: 568.2500
Límites IQR -> lim_inf: -905.8450 | lim_sup: 7174.9550
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -449.8350 | lim_sup: 749.7250
Límites IQR -> lim_inf: -34.6075 | lim_sup: 902.6125
Límites IQR -> lim_inf: -277.5000 | lim_sup: 462.5000
Límites IQR -> lim_inf: -147.1775 | lim_sup: 919.0225
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -217.2225 | lim_sup: 362.0375
Límites IQR -> lim_inf: -199.3550 | lim_sup: 1127.2450
Límites IQR -> lim_inf: -234.8438 | lim_sup: 391.4062
Límites IQR -> lim_inf: -120.1800 | lim_sup: 922.5800
Límites IQR -> lim_inf: -1053.7700 | lim_sup: 3449.6300
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -33.6300 | lim_sup: 541.8900
Límites IQR -> lim_inf: -843.6425 | lim_sup: 1563.7375
Límites IQR -> lim_inf: -2262.1713 | lim_sup: 11791.4388
Límites IQR -> lim_inf: -2421.8850 | lim_sup: 7049.1950
Límites IQR -> lim_inf: -264.0000 | lim_sup: 440.0000
Límites IQR -> lim_inf: -451.8313 | lim_sup: 3436.1588
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -892.4362 | lim_sup: 1487.3937
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -93.3800 | lim_sup: 1086.8600
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -71.7300 | lim_sup: 889.0300
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -121.5000 | lim_sup: 202.5000
Límites IQR -> lim_inf: -1358.8800 | lim_sup: 2487.5600
Límites IQR -> lim_inf: 381.4400 | lim_sup: 1408.4800
Límites IQR -> lim_inf: -471.7687 | lim_sup: 786.2812
Límites IQR -> lim_inf: -391.3087 | lim_sup: 2281.1812
Límites IQR -> lim_inf: 60.0000 | lim_sup: 60.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -315.9075 | lim_sup: 526.5125
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 125.3050 | lim_sup: 666.5450
Límites IQR -> lim_inf: -300.7500 | lim_sup: 501.2500
Límites IQR -> lim_inf: -280.7712 | lim_sup: 1406.4987
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -265.0400 | lim_sup: 596.8000
Límites IQR -> lim_inf: -123.6150 | lim_sup: 206.0250
Límites IQR -> lim_inf: -53.6513 | lim_sup: 344.5987
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -150.8550 | lim_sup: 251.4250
Límites IQR -> lim_inf: -414.3750 | lim_sup: 690.6250
Límites IQR -> lim_inf: 464.9200 | lim_sup: 464.9200
Límites IQR -> lim_inf: -835.3687 | lim_sup: 1392.2812
Límites IQR -> lim_inf: -467.7700 | lim_sup: 1207.7300
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -530.2125 | lim_sup: 883.6875
Límites IQR -> lim_inf: 283.2500 | lim_sup: 401.6500
Límites IQR -> lim_inf: -330.0000 | lim_sup: 550.0000
Límites IQR -> lim_inf: -302.5975 | lim_sup: 1388.8025
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -539.8125 | lim_sup: 899.6875
Límites IQR -> lim_inf: -1520.6925 | lim_sup: 3177.6875
Límites IQR -> lim_inf: 11.1050 | lim_sup: 883.3050
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -132.1900 | lim_sup: 1236.2300
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -331.6050 | lim_sup: 552.6750
Límites IQR -> lim_inf: -762.7500 | lim_sup: 1271.2500
Límites IQR -> lim_inf: 308.3000 | lim_sup: 437.1800
Límites IQR -> lim_inf: -210.0000 | lim_sup: 350.0000
Límites IQR -> lim_inf: -156.5237 | lim_sup: 1122.9462
Límites IQR -> lim_inf: -225.0000 | lim_sup: 375.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -992.6150 | lim_sup: 2118.9450
Límites IQR -> lim_inf: 1235.8300 | lim_sup: 1440.1500
Límites IQR -> lim_inf: -127.1850 | lim_sup: 211.9750
Límites IQR -> lim_inf: -104.5813 | lim_sup: 1234.0688
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -518.2313 | lim_sup: 863.7188
Límites IQR -> lim_inf: -2833.3700 | lim_sup: 8500.1100
Límites IQR -> lim_inf: -319.8750 | lim_sup: 1042.3250
Límites IQR -> lim_inf: -480.0000 | lim_sup: 800.0000
Límites IQR -> lim_inf: -3.4312 | lim_sup: 734.1987
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -339.9300 | lim_sup: 566.5500
Límites IQR -> lim_inf: 152.1350 | lim_sup: 736.2550
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -240.7050 | lim_sup: 909.8350
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -2065.5500 | lim_sup: 10120.4500
Límites IQR -> lim_inf: 289.4800 | lim_sup: 289.4800
Límites IQR -> lim_inf: -165.0000 | lim_sup: 355.0000
Límites IQR -> lim_inf: -311.8763 | lim_sup: 635.9738
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 15.0000 | lim_sup: 15.0000
Límites IQR -> lim_inf: 89.8400 | lim_sup: 127.3600
Límites IQR -> lim_inf: 0.8612 | lim_sup: 75.6713
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -45.0000 | lim_sup: 75.0000
Límites IQR -> lim_inf: -367.7887 | lim_sup: 612.9813
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 68.0000 | lim_sup: 68.0000
Límites IQR -> lim_inf: -194.9475 | lim_sup: 1601.6325
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -30.0000 | lim_sup: 50.0000
Límites IQR -> lim_inf: -298.2300 | lim_sup: 497.0500
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 121.4300 | lim_sup: 121.4300
Límites IQR -> lim_inf: -285.2587 | lim_sup: 2059.8312
Límites IQR -> lim_inf: 52.0100 | lim_sup: 52.0100
Límites IQR -> lim_inf: -270.0000 | lim_sup: 450.0000
Límites IQR -> lim_inf: -120.0000 | lim_sup: 200.0000
Límites IQR -> lim_inf: -510.4050 | lim_sup: 850.6750
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -816.0413 | lim_sup: 2693.6888
Límites IQR -> lim_inf: 57.3700 | lim_sup: 57.3700
Límites IQR -> lim_inf: -135.0000 | lim_sup: 225.0000
Límites IQR -> lim_inf: -120.0000 | lim_sup: 200.0000
Límites IQR -> lim_inf: -331.9800 | lim_sup: 553.3000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -294.0900 | lim_sup: 2412.0300
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -120.0000 | lim_sup: 200.0000
Límites IQR -> lim_inf: -382.7212 | lim_sup: 637.8687
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -57.0725 | lim_sup: 185.7875
Límites IQR -> lim_inf: -166.6313 | lim_sup: 1586.9588
Límites IQR -> lim_inf: 50.4000 | lim_sup: 50.4000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -120.0000 | lim_sup: 200.0000
Límites IQR -> lim_inf: -328.5113 | lim_sup: 547.5187
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -285.0750 | lim_sup: 2420.7050
Límites IQR -> lim_inf: 45.1400 | lim_sup: 45.1400
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -45.0000 | lim_sup: 75.0000
Límites IQR -> lim_inf: -265.7250 | lim_sup: 442.8750
Límites IQR -> lim_inf: -116.7300 | lim_sup: 194.5500
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -116.8188 | lim_sup: 1935.0113
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -585.4425 | lim_sup: 975.7375
Límites IQR -> lim_inf: 110.0000 | lim_sup: 110.0000
Límites IQR -> lim_inf: -102.0000 | lim_sup: 170.0000
Límites IQR -> lim_inf: -615.6800 | lim_sup: 2887.4600
Límites IQR -> lim_inf: 49.6000 | lim_sup: 49.6000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -30.0000 | lim_sup: 50.0000
Límites IQR -> lim_inf: -279.1875 | lim_sup: 719.0325
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -95.0475 | lim_sup: 2124.6925
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -30.0000 | lim_sup: 50.0000
Límites IQR -> lim_inf: -420.2925 | lim_sup: 700.4875
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 121.4300 | lim_sup: 121.4300
Límites IQR -> lim_inf: -1183.0775 | lim_sup: 3482.3825
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -202.5000 | lim_sup: 337.5000
Límites IQR -> lim_inf: -494.7750 | lim_sup: 824.6250
Límites IQR -> lim_inf: 43.7500 | lim_sup: 43.7500
Límites IQR -> lim_inf: 96.8750 | lim_sup: 117.8750
Límites IQR -> lim_inf: -45.5363 | lim_sup: 75.8938
Límites IQR -> lim_inf: -361.5350 | lim_sup: 2290.5450
Límites IQR -> lim_inf: 77.1600 | lim_sup: 77.1600
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -150.0000 | lim_sup: 250.0000
Límites IQR -> lim_inf: -566.8438 | lim_sup: 1363.5263
Límites IQR -> lim_inf: 110.0000 | lim_sup: 110.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -260.0725 | lim_sup: 3017.1275
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -88.1250 | lim_sup: 146.8750
Límites IQR -> lim_inf: -305.2912 | lim_sup: 508.8188
Límites IQR -> lim_inf: 110.0000 | lim_sup: 110.0000
Límites IQR -> lim_inf: -45.5363 | lim_sup: 75.8938
Límites IQR -> lim_inf: 138.6525 | lim_sup: 1647.8325
Límites IQR -> lim_inf: 59.9100 | lim_sup: 59.9100
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -120.0000 | lim_sup: 200.0000
Límites IQR -> lim_inf: -332.4600 | lim_sup: 554.1000
Límites IQR -> lim_inf: 110.0000 | lim_sup: 110.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -291.6137 | lim_sup: 2579.4762
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -120.0000 | lim_sup: 200.0000
Límites IQR -> lim_inf: -329.2500 | lim_sup: 548.7500
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -105.0450 | lim_sup: 175.0750
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -5.6913 | lim_sup: 1353.5387
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -120.0000 | lim_sup: 200.0000
Límites IQR -> lim_inf: -368.2800 | lim_sup: 613.8000
Límites IQR -> lim_inf: 110.0000 | lim_sup: 110.0000
Límites IQR -> lim_inf: -91.0725 | lim_sup: 151.7875
Límites IQR -> lim_inf: -204.5338 | lim_sup: 1704.9763
Límites IQR -> lim_inf: -95.2500 | lim_sup: 158.7500
Límites IQR -> lim_inf: -93.0000 | lim_sup: 155.0000
Límites IQR -> lim_inf: -217.5000 | lim_sup: 362.5000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 121.4300 | lim_sup: 121.4300
Límites IQR -> lim_inf: -127.1625 | lim_sup: 1515.7975
Límites IQR -> lim_inf: 44.1400 | lim_sup: 44.1400
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -110.1375 | lim_sup: 183.5625
Límites IQR -> lim_inf: -197.4975 | lim_sup: 329.1625
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 121.4300 | lim_sup: 121.4300
Límites IQR -> lim_inf: -165.1300 | lim_sup: 1942.0900
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -150.0000 | lim_sup: 250.0000
Límites IQR -> lim_inf: -631.0538 | lim_sup: 1051.7562
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -251.1162 | lim_sup: 2123.0138
Límites IQR -> lim_inf: 104.2800 | lim_sup: 104.2800
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -209.4750 | lim_sup: 349.1250
Límites IQR -> lim_inf: -608.9688 | lim_sup: 1074.8412
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -159.9600 | lim_sup: 706.6000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -1026.1950 | lim_sup: 4600.7850
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -135.4950 | lim_sup: 225.8250
Límites IQR -> lim_inf: -350.5238 | lim_sup: 584.2062
Límites IQR -> lim_inf: 32.7500 | lim_sup: 238.7500
Límites IQR -> lim_inf: -203.5725 | lim_sup: 339.2875
Límites IQR -> lim_inf: -270.7175 | lim_sup: 2244.8425
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -133.5750 | lim_sup: 222.6250
Límites IQR -> lim_inf: -200.6662 | lim_sup: 334.4438
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 110.0000 | lim_sup: 110.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -72.5350 | lim_sup: 1877.3450
Límites IQR -> lim_inf: 49.8800 | lim_sup: 49.8800
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -128.2500 | lim_sup: 213.7500
Límites IQR -> lim_inf: -440.6775 | lim_sup: 734.4625
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 110.0000 | lim_sup: 110.0000
Límites IQR -> lim_inf: -102.0000 | lim_sup: 170.0000
Límites IQR -> lim_inf: 30.8113 | lim_sup: 1904.5412
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -150.0000 | lim_sup: 250.0000
Límites IQR -> lim_inf: -225.9937 | lim_sup: 376.6562
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 6.1938 | lim_sup: 160.3438
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -16.0075 | lim_sup: 1741.7525
Límites IQR -> lim_inf: -313.1250 | lim_sup: 521.8750
Límites IQR -> lim_inf: -150.0000 | lim_sup: 250.0000
Límites IQR -> lim_inf: -400.4925 | lim_sup: 667.4875
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 121.4300 | lim_sup: 121.4300
Límites IQR -> lim_inf: -255.1075 | lim_sup: 2114.9725
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -137.0400 | lim_sup: 228.4000
Límites IQR -> lim_inf: -412.2013 | lim_sup: 777.2088
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 92.5000 | lim_sup: 120.5000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -136.4962 | lim_sup: 2299.1137
Límites IQR -> lim_inf: 48.7400 | lim_sup: 48.7400
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -164.7188 | lim_sup: 274.5312
Límites IQR -> lim_inf: -153.1650 | lim_sup: 255.2750
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 55.0000 | lim_sup: 55.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -479.5562 | lim_sup: 1934.9337
Límites IQR -> lim_inf: 55.7000 | lim_sup: 55.7000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -105.0000 | lim_sup: 175.0000
Límites IQR -> lim_inf: -405.5475 | lim_sup: 852.4725
Límites IQR -> lim_inf: -87.5475 | lim_sup: 145.9125
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -99.4400 | lim_sup: 2087.7800
Límites IQR -> lim_inf: 40.1700 | lim_sup: 40.1700
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -106.5000 | lim_sup: 177.5000
Límites IQR -> lim_inf: -357.3250 | lim_sup: 736.8750
Límites IQR -> lim_inf: -116.7300 | lim_sup: 194.5500
Límites IQR -> lim_inf: -102.0000 | lim_sup: 170.0000
Límites IQR -> lim_inf: -322.3438 | lim_sup: 2121.1463
Límites IQR -> lim_inf: 45.6600 | lim_sup: 45.6600
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -45.0000 | lim_sup: 75.0000
Límites IQR -> lim_inf: -427.6425 | lim_sup: 712.7375
Límites IQR -> lim_inf: -87.5475 | lim_sup: 145.9125
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -203.6937 | lim_sup: 1948.9762
Límites IQR -> lim_inf: 62.0300 | lim_sup: 62.0300
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -45.0000 | lim_sup: 75.0000
Límites IQR -> lim_inf: -445.1963 | lim_sup: 741.9938
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -355.5150 | lim_sup: 2656.7850
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -45.0000 | lim_sup: 75.0000
Límites IQR -> lim_inf: -323.7750 | lim_sup: 539.6250
Límites IQR -> lim_inf: -91.5000 | lim_sup: 152.5000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -292.5900 | lim_sup: 1822.4100
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -106.5000 | lim_sup: 177.5000
Límites IQR -> lim_inf: -340.8750 | lim_sup: 568.1250
Límites IQR -> lim_inf: -91.5000 | lim_sup: 152.5000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -141.9775 | lim_sup: 1563.4425
Límites IQR -> lim_inf: 72.0000 | lim_sup: 72.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -88.1250 | lim_sup: 146.8750
Límites IQR -> lim_inf: -604.5900 | lim_sup: 1007.6500
Límites IQR -> lim_inf: 110.0000 | lim_sup: 110.0000
Límites IQR -> lim_inf: -102.0000 | lim_sup: 170.0000
Límites IQR -> lim_inf: -505.2775 | lim_sup: 3428.9025
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -144.6600 | lim_sup: 241.1000
Límites IQR -> lim_inf: -322.2750 | lim_sup: 537.1250
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 102.7950 | lim_sup: 134.6750
Límites IQR -> lim_inf: -52.5000 | lim_sup: 87.5000
Límites IQR -> lim_inf: -21.2937 | lim_sup: 2468.4762
Límites IQR -> lim_inf: 50.6000 | lim_sup: 50.6000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -30.0000 | lim_sup: 50.0000
Límites IQR -> lim_inf: -516.1950 | lim_sup: 860.3250
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -76.5000 | lim_sup: 127.5000
Límites IQR -> lim_inf: -289.6950 | lim_sup: 2115.5050
Límites IQR -> lim_inf: 1112.0000 | lim_sup: 1112.0000
Límites IQR -> lim_inf: -200.2200 | lim_sup: 333.7000
Límites IQR -> lim_inf: -584.8500 | lim_sup: 974.7500
Límites IQR -> lim_inf: -956.7350 | lim_sup: 2099.8650
Límites IQR -> lim_inf: 361.9950 | lim_sup: 513.3150
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.8225 | lim_sup: 349.8025
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -120.0000 | lim_sup: 200.0000
Límites IQR -> lim_inf: -540.7500 | lim_sup: 901.2500
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -116.7300 | lim_sup: 194.5500
Límites IQR -> lim_inf: -102.0000 | lim_sup: 170.0000
Límites IQR -> lim_inf: -170.3413 | lim_sup: 1807.2088
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -174.0000 | lim_sup: 290.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -25.5000 | lim_sup: 42.5000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -190.6500 | lim_sup: 317.7500
Límites IQR -> lim_inf: -376.6163 | lim_sup: 627.6938
Límites IQR -> lim_inf: 110.0000 | lim_sup: 110.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -23.9438 | lim_sup: 1657.3663
Límites IQR -> lim_inf: 53.2100 | lim_sup: 53.2100
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -120.0000 | lim_sup: 200.0000
Límites IQR -> lim_inf: -340.2250 | lim_sup: 754.5350
Límites IQR -> lim_inf: 92.5000 | lim_sup: 120.5000
Límites IQR -> lim_inf: -45.5363 | lim_sup: 75.8938
Límites IQR -> lim_inf: -191.8325 | lim_sup: 2125.6275
Límites IQR -> lim_inf: -90.7500 | lim_sup: 151.2500
Límites IQR -> lim_inf: -60.0000 | lim_sup: 100.0000
Límites IQR -> lim_inf: -529.9750 | lim_sup: 1129.0650
Límites IQR -> lim_inf: 110.0000 | lim_sup: 110.0000
Límites IQR -> lim_inf: -187.6088 | lim_sup: 312.6813
Límites IQR -> lim_inf: -258.5287 | lim_sup: 2258.8812
Límites IQR -> lim_inf: 57.7100 | lim_sup: 57.7100
Límites IQR -> lim_inf: -90.7500 | lim_sup: 151.2500
Límites IQR -> lim_inf: -120.0000 | lim_sup: 200.0000
Límites IQR -> lim_inf: -426.7162 | lim_sup: 711.1937
Límites IQR -> lim_inf: 110.0000 | lim_sup: 110.0000
Límites IQR -> lim_inf: -102.1725 | lim_sup: 170.2875
Límites IQR -> lim_inf: -319.2900 | lim_sup: 2499.1700
Límites IQR -> lim_inf: 46.1400 | lim_sup: 46.1400
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -66.0000 | lim_sup: 110.0000
Límites IQR -> lim_inf: -429.1462 | lim_sup: 715.2437
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -290.3375 | lim_sup: 2020.0825
Límites IQR -> lim_inf: 40.4400 | lim_sup: 40.4400
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -120.0000 | lim_sup: 200.0000
Límites IQR -> lim_inf: -370.5750 | lim_sup: 617.6250
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 110.0000 | lim_sup: 110.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 19.1950 | lim_sup: 1657.5150
Límites IQR -> lim_inf: 45.2900 | lim_sup: 45.2900
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -120.0000 | lim_sup: 200.0000
Límites IQR -> lim_inf: -660.4725 | lim_sup: 1100.7875
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -146.0138 | lim_sup: 1962.2963
Límites IQR -> lim_inf: 59.5900 | lim_sup: 59.5900
Límites IQR -> lim_inf: -125.0000 | lim_sup: 375.0000
Límites IQR -> lim_inf: -13.2900 | lim_sup: 22.1500
Límites IQR -> lim_inf: -848.2500 | lim_sup: 1413.7500
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -79.0838 | lim_sup: 2399.9263
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -150.5475 | lim_sup: 250.9125
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 35.2488 | lim_sup: 1718.2987
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -120.0000 | lim_sup: 200.0000
Límites IQR -> lim_inf: -233.4150 | lim_sup: 389.0250
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -87.5362 | lim_sup: 145.8938
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -158.2400 | lim_sup: 1919.8000
Límites IQR -> lim_inf: -180.2775 | lim_sup: 300.4625
Límites IQR -> lim_inf: -1525.8600 | lim_sup: 2783.1000
Límites IQR -> lim_inf: -712.5425 | lim_sup: 1851.2375
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 1797.4900 | lim_sup: 5118.6100
Límites IQR -> lim_inf: -217.7775 | lim_sup: 362.9625
Límites IQR -> lim_inf: -1494.8288 | lim_sup: 7866.9213
Límites IQR -> lim_inf: 499.3900 | lim_sup: 499.3900
Límites IQR -> lim_inf: -75.0000 | lim_sup: 125.0000
Límites IQR -> lim_inf: -78.4900 | lim_sup: 1101.2900
Límites IQR -> lim_inf: -447.1200 | lim_sup: 2026.1200
Límites IQR -> lim_inf: 492.1600 | lim_sup: 492.1600
Límites IQR -> lim_inf: 4053.9725 | lim_sup: 7733.5125
Límites IQR -> lim_inf: -454.6200 | lim_sup: 757.7000
Límites IQR -> lim_inf: -2755.3350 | lim_sup: 20245.3650
Límites IQR -> lim_inf: 57.2300 | lim_sup: 57.2300
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -105.0000 | lim_sup: 175.0000
Límites IQR -> lim_inf: -198.0000 | lim_sup: 330.0000
Límites IQR -> lim_inf: -87.5475 | lim_sup: 145.9125
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -511.8400 | lim_sup: 2842.9400
Límites IQR -> lim_inf: -187.5000 | lim_sup: 562.5000
Límites IQR -> lim_inf: -64.5000 | lim_sup: 107.5000
Límites IQR -> lim_inf: -306.7500 | lim_sup: 511.2500
Límites IQR -> lim_inf: -68.6250 | lim_sup: 114.3750
Límites IQR -> lim_inf: 94.8000 | lim_sup: 94.8000
Límites IQR -> lim_inf: -325.7850 | lim_sup: 2373.5350
Límites IQR -> lim_inf: 79.8000 | lim_sup: 79.8000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -66.0000 | lim_sup: 110.0000
Límites IQR -> lim_inf: -276.0000 | lim_sup: 460.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -51.0075 | lim_sup: 85.0125
Límites IQR -> lim_inf: -409.6250 | lim_sup: 3023.4750
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 500.0000 | lim_sup: 500.0000
Límites IQR -> lim_inf: 85.2750 | lim_sup: 97.8750
Límites IQR -> lim_inf: 4.1087 | lim_sup: 236.4188
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -78.2000 | lim_sup: 363.4000
Límites IQR -> lim_inf: 250.0000 | lim_sup: 250.0000
Límites IQR -> lim_inf: -60.0000 | lim_sup: 100.0000
Límites IQR -> lim_inf: -509.8050 | lim_sup: 849.6750
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 60.3675 | lim_sup: 109.5075
Límites IQR -> lim_inf: 63.1500 | lim_sup: 63.1500
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -30.0000 | lim_sup: 50.0000
Límites IQR -> lim_inf: -694.0500 | lim_sup: 1156.7500
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -244.0088 | lim_sup: 2672.2013
Límites IQR -> lim_inf: 56.3000 | lim_sup: 56.3000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -328.9500 | lim_sup: 548.2500
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -87.5475 | lim_sup: 145.9125
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -149.3775 | lim_sup: 2298.2025
Límites IQR -> lim_inf: 259.1000 | lim_sup: 259.1000
Límites IQR -> lim_inf: -95.6250 | lim_sup: 159.3750
Límites IQR -> lim_inf: 30.8900 | lim_sup: 30.8900
Límites IQR -> lim_inf: 34.8700 | lim_sup: 34.8700
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -60.0000 | lim_sup: 100.0000
Límites IQR -> lim_inf: -437.2500 | lim_sup: 728.7500
Límites IQR -> lim_inf: 92.5000 | lim_sup: 120.5000
Límites IQR -> lim_inf: -25.5000 | lim_sup: 42.5000
Límites IQR -> lim_inf: -263.2650 | lim_sup: 1669.4550
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -146.1000 | lim_sup: 243.5000
Límites IQR -> lim_inf: -638.9250 | lim_sup: 1064.8750
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -235.6713 | lim_sup: 2385.8387
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -2866.8700 | lim_sup: 7031.2700
Límites IQR -> lim_inf: -3623.4200 | lim_sup: 13964.6800
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 16252.4512 | lim_sup: 22178.4813
Límites IQR -> lim_inf: -2398.3763 | lim_sup: 5225.8138
Límites IQR -> lim_inf: -6386.3850 | lim_sup: 62980.3950
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -15.0000 | lim_sup: 25.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 110.0000 | lim_sup: 110.0000
Límites IQR -> lim_inf: -84.0000 | lim_sup: 140.0000
Límites IQR -> lim_inf: -58.4862 | lim_sup: 2115.1237
Límites IQR -> lim_inf: -337.5000 | lim_sup: 562.5000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -128.4300 | lim_sup: 214.0500
Límites IQR -> lim_inf: 380.6650 | lim_sup: 455.5050
Límites IQR -> lim_inf: -207.7188 | lim_sup: 988.5713
Límites IQR -> lim_inf: 46.3600 | lim_sup: 46.3600
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -105.0000 | lim_sup: 175.0000
Límites IQR -> lim_inf: -373.2000 | lim_sup: 622.0000
Límites IQR -> lim_inf: -91.5000 | lim_sup: 152.5000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -300.9425 | lim_sup: 2147.5575
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -169.2975 | lim_sup: 282.1625
Límites IQR -> lim_inf: -405.9000 | lim_sup: 2381.3800
Límites IQR -> lim_inf: -1941.2175 | lim_sup: 3584.5225
Límites IQR -> lim_inf: -30.1275 | lim_sup: 50.2125
Límites IQR -> lim_inf: 2837.0700 | lim_sup: 5892.9100
Límites IQR -> lim_inf: -1088.9013 | lim_sup: 2425.1888
Límites IQR -> lim_inf: -6979.5900 | lim_sup: 24141.4900
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -150.7013 | lim_sup: 251.1688
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 180.0000 | lim_sup: 180.0000
Límites IQR -> lim_inf: -95.0188 | lim_sup: 752.7713
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 122.5400 | lim_sup: 173.7400
Límites IQR -> lim_inf: -45.8563 | lim_sup: 197.4938
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -65.0025 | lim_sup: 108.3375
Límites IQR -> lim_inf: 181.2000 | lim_sup: 256.9600
Límites IQR -> lim_inf: -30.4975 | lim_sup: 302.1625
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -174.4137 | lim_sup: 358.9963
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -66.6450 | lim_sup: 437.0750
Límites IQR -> lim_inf: 59.3750 | lim_sup: 534.3750
Límites IQR -> lim_inf: -269.3688 | lim_sup: 691.4813
Límites IQR -> lim_inf: 48.3700 | lim_sup: 48.3700
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -137.1750 | lim_sup: 228.6250
Límites IQR -> lim_inf: -422.9625 | lim_sup: 704.9375
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 110.0000 | lim_sup: 110.0000
Límites IQR -> lim_inf: -45.5363 | lim_sup: 75.8938
Límites IQR -> lim_inf: -53.2537 | lim_sup: 1918.4962
Límites IQR -> lim_inf: 67.2100 | lim_sup: 67.2100
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -162.7350 | lim_sup: 271.2250
Límites IQR -> lim_inf: -525.2400 | lim_sup: 875.4000
Límites IQR -> lim_inf: 110.0000 | lim_sup: 110.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -126.1250 | lim_sup: 2747.5150
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -1850.0062 | lim_sup: 5907.8237
Límites IQR -> lim_inf: -4016.0213 | lim_sup: 14363.3888
Límites IQR -> lim_inf: -2688.0212 | lim_sup: 7716.6687
Límites IQR -> lim_inf: 6871.5487 | lim_sup: 15718.7788
Límites IQR -> lim_inf: -1670.5725 | lim_sup: 2784.2875
Límites IQR -> lim_inf: -5462.8038 | lim_sup: 53688.6062
Límites IQR -> lim_inf: -78.4648 | lim_sup: 727.8898
Límites IQR -> lim_inf: 49.0909 | lim_sup: 100.0000
Límites IQR -> lim_inf: -473.6420 | lim_sup: 3737.8216
Límites IQR -> lim_inf: 4223.9602 | lim_sup: 4722.0057
Límites IQR -> lim_inf: -338.2159 | lim_sup: 563.6932
Límites IQR -> lim_inf: 11589.4614 | lim_sup: 17066.7159
Límites IQR -> lim_inf: 54.3602 | lim_sup: 506.5148
Límites IQR -> lim_inf: 49.0909 | lim_sup: 100.0000
Límites IQR -> lim_inf: -1165.9727 | lim_sup: 3651.3727
Límites IQR -> lim_inf: 2773.1591 | lim_sup: 7044.2500
Límites IQR -> lim_inf: -849.9682 | lim_sup: 1416.6136
Límites IQR -> lim_inf: 9924.4443 | lim_sup: 12992.1261
Límites IQR -> lim_inf: 80.9205 | lim_sup: 462.2477
Límites IQR -> lim_inf: 66.3636 | lim_sup: 131.8182
Límites IQR -> lim_inf: -1648.3989 | lim_sup: 4000.6648
Límites IQR -> lim_inf: 4403.5273 | lim_sup: 6487.4545
Límites IQR -> lim_inf: -772.8614 | lim_sup: 1288.1023
Límites IQR -> lim_inf: 7914.0795 | lim_sup: 12708.0068
Límites IQR -> lim_inf: 97.6750 | lim_sup: 514.2750
Límites IQR -> lim_inf: -472.7273 | lim_sup: 1003.6364
Límites IQR -> lim_inf: -1302.1648 | lim_sup: 5904.8080
Límites IQR -> lim_inf: 7186.9602 | lim_sup: 8112.3693
Límites IQR -> lim_inf: -941.8977 | lim_sup: 1569.8295
Límites IQR -> lim_inf: 15028.6500 | lim_sup: 21696.3591
Límites IQR -> lim_inf: -8.1420 | lim_sup: 222.5398
Límites IQR -> lim_inf: 66.3636 | lim_sup: 131.8182
Límites IQR -> lim_inf: -948.8216 | lim_sup: 2161.3693
Límites IQR -> lim_inf: 3446.7273 | lim_sup: 3810.3636
Límites IQR -> lim_inf: -115.9091 | lim_sup: 193.1818
Límites IQR -> lim_inf: 4688.1659 | lim_sup: 8851.8023
Límites IQR -> lim_inf: -78.7864 | lim_sup: 507.5773
Límites IQR -> lim_inf: 66.3636 | lim_sup: 131.8182
Límites IQR -> lim_inf: -947.5864 | lim_sup: 2449.4864
Límites IQR -> lim_inf: 3570.0455 | lim_sup: 3851.8636
Límites IQR -> lim_inf: -136.3636 | lim_sup: 227.2727
Límites IQR -> lim_inf: 6163.0011 | lim_sup: 10976.5011
Límites IQR -> lim_inf: -284.7716 | lim_sup: 838.8375
Límites IQR -> lim_inf: 88.6364 | lim_sup: 143.1818
Límites IQR -> lim_inf: -1328.3636 | lim_sup: 4016.5818
Límites IQR -> lim_inf: 5021.6830 | lim_sup: 5666.0011
Límites IQR -> lim_inf: -409.3977 | lim_sup: 682.3295
Límites IQR -> lim_inf: 8423.3534 | lim_sup: 12828.7625
Límites IQR -> lim_inf: 107.9136 | lim_sup: 320.6773
Límites IQR -> lim_inf: 88.6364 | lim_sup: 143.1818
Límites IQR -> lim_inf: -1369.6386 | lim_sup: 3682.5614
Límites IQR -> lim_inf: 4811.7455 | lim_sup: 5553.4909
Límites IQR -> lim_inf: -197.7273 | lim_sup: 329.5455
Límites IQR -> lim_inf: 6913.3420 | lim_sup: 11787.3875
Límites IQR -> lim_inf: 146.6932 | lim_sup: 521.9841
Límites IQR -> lim_inf: 1142.2727 | lim_sup: 1193.1818
Límites IQR -> lim_inf: -1832.0648 | lim_sup: 6377.8716
Límites IQR -> lim_inf: 5382.6705 | lim_sup: 5666.2159
Límites IQR -> lim_inf: -567.6136 | lim_sup: 946.0227
Límites IQR -> lim_inf: 15157.1000 | lim_sup: 27218.6818
Límites IQR -> lim_inf: -123.3182 | lim_sup: 802.6455
Límites IQR -> lim_inf: -104.5455 | lim_sup: 390.0000
Límites IQR -> lim_inf: -1075.1830 | lim_sup: 3544.4625
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 4423.0455 | lim_sup: 9191.2636
Límites IQR -> lim_inf: -545.4545 | lim_sup: 909.0909
Límites IQR -> lim_inf: 15465.7261 | lim_sup: 19526.0261
Límites IQR -> lim_inf: 89.4216 | lim_sup: 415.8125
Límites IQR -> lim_inf: 49.0909 | lim_sup: 100.0000
Límites IQR -> lim_inf: -2375.2227 | lim_sup: 6025.1045
Límites IQR -> lim_inf: 3860.3182 | lim_sup: 8106.5000
Límites IQR -> lim_inf: -255.6818 | lim_sup: 426.1364
Límites IQR -> lim_inf: 8730.4330 | lim_sup: 15897.4602
Límites IQR -> lim_inf: 146.6932 | lim_sup: 521.9841
Límites IQR -> lim_inf: 49.0909 | lim_sup: 100.0000
Límites IQR -> lim_inf: -536.2784 | lim_sup: 3668.2398
Límites IQR -> lim_inf: 6364.0455 | lim_sup: 6890.2273
Límites IQR -> lim_inf: -255.6818 | lim_sup: 426.1364
Límites IQR -> lim_inf: 15458.2125 | lim_sup: 19323.1489
Límites IQR -> lim_inf: 139.1705 | lim_sup: 438.4977
Límites IQR -> lim_inf: -209.0909 | lim_sup: 634.5455
Límites IQR -> lim_inf: -1145.5273 | lim_sup: 3088.7273
Límites IQR -> lim_inf: 4415.2955 | lim_sup: 6228.7864
Límites IQR -> lim_inf: -982.7455 | lim_sup: 1637.9091
Límites IQR -> lim_inf: 6271.6716 | lim_sup: 13027.6807
Límites IQR -> lim_inf: 17.4443 | lim_sup: 665.3261
Límites IQR -> lim_inf: -267.7273 | lim_sup: 732.2727
Límites IQR -> lim_inf: -708.7023 | lim_sup: 5273.9341
Límites IQR -> lim_inf: 6099.0466 | lim_sup: 6886.6011
Límites IQR -> lim_inf: -618.3136 | lim_sup: 1030.5227
Límites IQR -> lim_inf: 11479.3034 | lim_sup: 16475.0216
Límites IQR -> lim_inf: 15.8148 | lim_sup: 668.0420
Límites IQR -> lim_inf: 49.0909 | lim_sup: 100.0000
Límites IQR -> lim_inf: -2811.5080 | lim_sup: 6390.5375
Límites IQR -> lim_inf: 5933.7295 | lim_sup: 7636.3295
Límites IQR -> lim_inf: -1237.8182 | lim_sup: 2569.0909
Límites IQR -> lim_inf: 12384.0216 | lim_sup: 16511.0852
Límites IQR -> lim_inf: 90.6864 | lim_sup: 225.1227
Límites IQR -> lim_inf: 31.8182 | lim_sup: 68.1818
Límites IQR -> lim_inf: -947.8136 | lim_sup: 2184.0409
Límites IQR -> lim_inf: 3613.8989 | lim_sup: 3712.8352
Límites IQR -> lim_inf: -204.5455 | lim_sup: 340.9091
Límites IQR -> lim_inf: 5819.6932 | lim_sup: 12270.8568
Límites IQR -> lim_inf: 493.5364 | lim_sup: 493.5364
Límites IQR -> lim_inf: 49.0909 | lim_sup: 100.0000
Límites IQR -> lim_inf: -1749.9773 | lim_sup: 6816.7136
Límites IQR -> lim_inf: 63.6364 | lim_sup: 63.6364
Límites IQR -> lim_inf: 7163.3636 | lim_sup: 7327.7273
Límites IQR -> lim_inf: -606.3727 | lim_sup: 1478.5000
Límites IQR -> lim_inf: 16345.3420 | lim_sup: 22563.2420
Límites IQR -> lim_inf: 62.2341 | lim_sup: 735.5250
Límites IQR -> lim_inf: 155.0273 | lim_sup: 155.0273
Límites IQR -> lim_inf: -1013.2864 | lim_sup: 2421.2955
Límites IQR -> lim_inf: 3547.0682 | lim_sup: 3618.8864
Límites IQR -> lim_inf: -216.2045 | lim_sup: 360.3409
Límites IQR -> lim_inf: 4740.0614 | lim_sup: 12041.9523
Límites IQR -> lim_inf: 54.3636 | lim_sup: 506.5091
Límites IQR -> lim_inf: 297.0000 | lim_sup: 297.0000
Límites IQR -> lim_inf: -2198.3091 | lim_sup: 6257.9818
Límites IQR -> lim_inf: 3541.1955 | lim_sup: 7804.3591
Límites IQR -> lim_inf: -681.8182 | lim_sup: 1136.3636
Límites IQR -> lim_inf: 19210.0216 | lim_sup: 27446.9761
Límites IQR -> lim_inf: 158.3750 | lim_sup: 406.4659
Límites IQR -> lim_inf: -1759.3000 | lim_sup: 4447.3182
Límites IQR -> lim_inf: -1279.0148 | lim_sup: 8934.7580
Límites IQR -> lim_inf: 5952.3773 | lim_sup: 7305.0682
Límites IQR -> lim_inf: -1399.4318 | lim_sup: 2332.3864
Límites IQR -> lim_inf: 14728.9841 | lim_sup: 20924.2568
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -120.0000 | lim_sup: 200.0000
Límites IQR -> lim_inf: -337.1475 | lim_sup: 561.9125
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 121.4300 | lim_sup: 121.4300
Límites IQR -> lim_inf: -113.1263 | lim_sup: 1307.6038
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -136.2000 | lim_sup: 227.0000
Límites IQR -> lim_inf: -390.4162 | lim_sup: 650.6937
Límites IQR -> lim_inf: 110.0000 | lim_sup: 110.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -261.9800 | lim_sup: 1538.2600
Límites IQR -> lim_inf: 49.3600 | lim_sup: 49.3600
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -66.0000 | lim_sup: 110.0000
Límites IQR -> lim_inf: -946.7250 | lim_sup: 1577.8750
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -102.0000 | lim_sup: 170.0000
Límites IQR -> lim_inf: -254.2625 | lim_sup: 2258.5375
Límites IQR -> lim_inf: 53.8000 | lim_sup: 53.8000
Límites IQR -> lim_inf: -219.3750 | lim_sup: 365.6250
Límites IQR -> lim_inf: -66.0000 | lim_sup: 110.0000
Límites IQR -> lim_inf: -508.8750 | lim_sup: 848.1250
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -497.2663 | lim_sup: 2576.1038
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -105.0000 | lim_sup: 175.0000
Límites IQR -> lim_inf: -1502.1700 | lim_sup: 3344.9500
Límites IQR -> lim_inf: 74.4200 | lim_sup: 74.4200
Límites IQR -> lim_inf: 74.4200 | lim_sup: 74.4200
Límites IQR -> lim_inf: -20.9000 | lim_sup: 252.4000
Límites IQR -> lim_inf: 63.3400 | lim_sup: 63.3400
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -15.0000 | lim_sup: 25.0000
Límites IQR -> lim_inf: -204.3750 | lim_sup: 340.6250
Límites IQR -> lim_inf: 55.5350 | lim_sup: 200.7750
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -205.6362 | lim_sup: 2659.4137
Límites IQR -> lim_inf: 63.4800 | lim_sup: 63.4800
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -66.0000 | lim_sup: 110.0000
Límites IQR -> lim_inf: -793.6350 | lim_sup: 1322.7250
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 121.4300 | lim_sup: 121.4300
Límites IQR -> lim_inf: -402.5250 | lim_sup: 2861.2750
Límites IQR -> lim_inf: -11.5216 | lim_sup: 725.6511
Límites IQR -> lim_inf: 49.0909 | lim_sup: 100.0000
Límites IQR -> lim_inf: -1882.0568 | lim_sup: 4329.4886
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 4839.2909 | lim_sup: 5052.7455
Límites IQR -> lim_inf: -383.5227 | lim_sup: 639.2045
Límites IQR -> lim_inf: 10837.6182 | lim_sup: 16536.1818
Límites IQR -> lim_inf: 70.3364 | lim_sup: 210.1182
Límites IQR -> lim_inf: 454.5455 | lim_sup: 454.5455
Límites IQR -> lim_inf: -781.5443 | lim_sup: 1302.5739
Límites IQR -> lim_inf: 990.3273 | lim_sup: 1012.5091
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 2163.8898 | lim_sup: 6641.4261
Límites IQR -> lim_inf: -90.2466 | lim_sup: 593.7807
Límites IQR -> lim_inf: 500.0000 | lim_sup: 500.0000
Límites IQR -> lim_inf: -1490.7716 | lim_sup: 4042.0375
Límites IQR -> lim_inf: 4857.6909 | lim_sup: 6371.7273
Límites IQR -> lim_inf: -514.3602 | lim_sup: 857.2670
Límites IQR -> lim_inf: 23130.3398 | lim_sup: 30480.4580
Límites IQR -> lim_inf: -33.2913 | lim_sup: 233.0391
Límites IQR -> lim_inf: -718.3863 | lim_sup: 1784.0059
Límites IQR -> lim_inf: 600.0000 | lim_sup: 600.0000
Límites IQR -> lim_inf: -203.8134 | lim_sup: 375.3050
Límites IQR -> lim_inf: 1524.8668 | lim_sup: 9757.3909
Límites IQR -> lim_inf: -158.2797 | lim_sup: 1107.9579
Límites IQR -> lim_inf: 481.3956 | lim_sup: 3502.6966
Límites IQR -> lim_inf: 600.0000 | lim_sup: 600.0000
Límites IQR -> lim_inf: -576.0102 | lim_sup: 1060.8170
Límites IQR -> lim_inf: 678.8923 | lim_sup: 21975.3839
Límites IQR -> lim_inf: -47.3087 | lim_sup: 331.1606
Límites IQR -> lim_inf: -261.1628 | lim_sup: 925.8316
Límites IQR -> lim_inf: -117.3223 | lim_sup: 199.9373
Límites IQR -> lim_inf: -16688.1843 | lim_sup: 36930.8978
Límites IQR -> lim_inf: -35.6274 | lim_sup: 249.3922
Límites IQR -> lim_inf: -418.2738 | lim_sup: 1691.8452
Límites IQR -> lim_inf: 21.8218 | lim_sup: 21.8218
Límites IQR -> lim_inf: -109.4932 | lim_sup: 218.8886
Límites IQR -> lim_inf: 719.5326 | lim_sup: 3930.2477
Límites IQR -> lim_inf: -23.3623 | lim_sup: 163.5361
Límites IQR -> lim_inf: -311.5708 | lim_sup: 769.7238
Límites IQR -> lim_inf: -60.4950 | lim_sup: 100.8250
Límites IQR -> lim_inf: 262.7332 | lim_sup: 4374.9385
Límites IQR -> lim_inf: -217.2699 | lim_sup: 362.1165
Límites IQR -> lim_inf: -360.4392 | lim_sup: 833.2920
Límites IQR -> lim_inf: -66.2400 | lim_sup: 110.4000
Límites IQR -> lim_inf: 841.6291 | lim_sup: 4622.6102
Límites IQR -> lim_inf: -157.6953 | lim_sup: 262.8255
Límites IQR -> lim_inf: -381.4738 | lim_sup: 1230.3306
Límites IQR -> lim_inf: 19.4718 | lim_sup: 19.4718
Límites IQR -> lim_inf: -63.7874 | lim_sup: 149.6190
Límites IQR -> lim_inf: 776.2703 | lim_sup: 5358.2934
Límites IQR -> lim_inf: 50.3649 | lim_sup: 140.0585
Límites IQR -> lim_inf: -1138.2285 | lim_sup: 3333.4507
Límites IQR -> lim_inf: -276.4539 | lim_sup: 513.5565
Límites IQR -> lim_inf: 42.3995 | lim_sup: 11437.4144
Límites IQR -> lim_inf: 128.4932 | lim_sup: 128.4932
Límites IQR -> lim_inf: -844.5531 | lim_sup: 2164.3242
Límites IQR -> lim_inf: 540.0000 | lim_sup: 540.0000
Límites IQR -> lim_inf: 4482.9500 | lim_sup: 5048.2300
Límites IQR -> lim_inf: -335.6320 | lim_sup: 761.9947
Límites IQR -> lim_inf: -3916.3894 | lim_sup: 8592.0772
Límites IQR -> lim_inf: -15.7695 | lim_sup: 110.3869
Límites IQR -> lim_inf: -424.9990 | lim_sup: 837.3802
Límites IQR -> lim_inf: -54.7800 | lim_sup: 91.3000
Límites IQR -> lim_inf: -6327.3608 | lim_sup: 13746.4885
Límites IQR -> lim_inf: 272.2647 | lim_sup: 998.3039
Límites IQR -> lim_inf: -1110.7491 | lim_sup: 5542.1039
Límites IQR -> lim_inf: -51.3009 | lim_sup: 708.3003
Límites IQR -> lim_inf: -742.0072 | lim_sup: 1394.0680
Límites IQR -> lim_inf: 3347.0410 | lim_sup: 18768.4620
Límites IQR -> lim_inf: -94.6170 | lim_sup: 157.6950
Límites IQR -> lim_inf: -221.2094 | lim_sup: 521.2239
Límites IQR -> lim_inf: 600.0000 | lim_sup: 600.0000
Límites IQR -> lim_inf: -56.4450 | lim_sup: 107.5150
Límites IQR -> lim_inf: -99.1214 | lim_sup: 1713.1180
Límites IQR -> lim_inf: -70.0869 | lim_sup: 116.8115
Límites IQR -> lim_inf: -341.1032 | lim_sup: 775.3864
Límites IQR -> lim_inf: 600.0000 | lim_sup: 600.0000
Límites IQR -> lim_inf: 9.0664 | lim_sup: 9.0664
Límites IQR -> lim_inf: -22.3114 | lim_sup: 51.3126
Límites IQR -> lim_inf: 55.5272 | lim_sup: 1742.0928
Límites IQR -> lim_inf: -186.3538 | lim_sup: 1304.4766
Límites IQR -> lim_inf: -334.4193 | lim_sup: 5377.0956
Límites IQR -> lim_inf: -678.1050 | lim_sup: 1351.4150
Límites IQR -> lim_inf: 4996.9363 | lim_sup: 38972.5667
Límites IQR -> lim_inf: -112.6199 | lim_sup: 788.3397
Límites IQR -> lim_inf: -907.1314 | lim_sup: 5419.2924
Límites IQR -> lim_inf: -400.7044 | lim_sup: 852.1826
Límites IQR -> lim_inf: 1031.1659 | lim_sup: 28946.6712
Límites IQR -> lim_inf: -333.5667 | lim_sup: 555.9445
Límites IQR -> lim_inf: -612.5437 | lim_sup: 1162.0137
Límites IQR -> lim_inf: -97.0500 | lim_sup: 161.7500
Límites IQR -> lim_inf: 129.3809 | lim_sup: 5845.1243
Límites IQR -> lim_inf: 148.2518 | lim_sup: 148.2518
Límites IQR -> lim_inf: -511.7603 | lim_sup: 1424.6309
Límites IQR -> lim_inf: -139.7042 | lim_sup: 264.1993
Límites IQR -> lim_inf: 700.8835 | lim_sup: 8748.4127
Límites IQR -> lim_inf: 332.5078 | lim_sup: 332.5078
Límites IQR -> lim_inf: -1161.4739 | lim_sup: 3013.3435
Límites IQR -> lim_inf: -155.5613 | lim_sup: 318.3544
Límites IQR -> lim_inf: 470.8174 | lim_sup: 11285.0646
Límites IQR -> lim_inf: -127.0734 | lim_sup: 211.7890
Límites IQR -> lim_inf: -880.1862 | lim_sup: 1915.3694
Límites IQR -> lim_inf: -60.1740 | lim_sup: 124.2900
Límites IQR -> lim_inf: 286.5616 | lim_sup: 3863.3696
Límites IQR -> lim_inf: -77.0955 | lim_sup: 128.4925
Límites IQR -> lim_inf: -276.5793 | lim_sup: 645.2618
Límites IQR -> lim_inf: 9.0664 | lim_sup: 9.0664
Límites IQR -> lim_inf: -37.6544 | lim_sup: 71.1200
Límites IQR -> lim_inf: 67.8274 | lim_sup: 1908.7162
Límites IQR -> lim_inf: -127.0731 | lim_sup: 211.7885
Límites IQR -> lim_inf: -265.9262 | lim_sup: 913.6388
Límites IQR -> lim_inf: -52.4550 | lim_sup: 87.4250
Límites IQR -> lim_inf: 749.7281 | lim_sup: 4717.6644
Límites IQR -> lim_inf: -222.3777 | lim_sup: 370.6295
Límites IQR -> lim_inf: -304.2792 | lim_sup: 741.5398
Límites IQR -> lim_inf: -61.5000 | lim_sup: 102.5000
Límites IQR -> lim_inf: 949.0393 | lim_sup: 4414.3407
Límites IQR -> lim_inf: -231.2880 | lim_sup: 385.4800
Límites IQR -> lim_inf: -541.9647 | lim_sup: 1158.1929
Límites IQR -> lim_inf: 16.1146 | lim_sup: 16.1146
Límites IQR -> lim_inf: -90.7163 | lim_sup: 151.1938
Límites IQR -> lim_inf: 827.5626 | lim_sup: 2819.3616
Límites IQR -> lim_inf: -109.0487 | lim_sup: 181.7477
Límites IQR -> lim_inf: -341.9025 | lim_sup: 848.8116
Límites IQR -> lim_inf: 9.0664 | lim_sup: 9.0664
Límites IQR -> lim_inf: -70.7147 | lim_sup: 138.0178
Límites IQR -> lim_inf: -98.3517 | lim_sup: 2488.1572
Límites IQR -> lim_inf: -291.2778 | lim_sup: 485.4630
Límites IQR -> lim_inf: -421.1808 | lim_sup: 1052.6701
Límites IQR -> lim_inf: 16.7860 | lim_sup: 16.7860
Límites IQR -> lim_inf: -154.3500 | lim_sup: 257.2500
Límites IQR -> lim_inf: 660.3736 | lim_sup: 3256.8457
Límites IQR -> lim_inf: -19.0050 | lim_sup: 133.0350
Límites IQR -> lim_inf: -183.8153 | lim_sup: 537.3483
Límites IQR -> lim_inf: -51.1500 | lim_sup: 136.8500
Límites IQR -> lim_inf: 829.5660 | lim_sup: 4243.1468
Límites IQR -> lim_inf: -229.6203 | lim_sup: 534.4755
Límites IQR -> lim_inf: -50.0900 | lim_sup: 130.1500
Límites IQR -> lim_inf: 564.3260 | lim_sup: 2403.7551
Límites IQR -> lim_inf: -58.0000 | lim_sup: 406.0000
Límites IQR -> lim_inf: -638.6961 | lim_sup: 2332.5657
Límites IQR -> lim_inf: -106.7716 | lim_sup: 241.9526
Límites IQR -> lim_inf: 2243.6962 | lim_sup: 13536.4277
Límites IQR -> lim_inf: -20.0050 | lim_sup: 140.0350
Límites IQR -> lim_inf: -335.7836 | lim_sup: 776.2780
Límites IQR -> lim_inf: -45.1500 | lim_sup: 75.2500
Límites IQR -> lim_inf: 701.9466 | lim_sup: 4255.6957
Límites IQR -> lim_inf: -350.3445 | lim_sup: 583.9075
Límites IQR -> lim_inf: -1026.2798 | lim_sup: 1959.0736
Límites IQR -> lim_inf: 600.0000 | lim_sup: 600.0000
Límites IQR -> lim_inf: -49.0215 | lim_sup: 81.7025
Límites IQR -> lim_inf: -979.0801 | lim_sup: 6534.4252
Límites IQR -> lim_inf: -72.0000 | lim_sup: 504.0000
Límites IQR -> lim_inf: -240.5427 | lim_sup: 1565.7370
Límites IQR -> lim_inf: -82.7485 | lim_sup: 184.4475
Límites IQR -> lim_inf: 926.0737 | lim_sup: 10537.6305
Límites IQR -> lim_inf: -46.9008 | lim_sup: 328.3056
Límites IQR -> lim_inf: -296.7311 | lim_sup: 989.0867
Límites IQR -> lim_inf: -61.5000 | lim_sup: 102.5000
Límites IQR -> lim_inf: 1297.2673 | lim_sup: 8991.2037
Límites IQR -> lim_inf: -37.8497 | lim_sup: 264.9479
Límites IQR -> lim_inf: -443.3872 | lim_sup: 1038.0243
Límites IQR -> lim_inf: -46.8100 | lim_sup: 101.3500
Límites IQR -> lim_inf: 2018.1117 | lim_sup: 7013.9297
Límites IQR -> lim_inf: -343.6431 | lim_sup: 572.7385
Límites IQR -> lim_inf: -165.2105 | lim_sup: 989.6028
Límites IQR -> lim_inf: -290.0700 | lim_sup: 483.4500
Límites IQR -> lim_inf: 392.1476 | lim_sup: 5565.4242
Límites IQR -> lim_inf: -335.1810 | lim_sup: 558.6350
Límites IQR -> lim_inf: -705.6902 | lim_sup: 1907.9418
Límites IQR -> lim_inf: -95.6755 | lim_sup: 265.5925
Límites IQR -> lim_inf: 1086.8251 | lim_sup: 6543.4780
Límites IQR -> lim_inf: -162.0000 | lim_sup: 270.0000
Límites IQR -> lim_inf: -483.9000 | lim_sup: 1046.4200
Límites IQR -> lim_inf: -66.0000 | lim_sup: 110.0000
Límites IQR -> lim_inf: 400.3801 | lim_sup: 4204.8891
Límites IQR -> lim_inf: -173.3781 | lim_sup: 288.9635
Límites IQR -> lim_inf: -345.8750 | lim_sup: 1374.3250
Límites IQR -> lim_inf: -159.6000 | lim_sup: 266.0000
Límites IQR -> lim_inf: 340.3329 | lim_sup: 4949.5072
Límites IQR -> lim_inf: -111.8865 | lim_sup: 186.4775
Límites IQR -> lim_inf: -437.4019 | lim_sup: 951.6571
Límites IQR -> lim_inf: -45.7500 | lim_sup: 76.2500
Límites IQR -> lim_inf: 505.7838 | lim_sup: 2604.0563
Límites IQR -> lim_inf: -135.0600 | lim_sup: 225.1000
Límites IQR -> lim_inf: -407.6751 | lim_sup: 1033.3835
Límites IQR -> lim_inf: 600.0000 | lim_sup: 600.0000
Límites IQR -> lim_inf: -106.5180 | lim_sup: 177.5300
Límites IQR -> lim_inf: 389.4878 | lim_sup: 3662.8607
Límites IQR -> lim_inf: -627.0000 | lim_sup: 1045.0000
Límites IQR -> lim_inf: -302.0639 | lim_sup: 2105.6653
Límites IQR -> lim_inf: 600.0000 | lim_sup: 600.0000
Límites IQR -> lim_inf: -210.4650 | lim_sup: 403.5750
Límites IQR -> lim_inf: 1166.0206 | lim_sup: 11666.8983
Límites IQR -> lim_inf: 174.0000 | lim_sup: 174.0000
Límites IQR -> lim_inf: -231.5199 | lim_sup: 876.2680
Límites IQR -> lim_inf: -59.3670 | lim_sup: 98.9450
Límites IQR -> lim_inf: 273.8211 | lim_sup: 5180.2838
Límites IQR -> lim_inf: -156.8502 | lim_sup: 261.4170
Límites IQR -> lim_inf: -285.5328 | lim_sup: 545.2214
Límites IQR -> lim_inf: -48.9060 | lim_sup: 81.5100
Límites IQR -> lim_inf: 266.5693 | lim_sup: 1275.3148
Límites IQR -> lim_inf: 168.6800 | lim_sup: 168.6800
Límites IQR -> lim_inf: -481.2080 | lim_sup: 1500.3869
Límites IQR -> lim_inf: 600.0000 | lim_sup: 600.0000
Límites IQR -> lim_inf: -178.5500 | lim_sup: 356.2500
Límites IQR -> lim_inf: 1044.6321 | lim_sup: 5760.5788
Límites IQR -> lim_inf: -81.7680 | lim_sup: 572.3756
Límites IQR -> lim_inf: -700.5010 | lim_sup: 1970.4350
Límites IQR -> lim_inf: 20.1214 | lim_sup: 20.1214
Límites IQR -> lim_inf: -91.4810 | lim_sup: 186.0683
Límites IQR -> lim_inf: -957.6424 | lim_sup: 7385.6745
Límites IQR -> lim_inf: -714.6867 | lim_sup: 1191.1445
Límites IQR -> lim_inf: -821.6814 | lim_sup: 2984.1356
Límites IQR -> lim_inf: -571.7395 | lim_sup: 1292.5735
Límites IQR -> lim_inf: 1893.7400 | lim_sup: 13766.8407
Límites IQR -> lim_inf: -234.1854 | lim_sup: 390.3090
Límites IQR -> lim_inf: -496.4801 | lim_sup: 1282.6945
Límites IQR -> lim_inf: -94.3170 | lim_sup: 157.1950
Límites IQR -> lim_inf: 512.7268 | lim_sup: 3499.4292
Límites IQR -> lim_inf: -195.4479 | lim_sup: 325.7465
Límites IQR -> lim_inf: -266.3696 | lim_sup: 1065.4494
Límites IQR -> lim_inf: -77.5125 | lim_sup: 161.1875
Límites IQR -> lim_inf: 1346.0170 | lim_sup: 4130.7403
Límites IQR -> lim_inf: -434.5404 | lim_sup: 724.2340
Límites IQR -> lim_inf: -609.5787 | lim_sup: 2432.9287
Límites IQR -> lim_inf: 728.0000 | lim_sup: 728.0000
Límites IQR -> lim_inf: -42.8675 | lim_sup: 78.9125
Límites IQR -> lim_inf: 593.1029 | lim_sup: 8633.6112
Límites IQR -> lim_inf: -241.4388 | lim_sup: 402.3980
Límites IQR -> lim_inf: -414.5511 | lim_sup: 1166.6647
Límites IQR -> lim_inf: 3570.8524 | lim_sup: 3570.8524
Límites IQR -> lim_inf: -50.1025 | lim_sup: 109.3175
Límites IQR -> lim_inf: 1386.3636 | lim_sup: 5258.7166
Límites IQR -> lim_inf: -26.9857 | lim_sup: 188.8996
Límites IQR -> lim_inf: -868.1122 | lim_sup: 2129.5316
Límites IQR -> lim_inf: -58.3525 | lim_sup: 131.7075
Límites IQR -> lim_inf: 831.6707 | lim_sup: 5503.3348
Límites IQR -> lim_inf: -149.0352 | lim_sup: 248.3920
Límites IQR -> lim_inf: -409.2908 | lim_sup: 906.7342
Límites IQR -> lim_inf: -47.8455 | lim_sup: 79.7425
Límites IQR -> lim_inf: 537.7572 | lim_sup: 3102.0457
Límites IQR -> lim_inf: -194.9445 | lim_sup: 324.9075
Límites IQR -> lim_inf: -241.7390 | lim_sup: 752.1823
Límites IQR -> lim_inf: -155.0332 | lim_sup: 258.3886
Límites IQR -> lim_inf: 661.1363 | lim_sup: 3620.4024
Límites IQR -> lim_inf: 186.2136 | lim_sup: 186.2136
Límites IQR -> lim_inf: -533.4899 | lim_sup: 1082.0439
Límites IQR -> lim_inf: -94.5360 | lim_sup: 157.5600
Límites IQR -> lim_inf: 476.4825 | lim_sup: 2442.2941
Límites IQR -> lim_inf: -487.2789 | lim_sup: 812.1315
Límites IQR -> lim_inf: -1052.0436 | lim_sup: 5848.1686
Límites IQR -> lim_inf: 3651.8672 | lim_sup: 3651.8672
Límites IQR -> lim_inf: -260.3801 | lim_sup: 1689.9668
Límites IQR -> lim_inf: -4876.2243 | lim_sup: 17467.9138
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -143.6778 | lim_sup: 1005.7442
Límites IQR -> lim_inf: -1169.3594 | lim_sup: 4922.8686
Límites IQR -> lim_inf: -396.2008 | lim_sup: 738.3614
Límites IQR -> lim_inf: -253.1984 | lim_sup: 13458.3776
Límites IQR -> lim_inf: -73.5913 | lim_sup: 515.1391
Límites IQR -> lim_inf: -1185.0446 | lim_sup: 4522.1446
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -340.2436 | lim_sup: 934.7706
Límites IQR -> lim_inf: 2884.6029 | lim_sup: 17479.8962
Límites IQR -> lim_inf: -465.0300 | lim_sup: 775.0500
Límites IQR -> lim_inf: 379.9358 | lim_sup: 2861.1980
Límites IQR -> lim_inf: -465.7900 | lim_sup: 826.9700
Límites IQR -> lim_inf: 584.5825 | lim_sup: 19484.8433
Límites IQR -> lim_inf: -336.5400 | lim_sup: 560.9000
Límites IQR -> lim_inf: -35.1786 | lim_sup: 58.6310
Límites IQR -> lim_inf: 104.3332 | lim_sup: 586.9484
Límites IQR -> lim_inf: -32.0100 | lim_sup: 224.0700
Límites IQR -> lim_inf: -459.8602 | lim_sup: 1194.6662
Límites IQR -> lim_inf: 600.0000 | lim_sup: 600.0000
Límites IQR -> lim_inf: -81.4000 | lim_sup: 189.0000
Límites IQR -> lim_inf: 73.7798 | lim_sup: 7146.1719
Límites IQR -> lim_inf: -570.6009 | lim_sup: 951.0015
Límites IQR -> lim_inf: -887.4266 | lim_sup: 4641.7528
Límites IQR -> lim_inf: 600.0000 | lim_sup: 600.0000
Límites IQR -> lim_inf: -252.6766 | lim_sup: 702.9282
Límites IQR -> lim_inf: 2968.0838 | lim_sup: 13410.0592
Límites IQR -> lim_inf: -495.0000 | lim_sup: 825.0000
Límites IQR -> lim_inf: -580.7972 | lim_sup: 3940.9784
Límites IQR -> lim_inf: -183.0685 | lim_sup: 352.0475
Límites IQR -> lim_inf: 689.1187 | lim_sup: 18632.4870
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -558.8550 | lim_sup: 931.4250
Límites IQR -> lim_inf: -220.5750 | lim_sup: 786.5050
Límites IQR -> lim_inf: -46.2375 | lim_sup: 77.0625
Límites IQR -> lim_inf: -194.6587 | lim_sup: 1164.9913
Límites IQR -> lim_inf: 37.6900 | lim_sup: 37.6900
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -121.1475 | lim_sup: 201.9125
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 121.4300 | lim_sup: 121.4300
Límites IQR -> lim_inf: 2.2512 | lim_sup: 1529.2812
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 122.4100 | lim_sup: 122.4100
Límites IQR -> lim_inf: 728.2700 | lim_sup: 728.2700
Límites IQR -> lim_inf: 963.0000 | lim_sup: 963.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 885.9050 | lim_sup: 1521.8650
Límites IQR -> lim_inf: -1133.6600 | lim_sup: 9278.6400
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -139.2250 | lim_sup: 525.3750
Límites IQR -> lim_inf: -1264.8750 | lim_sup: 2108.1250
Límites IQR -> lim_inf: -836.0725 | lim_sup: 35642.0675
Límites IQR -> lim_inf: 580.8900 | lim_sup: 580.8900
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -116.4375 | lim_sup: 194.0625
Límites IQR -> lim_inf: -3.4650 | lim_sup: 545.7750
Límites IQR -> lim_inf: -218.6400 | lim_sup: 364.4000
Límites IQR -> lim_inf: -78.3075 | lim_sup: 395.7925
Límites IQR -> lim_inf: 623.0000 | lim_sup: 623.0000
Límites IQR -> lim_inf: -499.7250 | lim_sup: 832.8750
Límites IQR -> lim_inf: -1269.8825 | lim_sup: 5284.3375
Límites IQR -> lim_inf: -1322.4775 | lim_sup: 4629.3425
Límites IQR -> lim_inf: 3967.5900 | lim_sup: 5626.2300
Límites IQR -> lim_inf: -714.4725 | lim_sup: 1190.7875
Límites IQR -> lim_inf: 291.1325 | lim_sup: 20970.4125
Límites IQR -> lim_inf: -51.8909 | lim_sup: 683.6000
Límites IQR -> lim_inf: 49.0909 | lim_sup: 100.0000
Límites IQR -> lim_inf: -1080.2193 | lim_sup: 2780.5716
Límites IQR -> lim_inf: 3778.4432 | lim_sup: 6773.5523
Límites IQR -> lim_inf: -92.0455 | lim_sup: 153.4091
Límites IQR -> lim_inf: 7109.1352 | lim_sup: 12021.1625
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -66.0000 | lim_sup: 110.0000
Límites IQR -> lim_inf: -218.0400 | lim_sup: 363.4000
Límites IQR -> lim_inf: 240.0000 | lim_sup: 240.0000
Límites IQR -> lim_inf: -32.6850 | lim_sup: 54.4750
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -84.8600 | lim_sup: 1005.2200
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -320.4050 | lim_sup: 719.5950
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 128.0000 | lim_sup: 128.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -196.8750 | lim_sup: 328.1250
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -264.1650 | lim_sup: 440.2750
Límites IQR -> lim_inf: -236.0250 | lim_sup: 393.3750
Límites IQR -> lim_inf: -76.5000 | lim_sup: 127.5000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -120.0000 | lim_sup: 200.0000
Límites IQR -> lim_inf: -326.7325 | lim_sup: 752.5875
Límites IQR -> lim_inf: 57.5000 | lim_sup: 141.5000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -38.2137 | lim_sup: 1273.0963
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -120.0000 | lim_sup: 200.0000
Límites IQR -> lim_inf: -236.9700 | lim_sup: 394.9500
Límites IQR -> lim_inf: 110.0000 | lim_sup: 110.0000
Límites IQR -> lim_inf: -91.0725 | lim_sup: 151.7875
Límites IQR -> lim_inf: -174.2562 | lim_sup: 1372.7737
Límites IQR -> lim_inf: -135.0000 | lim_sup: 405.0000
Límites IQR -> lim_inf: -120.0000 | lim_sup: 200.0000
Límites IQR -> lim_inf: -310.3162 | lim_sup: 517.1937
Límites IQR -> lim_inf: 83.7500 | lim_sup: 125.7500
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 32.3838 | lim_sup: 1155.9938
Límites IQR -> lim_inf: 125.0000 | lim_sup: 125.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -299.2500 | lim_sup: 498.7500
Límites IQR -> lim_inf: 121.4300 | lim_sup: 121.4300
Límites IQR -> lim_inf: -137.8775 | lim_sup: 639.2425
Límites IQR -> lim_inf: -201.1994 | lim_sup: 1408.3954
Límites IQR -> lim_inf: -384.8433 | lim_sup: 3164.3068
Límites IQR -> lim_inf: -264.0350 | lim_sup: 535.7650
Límites IQR -> lim_inf: 4381.5471 | lim_sup: 27170.4688
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -393.2250 | lim_sup: 655.3750
Límites IQR -> lim_inf: -2219.4288 | lim_sup: 7252.3813
Límites IQR -> lim_inf: 192.4700 | lim_sup: 192.4700
Límites IQR -> lim_inf: -76.5000 | lim_sup: 127.5000
Límites IQR -> lim_inf: -157.8650 | lim_sup: 519.3950
Límites IQR -> lim_inf: 14.4688 | lim_sup: 207.8787
Límites IQR -> lim_inf: -362.7675 | lim_sup: 1731.6125
Límites IQR -> lim_inf: -281.2500 | lim_sup: 468.7500
Límites IQR -> lim_inf: -90.0000 | lim_sup: 150.0000
Límites IQR -> lim_inf: -525.4987 | lim_sup: 875.8312
Límites IQR -> lim_inf: -37.5000 | lim_sup: 62.5000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -65.6375 | lim_sup: 1738.7825
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -1443.0000 | lim_sup: 2405.0000
Límites IQR -> lim_inf: -1301.8663 | lim_sup: 3220.1237
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 4161.2000 | lim_sup: 5175.1200
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -498.0763 | lim_sup: 8530.2938
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 73.3163 | lim_sup: 744.4262
Límites IQR -> lim_inf: -1445.5875 | lim_sup: 3643.7925
Límites IQR -> lim_inf: 6655.3763 | lim_sup: 8774.5863
Límites IQR -> lim_inf: -102.0000 | lim_sup: 170.0000
Límites IQR -> lim_inf: -1261.6275 | lim_sup: 13266.7325
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -262.5000 | lim_sup: 437.5000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 490.9900 | lim_sup: 490.9900
Límites IQR -> lim_inf: 68.0000 | lim_sup: 68.0000
Límites IQR -> lim_inf: -64.4288 | lim_sup: 483.2613
Límites IQR -> lim_inf: -143.2463 | lim_sup: 595.9038
Límites IQR -> lim_inf: -112.5000 | lim_sup: 187.5000
Límites IQR -> lim_inf: -768.5512 | lim_sup: 1365.8387
Límites IQR -> lim_inf: -77.2500 | lim_sup: 128.7500
Límites IQR -> lim_inf: -182.1450 | lim_sup: 303.5750
Límites IQR -> lim_inf: -259.1050 | lim_sup: 2376.8150
Límites IQR -> lim_inf: -209.8300 | lim_sup: 2542.4500
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -33.7500 | lim_sup: 56.2500
Límites IQR -> lim_inf: -90.0000 | lim_sup: 150.0000
Límites IQR -> lim_inf: 1347.6300 | lim_sup: 1347.6300
Límites IQR -> lim_inf: -70.9287 | lim_sup: 1752.0212
Límites IQR -> lim_inf: 133.8500 | lim_sup: 133.8500
Límites IQR -> lim_inf: -1043.6250 | lim_sup: 1739.3750
Límites IQR -> lim_inf: 149.3000 | lim_sup: 170.4200
Límites IQR -> lim_inf: -75.0000 | lim_sup: 125.0000
Límites IQR -> lim_inf: 42.9725 | lim_sup: 255.9925
Límites IQR -> lim_inf: -67.2550 | lim_sup: 200.2650
Límites IQR -> lim_inf: -827.4150 | lim_sup: 2886.0850
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -130.2600 | lim_sup: 217.1000
Límites IQR -> lim_inf: -242.0850 | lim_sup: 403.4750
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 134.0950 | lim_sup: 1109.5950
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -232.5525 | lim_sup: 387.5875
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -102.0000 | lim_sup: 170.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -162.7887 | lim_sup: 1714.5212
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -585.0000 | lim_sup: 975.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 41.2850 | lim_sup: 148.1450
Límites IQR -> lim_inf: -171.2050 | lim_sup: 2151.8750
Límites IQR -> lim_inf: 8.6475 | lim_sup: 1490.2075
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -967.5450 | lim_sup: 1612.5750
Límites IQR -> lim_inf: 222.4500 | lim_sup: 222.4500
Límites IQR -> lim_inf: -71.3000 | lim_sup: 628.4600
Límites IQR -> lim_inf: -43.0813 | lim_sup: 2124.8288
Límites IQR -> lim_inf: -151.8775 | lim_sup: 1485.8025
Límites IQR -> lim_inf: -318.7937 | lim_sup: 797.9895
Límites IQR -> lim_inf: -461.3539 | lim_sup: 1198.1499
Límites IQR -> lim_inf: -522.4154 | lim_sup: 870.6923
Límites IQR -> lim_inf: 35.2946 | lim_sup: 649.7694
Límites IQR -> lim_inf: -638.4982 | lim_sup: 3287.5965
Límites IQR -> lim_inf: -1637.9151 | lim_sup: 3007.7281
Límites IQR -> lim_inf: -67.0829 | lim_sup: 2872.2447
Límites IQR -> lim_inf: 424.2097 | lim_sup: 1187.0656
Límites IQR -> lim_inf: 545.7771 | lim_sup: 2443.9108
Límites IQR -> lim_inf: -22106.9368 | lim_sup: 42355.7637
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 7215.8293 | lim_sup: 13263.5111
Límites IQR -> lim_inf: -1225.6041 | lim_sup: 20899.9174
Límites IQR -> lim_inf: 8135.5775 | lim_sup: 37699.6975
Límites IQR -> lim_inf: -4675.0834 | lim_sup: 12358.7371
Límites IQR -> lim_inf: 23881.0866 | lim_sup: 36322.8821
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 1496.0625 | lim_sup: 2972.6265
Límites IQR -> lim_inf: -1154.3155 | lim_sup: 4569.4105
Límites IQR -> lim_inf: 1734.9315 | lim_sup: 13118.5635
Límites IQR -> lim_inf: -789.5250 | lim_sup: 1315.8750
Límites IQR -> lim_inf: 6858.9603 | lim_sup: 16588.5708
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 2020.5182 | lim_sup: 4054.6723
Límites IQR -> lim_inf: -2130.3115 | lim_sup: 7046.3825
Límites IQR -> lim_inf: 1213.8072 | lim_sup: 11199.0292
Límites IQR -> lim_inf: -446.1937 | lim_sup: 743.6562
Límites IQR -> lim_inf: -34.1898 | lim_sup: 20933.2217
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 1393.0722 | lim_sup: 2588.9743
Límites IQR -> lim_inf: -1289.2332 | lim_sup: 4440.0993
Límites IQR -> lim_inf: 887.1992 | lim_sup: 12994.9533
Límites IQR -> lim_inf: -439.2000 | lim_sup: 765.6000
Límites IQR -> lim_inf: 4698.4569 | lim_sup: 11411.6159
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 1308.2203 | lim_sup: 2585.3102
Límites IQR -> lim_inf: -586.7794 | lim_sup: 2728.2186
Límites IQR -> lim_inf: -455.5115 | lim_sup: 9462.9005
Límites IQR -> lim_inf: -851.9474 | lim_sup: 2072.4136
Límites IQR -> lim_inf: 2625.4943 | lim_sup: 7437.9418
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 1611.2515 | lim_sup: 3210.3595
Límites IQR -> lim_inf: -914.5504 | lim_sup: 4157.1136
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 138.4235 | lim_sup: 14203.0075
Límites IQR -> lim_inf: -2580.7500 | lim_sup: 5133.4900
Límites IQR -> lim_inf: 3496.8356 | lim_sup: 8313.3641
Límites IQR -> lim_inf: -29.9337 | lim_sup: 89.8012
Límites IQR -> lim_inf: 1946.9631 | lim_sup: 2526.5140
Límites IQR -> lim_inf: -1257.7777 | lim_sup: 2096.2962
Límites IQR -> lim_inf: 1498.1733 | lim_sup: 5755.3952
Límites IQR -> lim_inf: -300.0000 | lim_sup: 500.0000
Límites IQR -> lim_inf: -9001.4993 | lim_sup: 18290.2808
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 1218.8029 | lim_sup: 2323.8524
Límites IQR -> lim_inf: -460.8044 | lim_sup: 2592.4976
Límites IQR -> lim_inf: 586.9685 | lim_sup: 9218.8245
Límites IQR -> lim_inf: -300.0000 | lim_sup: 500.0000
Límites IQR -> lim_inf: 6383.9218 | lim_sup: 14749.5357
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 3767.4883 | lim_sup: 7093.0094
Límites IQR -> lim_inf: -204.5977 | lim_sup: 4867.6628
Límites IQR -> lim_inf: 2831.1945 | lim_sup: 18016.9265
Límites IQR -> lim_inf: 95.1500 | lim_sup: 814.7500
Límites IQR -> lim_inf: 14399.3074 | lim_sup: 28060.3544
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 2237.5606 | lim_sup: 4328.3902
Límites IQR -> lim_inf: 1371.4962 | lim_sup: 3338.4372
Límites IQR -> lim_inf: 1173.4317 | lim_sup: 12481.7298
Límites IQR -> lim_inf: -956.6250 | lim_sup: 1594.3750
Límites IQR -> lim_inf: 10277.0139 | lim_sup: 18535.5399
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -610.2143 | lim_sup: 1017.0238
Límites IQR -> lim_inf: 678.5714 | lim_sup: 678.5714
Límites IQR -> lim_inf: 46.3595 | lim_sup: 81.4643
Límites IQR -> lim_inf: -166.0714 | lim_sup: 276.7857
Límites IQR -> lim_inf: 1299.4101 | lim_sup: 2803.4958
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -840.3271 | lim_sup: 1932.0896
Límites IQR -> lim_inf: -122.0286 | lim_sup: 203.3810
Límites IQR -> lim_inf: -84.0976 | lim_sup: 219.2738
Límites IQR -> lim_inf: 1354.5000 | lim_sup: 9111.0381
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -532.1429 | lim_sup: 886.9048
Límites IQR -> lim_inf: -871.9164 | lim_sup: 2492.9336
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -273.8298 | lim_sup: 596.5798
Límites IQR -> lim_inf: -143.7500 | lim_sup: 239.5833
Límites IQR -> lim_inf: 3274.3714 | lim_sup: 8895.5810
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 1545.3795 | lim_sup: 3007.2435
Límites IQR -> lim_inf: 283.6354 | lim_sup: 1798.0424
Límites IQR -> lim_inf: 1131.2415 | lim_sup: 7206.9895
Límites IQR -> lim_inf: -477.0000 | lim_sup: 795.0000
Límites IQR -> lim_inf: 4338.0351 | lim_sup: 10026.5596
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -1077.9196 | lim_sup: 1842.9423
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -11.6488 | lim_sup: 165.4464
Límites IQR -> lim_inf: -24.2134 | lim_sup: 40.3557
Límites IQR -> lim_inf: 1961.5929 | lim_sup: 4743.4500
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -194.7467 | lim_sup: 502.9509
Límites IQR -> lim_inf: -93.0000 | lim_sup: 155.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 1148.9101 | lim_sup: 3365.2292
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -978.0560 | lim_sup: 1793.5345
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -148.5536 | lim_sup: 247.5893
Límites IQR -> lim_inf: -9.4446 | lim_sup: 15.7411
Límites IQR -> lim_inf: 1984.2527 | lim_sup: 3913.0503
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 951.1523 | lim_sup: 3272.6982
Límites IQR -> lim_inf: -1060.7377 | lim_sup: 2822.5445
Límites IQR -> lim_inf: 809.0523 | lim_sup: 9344.9462
Límites IQR -> lim_inf: -1151.8069 | lim_sup: 2392.4781
Límites IQR -> lim_inf: 5165.2600 | lim_sup: 9315.9340
Límites IQR -> lim_inf: -250.0000 | lim_sup: 416.6667
Límites IQR -> lim_inf: -396.9643 | lim_sup: 1175.8929
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 99.0357 | lim_sup: 99.0357
Límites IQR -> lim_inf: -8.1857 | lim_sup: 13.6429
Límites IQR -> lim_inf: 3941.0583 | lim_sup: 8546.7964
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 1935.4921 | lim_sup: 3641.3843
Límites IQR -> lim_inf: -1714.2721 | lim_sup: 4953.9329
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 1233.7462 | lim_sup: 10734.8603
Límites IQR -> lim_inf: -418.7160 | lim_sup: 697.8600
Límites IQR -> lim_inf: 7810.8384 | lim_sup: 13905.7674
Límites IQR -> lim_inf: -646.4286 | lim_sup: 1077.3810
Límites IQR -> lim_inf: -952.4155 | lim_sup: 4285.6607
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -145.3488 | lim_sup: 448.3560
Límites IQR -> lim_inf: -127.8006 | lim_sup: 301.8899
Límites IQR -> lim_inf: 4490.6199 | lim_sup: 14753.3842
Límites IQR -> lim_inf: -589.2857 | lim_sup: 982.1429
Límites IQR -> lim_inf: -636.9702 | lim_sup: 3856.5107
Límites IQR -> lim_inf: 53.9143 | lim_sup: 198.0476
Límites IQR -> lim_inf: -289.8199 | lim_sup: 623.4205
Límites IQR -> lim_inf: 1137.4872 | lim_sup: 9710.1753
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 1635.5222 | lim_sup: 3297.9243
Límites IQR -> lim_inf: -1396.4631 | lim_sup: 4025.0964
Límites IQR -> lim_inf: 505.7418 | lim_sup: 5088.0597
Límites IQR -> lim_inf: -565.9875 | lim_sup: 943.3125
Límites IQR -> lim_inf: 1894.0085 | lim_sup: 13474.9085
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 1778.8872 | lim_sup: 4413.6493
Límites IQR -> lim_inf: -1765.1070 | lim_sup: 6316.4710
Límites IQR -> lim_inf: 1205.1735 | lim_sup: 7301.9055
Límites IQR -> lim_inf: -493.6125 | lim_sup: 822.6875
Límites IQR -> lim_inf: 1100.8856 | lim_sup: 18134.4131
Límites IQR -> lim_inf: -233.4150 | lim_sup: 389.0250
Límites IQR -> lim_inf: 4884.3655 | lim_sup: 9538.3473
Límites IQR -> lim_inf: -1022.8106 | lim_sup: 10812.3787
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 8363.5725 | lim_sup: 18420.4405
Límites IQR -> lim_inf: -1245.8925 | lim_sup: 2076.4875
Límites IQR -> lim_inf: 16654.8824 | lim_sup: 40057.2029
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 1609.4072 | lim_sup: 3017.4453
Límites IQR -> lim_inf: -1748.9474 | lim_sup: 4768.3411
Límites IQR -> lim_inf: 1258.3353 | lim_sup: 11832.0652
Límites IQR -> lim_inf: -736.5000 | lim_sup: 1227.5000
Límites IQR -> lim_inf: 6346.6662 | lim_sup: 15566.6483
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -31.2612 | lim_sup: 197.5167
Límites IQR -> lim_inf: -952.6012 | lim_sup: 1587.6687
Límites IQR -> lim_inf: 297.9223 | lim_sup: 1414.8722
Límites IQR -> lim_inf: 172.0000 | lim_sup: 172.0000
Límites IQR -> lim_inf: 459.2051 | lim_sup: 710.5691
Límites IQR -> lim_inf: -154.5673 | lim_sup: 289.6635
Límites IQR -> lim_inf: -178.8658 | lim_sup: 829.2407
Límites IQR -> lim_inf: 264.2577 | lim_sup: 341.5192
Límites IQR -> lim_inf: 66.5180 | lim_sup: 747.6542
Límites IQR -> lim_inf: -556.7966 | lim_sup: 7593.7100
Límites IQR -> lim_inf: -28.8462 | lim_sup: 248.0769
Límites IQR -> lim_inf: -729.2069 | lim_sup: 2412.2394
Límites IQR -> lim_inf: 383.3538 | lim_sup: 474.9231
Límites IQR -> lim_inf: 324.4446 | lim_sup: 1373.9458
Límites IQR -> lim_inf: -1742.8088 | lim_sup: 21296.3054
Límites IQR -> lim_inf: 62.5000 | lim_sup: 162.5000
Límites IQR -> lim_inf: -397.6913 | lim_sup: 662.8189
Límites IQR -> lim_inf: -156.2308 | lim_sup: 260.3846
Límites IQR -> lim_inf: -169.1215 | lim_sup: 281.8692
Límites IQR -> lim_inf: 427.5192 | lim_sup: 2339.3164
Límites IQR -> lim_inf: -392.8500 | lim_sup: 654.7500
Límites IQR -> lim_inf: -369.5280 | lim_sup: 773.8866
Límites IQR -> lim_inf: 208.9492 | lim_sup: 299.5462
Límites IQR -> lim_inf: -122.5904 | lim_sup: 490.8143
Límites IQR -> lim_inf: -936.9137 | lim_sup: 8912.2846
Límites IQR -> lim_inf: -621.2981 | lim_sup: 1609.8558
Límites IQR -> lim_inf: -765.6118 | lim_sup: 2111.4789
Límites IQR -> lim_inf: 486.0077 | lim_sup: 606.1923
Límites IQR -> lim_inf: 367.9668 | lim_sup: 1949.2740
Límites IQR -> lim_inf: -1087.5495 | lim_sup: 25144.6128
Límites IQR -> lim_inf: -86.5385 | lim_sup: 144.2308
Límites IQR -> lim_inf: -346.3145 | lim_sup: 961.9156
Límites IQR -> lim_inf: 335.0308 | lim_sup: 415.1538
Límites IQR -> lim_inf: 92.8057 | lim_sup: 869.1534
Límites IQR -> lim_inf: 2018.0516 | lim_sup: 7046.8290
Límites IQR -> lim_inf: -165.4196 | lim_sup: 521.8531
Límites IQR -> lim_inf: -696.6983 | lim_sup: 16389.2792
Límites IQR -> lim_inf: 5132.7723 | lim_sup: 18518.4942
Límites IQR -> lim_inf: -665.2702 | lim_sup: 1108.7837
Límites IQR -> lim_inf: 20846.3565 | lim_sup: 43239.3970
Límites IQR -> lim_inf: -180.0000 | lim_sup: 300.0000
Límites IQR -> lim_inf: 1851.2023 | lim_sup: 3713.8962
Límites IQR -> lim_inf: -1935.5948 | lim_sup: 7564.4409
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -2374.4112 | lim_sup: 16437.1908
Límites IQR -> lim_inf: -1265.7606 | lim_sup: 2109.6009
Límites IQR -> lim_inf: 7816.6240 | lim_sup: 18670.4200
Límites IQR -> lim_inf: -101.1189 | lim_sup: 291.6084
Límites IQR -> lim_inf: -1442.1884 | lim_sup: 4110.8267
Límites IQR -> lim_inf: 2398.7115 | lim_sup: 4794.9885
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 2776.0965 | lim_sup: 11316.9181
Límites IQR -> lim_inf: -101.1189 | lim_sup: 291.6084
Límites IQR -> lim_inf: -1918.8150 | lim_sup: 6024.9268
Límites IQR -> lim_inf: 6627.5923 | lim_sup: 11541.5000
Límites IQR -> lim_inf: -24.1087 | lim_sup: 40.1812
Límites IQR -> lim_inf: 14040.8347 | lim_sup: 28314.8155
Límites IQR -> lim_inf: 74.1600 | lim_sup: 74.1600
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -95.8350 | lim_sup: 159.7250
Límites IQR -> lim_inf: -465.4050 | lim_sup: 775.6750
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 110.0000 | lim_sup: 110.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -615.0750 | lim_sup: 3689.3850
Límites IQR -> lim_inf: 54.8600 | lim_sup: 54.8600
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -140.5688 | lim_sup: 234.2813
Límites IQR -> lim_inf: -336.9412 | lim_sup: 561.5687
Límites IQR -> lim_inf: 110.0000 | lim_sup: 110.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -906.3775 | lim_sup: 3535.5025
Límites IQR -> lim_inf: 54.8600 | lim_sup: 54.8600
Límites IQR -> lim_inf: 221.0500 | lim_sup: 221.0500
Límites IQR -> lim_inf: -161.3775 | lim_sup: 268.9625
Límites IQR -> lim_inf: -209.7000 | lim_sup: 349.5000
Límites IQR -> lim_inf: 110.0000 | lim_sup: 110.0000
Límites IQR -> lim_inf: -45.5550 | lim_sup: 75.9250
Límites IQR -> lim_inf: -319.0263 | lim_sup: 2236.3238
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -120.0000 | lim_sup: 200.0000
Límites IQR -> lim_inf: -364.2262 | lim_sup: 607.0438
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 31.0925 | lim_sup: 1356.8325
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -130.2600 | lim_sup: 217.1000
Límites IQR -> lim_inf: -379.8750 | lim_sup: 633.1250
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 110.0000 | lim_sup: 110.0000
Límites IQR -> lim_inf: -217.2337 | lim_sup: 362.0562
Límites IQR -> lim_inf: -294.4738 | lim_sup: 2620.9563
Límites IQR -> lim_inf: 46.1538 | lim_sup: 46.1538
Límites IQR -> lim_inf: -265.3846 | lim_sup: 442.3077
Límites IQR -> lim_inf: 276.9231 | lim_sup: 523.0769
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -653.6159 | lim_sup: 1585.8079
Límites IQR -> lim_inf: -101.1189 | lim_sup: 291.6084
Límites IQR -> lim_inf: -942.9461 | lim_sup: 2111.7199
Límites IQR -> lim_inf: 4643.4385 | lim_sup: 7180.3615
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 4581.4446 | lim_sup: 8809.2348
Límites IQR -> lim_inf: -101.1189 | lim_sup: 291.6084
Límites IQR -> lim_inf: -1007.0168 | lim_sup: 3937.0039
Límites IQR -> lim_inf: 2603.9096 | lim_sup: 5174.7619
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 2879.3087 | lim_sup: 10297.7595
Límites IQR -> lim_inf: 53.5700 | lim_sup: 53.5700
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -682.5000 | lim_sup: 1137.5000
Límites IQR -> lim_inf: -501.6000 | lim_sup: 879.8000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 110.0000 | lim_sup: 110.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 117.0938 | lim_sup: 2036.7837
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -120.0000 | lim_sup: 200.0000
Límites IQR -> lim_inf: -562.2938 | lim_sup: 937.1562
Límites IQR -> lim_inf: -192.0000 | lim_sup: 320.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 111.8962 | lim_sup: 1548.3063
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -120.0000 | lim_sup: 200.0000
Límites IQR -> lim_inf: -253.4663 | lim_sup: 422.4438
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -471.9325 | lim_sup: 2092.1675
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -181.3200 | lim_sup: 302.2000
Límites IQR -> lim_inf: -438.0000 | lim_sup: 730.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 189.4300 | lim_sup: 189.4300
Límites IQR -> lim_inf: -142.7800 | lim_sup: 1746.5400
Límites IQR -> lim_inf: 66.4600 | lim_sup: 66.4600
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -303.3000 | lim_sup: 505.5000
Límites IQR -> lim_inf: -474.8025 | lim_sup: 791.3375
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -345.0375 | lim_sup: 2851.9825
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -657.2110 | lim_sup: 1521.0057
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -175.2393 | lim_sup: 292.0655
Límites IQR -> lim_inf: -145.3598 | lim_sup: 242.2664
Límites IQR -> lim_inf: 4329.0557 | lim_sup: 9224.7152
Límites IQR -> lim_inf: -385.4799 | lim_sup: 642.4665
Límites IQR -> lim_inf: -629.1030 | lim_sup: 1705.9232
Límites IQR -> lim_inf: 182.8480 | lim_sup: 182.8480
Límites IQR -> lim_inf: -123.0686 | lim_sup: 333.7810
Límites IQR -> lim_inf: 549.1928 | lim_sup: 7759.6426
Límites IQR -> lim_inf: 48.3100 | lim_sup: 48.3100
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -60.0000 | lim_sup: 100.0000
Límites IQR -> lim_inf: -411.3150 | lim_sup: 975.5250
Límites IQR -> lim_inf: -91.5000 | lim_sup: 152.5000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -59.9400 | lim_sup: 2004.4800
Límites IQR -> lim_inf: 69.2800 | lim_sup: 69.2800
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -231.0225 | lim_sup: 385.0375
Límites IQR -> lim_inf: -456.0000 | lim_sup: 928.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 70.0150 | lim_sup: 2726.0550
Límites IQR -> lim_inf: -538.9209 | lim_sup: 898.2015
Límites IQR -> lim_inf: -319.7062 | lim_sup: 1604.1914
Límites IQR -> lim_inf: -268.9053 | lim_sup: 448.1755
Límites IQR -> lim_inf: 684.6996 | lim_sup: 10722.3682
Límites IQR -> lim_inf: 58.4300 | lim_sup: 58.4300
Límites IQR -> lim_inf: 290.0000 | lim_sup: 290.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -562.9050 | lim_sup: 938.1750
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 110.0000 | lim_sup: 110.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -395.2387 | lim_sup: 2811.2712
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -154.3200 | lim_sup: 257.2000
Límites IQR -> lim_inf: -360.0000 | lim_sup: 600.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -182.1450 | lim_sup: 303.5750
Límites IQR -> lim_inf: -324.1238 | lim_sup: 2328.0663
Límites IQR -> lim_inf: 220.3700 | lim_sup: 220.3700
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -1633.5000 | lim_sup: 2722.5000
Límites IQR -> lim_inf: -849.2250 | lim_sup: 4244.9950
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 1079.6975 | lim_sup: 1887.5575
Límites IQR -> lim_inf: -2812.2000 | lim_sup: 4687.0000
Límites IQR -> lim_inf: 20.2162 | lim_sup: 8347.7863
Límites IQR -> lim_inf: -43.2202 | lim_sup: 302.5414
Límites IQR -> lim_inf: -412.8996 | lim_sup: 1863.4498
Límites IQR -> lim_inf: 21.5532 | lim_sup: 21.5532
Límites IQR -> lim_inf: -225.6840 | lim_sup: 376.1400
Límites IQR -> lim_inf: 1274.7601 | lim_sup: 8867.3855
Límites IQR -> lim_inf: -7.2237 | lim_sup: 97.8462
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 8.5325 | lim_sup: 49.6125
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -171.7977 | lim_sup: 999.2122
Límites IQR -> lim_inf: -871.0388 | lim_sup: 4058.7227
Límites IQR -> lim_inf: 2539.6867 | lim_sup: 10814.6768
Límites IQR -> lim_inf: -506.2500 | lim_sup: 843.7500
Límites IQR -> lim_inf: 10973.3791 | lim_sup: 23192.5846
Límites IQR -> lim_inf: 128.7807 | lim_sup: 467.8625
Límites IQR -> lim_inf: -271.9318 | lim_sup: 961.7045
Límites IQR -> lim_inf: -1887.0784 | lim_sup: 5287.7125
Límites IQR -> lim_inf: 4800.0739 | lim_sup: 5403.3375
Límites IQR -> lim_inf: -600.0000 | lim_sup: 1000.0000
Límites IQR -> lim_inf: 10248.5932 | lim_sup: 12804.7750
Límites IQR -> lim_inf: 53.2100 | lim_sup: 53.2100
Límites IQR -> lim_inf: 530.0000 | lim_sup: 530.0000
Límites IQR -> lim_inf: -60.0000 | lim_sup: 100.0000
Límites IQR -> lim_inf: -670.8387 | lim_sup: 1488.5113
Límites IQR -> lim_inf: -91.5000 | lim_sup: 152.5000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -177.0025 | lim_sup: 2273.3375
Límites IQR -> lim_inf: 86.0955 | lim_sup: 453.6227
Límites IQR -> lim_inf: 181.8182 | lim_sup: 181.8182
Límites IQR -> lim_inf: -374.2523 | lim_sup: 4748.6023
Límites IQR -> lim_inf: -7448.7107 | lim_sup: 17031.9420
Límites IQR -> lim_inf: -234.5455 | lim_sup: 390.9091
Límites IQR -> lim_inf: 4884.2307 | lim_sup: 7058.4761
Límites IQR -> lim_inf: 28.9625 | lim_sup: 392.3352
Límites IQR -> lim_inf: 49.0909 | lim_sup: 100.0000
Límites IQR -> lim_inf: -2042.7284 | lim_sup: 6577.0080
Límites IQR -> lim_inf: 3060.7409 | lim_sup: 7911.5409
Límites IQR -> lim_inf: -329.2364 | lim_sup: 548.7273
Límites IQR -> lim_inf: 8498.6830 | lim_sup: 10799.4375
Límites IQR -> lim_inf: 280.0000 | lim_sup: 280.0000
Límites IQR -> lim_inf: -60.0000 | lim_sup: 100.0000
Límites IQR -> lim_inf: -335.9100 | lim_sup: 559.8500
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 195.8675 | lim_sup: 2160.2875
Límites IQR -> lim_inf: 45.4800 | lim_sup: 45.4800
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -286.5000 | lim_sup: 477.5000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -257.4212 | lim_sup: 2132.1888
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 1922.0888 | lim_sup: 3912.1507
Límites IQR -> lim_inf: -4210.4742 | lim_sup: 9257.6313
Límites IQR -> lim_inf: 1879.3740 | lim_sup: 12389.9300
Límites IQR -> lim_inf: -309.3750 | lim_sup: 515.6250
Límites IQR -> lim_inf: 7752.1877 | lim_sup: 19190.7218
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -93.7500 | lim_sup: 156.2500
Límites IQR -> lim_inf: -756.2699 | lim_sup: 2374.6658
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -4574.4750 | lim_sup: 7624.1250
Límites IQR -> lim_inf: -244.4518 | lim_sup: 407.4196
Límites IQR -> lim_inf: 571.4235 | lim_sup: 1240.3640
Límites IQR -> lim_inf: 195.1864 | lim_sup: 357.1864
Límites IQR -> lim_inf: -209.0909 | lim_sup: 634.5455
Límites IQR -> lim_inf: -2038.7875 | lim_sup: 4464.1670
Límites IQR -> lim_inf: 2303.7909 | lim_sup: 6463.7182
Límites IQR -> lim_inf: -535.8068 | lim_sup: 893.0114
Límites IQR -> lim_inf: 7956.1943 | lim_sup: 10095.0761
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 293.8500 | lim_sup: 293.8500
Límites IQR -> lim_inf: 868.6400 | lim_sup: 868.6400
Límites IQR -> lim_inf: -74.9350 | lim_sup: 343.6650
Límites IQR -> lim_inf: 40.4500 | lim_sup: 40.4500
Límites IQR -> lim_inf: 280.0000 | lim_sup: 280.0000
Límites IQR -> lim_inf: -120.0000 | lim_sup: 200.0000
Límites IQR -> lim_inf: -468.4613 | lim_sup: 780.7687
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -106.0413 | lim_sup: 1998.0688
Límites IQR -> lim_inf: -608.8138 | lim_sup: 2654.7762
Límites IQR -> lim_inf: -668.2600 | lim_sup: 1451.3400
Límites IQR -> lim_inf: -463.0600 | lim_sup: 3456.9400
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -238.3825 | lim_sup: 2650.6375
Límites IQR -> lim_inf: -512.1900 | lim_sup: 2390.5100
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -82.5000 | lim_sup: 137.5000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -1753.3437 | lim_sup: 10232.9263
Límites IQR -> lim_inf: -84.7650 | lim_sup: 187.9350
Límites IQR -> lim_inf: 39.6750 | lim_sup: 864.4350
Límites IQR -> lim_inf: 76.5550 | lim_sup: 883.3950
Límites IQR -> lim_inf: -128.0150 | lim_sup: 1718.0250
Límites IQR -> lim_inf: -148.6950 | lim_sup: 2065.2250
Límites IQR -> lim_inf: -295.6438 | lim_sup: 1840.8863
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -1271.1077 | lim_sup: 3598.6923
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -28.8571 | lim_sup: 175.7714
Límites IQR -> lim_inf: -118.5176 | lim_sup: 233.0848
Límites IQR -> lim_inf: 2968.1815 | lim_sup: 11489.9435
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -615.0057 | lim_sup: 2166.6491
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -148.5571 | lim_sup: 247.5952
Límites IQR -> lim_inf: -39.9592 | lim_sup: 287.2479
Límites IQR -> lim_inf: 2946.2187 | lim_sup: 9741.5307
Límites IQR -> lim_inf: 1071.4286 | lim_sup: 1071.4286
Límites IQR -> lim_inf: -892.3054 | lim_sup: 1487.1756
Límites IQR -> lim_inf: -345.5568 | lim_sup: 3074.4503
Límites IQR -> lim_inf: -125.8098 | lim_sup: 337.3592
Límites IQR -> lim_inf: -185.4646 | lim_sup: 506.8854
Límites IQR -> lim_inf: 4002.2655 | lim_sup: 7640.9893
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -478.6905 | lim_sup: 1068.1667
Límites IQR -> lim_inf: -1091.0092 | lim_sup: 5145.1122
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -535.4500 | lim_sup: 1020.0929
Límites IQR -> lim_inf: -291.6423 | lim_sup: 755.4958
Límites IQR -> lim_inf: 3908.0664 | lim_sup: 9492.7021
Límites IQR -> lim_inf: -80.3571 | lim_sup: 133.9286
Límites IQR -> lim_inf: -725.1530 | lim_sup: 1841.6661
Límites IQR -> lim_inf: -122.0286 | lim_sup: 203.3810
Límites IQR -> lim_inf: -94.4408 | lim_sup: 258.2330
Límites IQR -> lim_inf: 1223.9348 | lim_sup: 5704.1610
Límites IQR -> lim_inf: -114.7955 | lim_sup: 191.3259
Límites IQR -> lim_inf: -527.4030 | lim_sup: 1498.5875
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -177.7214 | lim_sup: 296.2024
Límites IQR -> lim_inf: -155.8143 | lim_sup: 259.6905
Límites IQR -> lim_inf: 1320.5827 | lim_sup: 6331.1970
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -170.8545 | lim_sup: 284.7574
Límites IQR -> lim_inf: -463.8586 | lim_sup: 1158.3152
Límites IQR -> lim_inf: -119.0536 | lim_sup: 198.4226
Límites IQR -> lim_inf: -111.9643 | lim_sup: 186.6071
Límites IQR -> lim_inf: 2182.2440 | lim_sup: 4548.0536
Límites IQR -> lim_inf: -355.3571 | lim_sup: 592.2619
Límites IQR -> lim_inf: -667.8232 | lim_sup: 2094.0196
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -148.5536 | lim_sup: 247.5893
Límites IQR -> lim_inf: -166.6155 | lim_sup: 356.8036
Límites IQR -> lim_inf: 2001.5976 | lim_sup: 6575.6214
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -526.0232 | lim_sup: 876.7054
Límites IQR -> lim_inf: -97.3589 | lim_sup: 162.2649
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 1438.8202 | lim_sup: 5023.5155
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -182.7950 | lim_sup: 1990.2650
Límites IQR -> lim_inf: -198.9663 | lim_sup: 1805.8238
Límites IQR -> lim_inf: -577.0950 | lim_sup: 2533.6850
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -631.4100 | lim_sup: 1052.3500
Límites IQR -> lim_inf: -77.2500 | lim_sup: 128.7500
Límites IQR -> lim_inf: -182.1450 | lim_sup: 303.5750
Límites IQR -> lim_inf: -183.4538 | lim_sup: 1839.9362
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -104.0625 | lim_sup: 173.4375
Límites IQR -> lim_inf: -118.6875 | lim_sup: 197.8125
Límites IQR -> lim_inf: 121.4300 | lim_sup: 121.4300
Límites IQR -> lim_inf: -52.4050 | lim_sup: 1984.8350
Límites IQR -> lim_inf: -186.9887 | lim_sup: 1852.3412
Límites IQR -> lim_inf: -371.5188 | lim_sup: 4299.1313
Límites IQR -> lim_inf: -628.4450 | lim_sup: 2650.2150
Límites IQR -> lim_inf: -794.7400 | lim_sup: 1765.5000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -336.0000 | lim_sup: 560.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -76.5000 | lim_sup: 127.5000
Límites IQR -> lim_inf: -189.4462 | lim_sup: 1825.2837
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -171.4125 | lim_sup: 285.6875
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 121.4300 | lim_sup: 121.4300
Límites IQR -> lim_inf: -142.4000 | lim_sup: 2213.8200
Límites IQR -> lim_inf: -342.8487 | lim_sup: 2803.5612
Límites IQR -> lim_inf: -367.2000 | lim_sup: 612.0000
Límites IQR -> lim_inf: -179.1000 | lim_sup: 298.5000
Límites IQR -> lim_inf: -181.8375 | lim_sup: 476.7025
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -38.6250 | lim_sup: 64.3750
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -607.8688 | lim_sup: 3361.0013
Límites IQR -> lim_inf: 79.2300 | lim_sup: 79.2300
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -137.5125 | lim_sup: 229.1875
Límites IQR -> lim_inf: -291.9000 | lim_sup: 486.5000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 90.6875 | lim_sup: 142.1875
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -727.3688 | lim_sup: 3999.8613
Límites IQR -> lim_inf: -73.4888 | lim_sup: 321.3613
Límites IQR -> lim_inf: -109.7288 | lim_sup: 2198.9213
Límites IQR -> lim_inf: -869.3025 | lim_sup: 3458.1375
Límites IQR -> lim_inf: -65.6463 | lim_sup: 534.8038
Límites IQR -> lim_inf: -29.9700 | lim_sup: 2914.6700
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -570.0000 | lim_sup: 950.0000
Límites IQR -> lim_inf: 135.3575 | lim_sup: 746.5775
Límites IQR -> lim_inf: -246.8363 | lim_sup: 411.3938
Límites IQR -> lim_inf: -126.6112 | lim_sup: 687.7387
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -208.5000 | lim_sup: 347.5000
Límites IQR -> lim_inf: -1062.9250 | lim_sup: 1835.6750
Límites IQR -> lim_inf: 204.3950 | lim_sup: 1323.8750
Límites IQR -> lim_inf: -362.2500 | lim_sup: 603.7500
Límites IQR -> lim_inf: -414.7387 | lim_sup: 1655.1912
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -268.8025 | lim_sup: 656.2175
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -404.2125 | lim_sup: 673.6875
Límites IQR -> lim_inf: -102.9375 | lim_sup: 171.5625
Límites IQR -> lim_inf: -39.6463 | lim_sup: 1068.8438
Límites IQR -> lim_inf: 80.0000 | lim_sup: 80.0000
Límites IQR -> lim_inf: -204.4050 | lim_sup: 340.6750
Límites IQR -> lim_inf: -632.6400 | lim_sup: 1054.4000
Límites IQR -> lim_inf: 739.1000 | lim_sup: 739.1000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -237.7650 | lim_sup: 1625.8350
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -123.4425 | lim_sup: 205.7375
Límites IQR -> lim_inf: -469.3388 | lim_sup: 782.2313
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 121.4300 | lim_sup: 121.4300
Límites IQR -> lim_inf: -466.3088 | lim_sup: 2252.4812
Límites IQR -> lim_inf: 144.1864 | lim_sup: 442.1864
Límites IQR -> lim_inf: -373.6364 | lim_sup: 804.5455
Límites IQR -> lim_inf: -2082.1989 | lim_sup: 5311.3193
Límites IQR -> lim_inf: 5515.3636 | lim_sup: 5609.9091
Límites IQR -> lim_inf: -340.9091 | lim_sup: 568.1818
Límites IQR -> lim_inf: 14214.4625 | lim_sup: 18688.8352
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -171.0637 | lim_sup: 285.1062
Límites IQR -> lim_inf: -87.7425 | lim_sup: 146.2375
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -139.2250 | lim_sup: 1292.5550
Límites IQR -> lim_inf: 95.1600 | lim_sup: 95.1600
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -148.5637 | lim_sup: 247.6062
Límites IQR -> lim_inf: -325.9162 | lim_sup: 543.1938
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -258.6488 | lim_sup: 461.9612
Límites IQR -> lim_inf: -75.3000 | lim_sup: 1243.7400
Límites IQR -> lim_inf: 65.9350 | lim_sup: 889.8950
Límites IQR -> lim_inf: -51.6700 | lim_sup: 1368.1700
Límites IQR -> lim_inf: 142.0850 | lim_sup: 1413.9650
Límites IQR -> lim_inf: -26.3500 | lim_sup: 1652.7700
Límites IQR -> lim_inf: 7264.5945 | lim_sup: 7264.5945
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 1094.7757 | lim_sup: 4317.5779
Límites IQR -> lim_inf: -8.9432 | lim_sup: 7699.3567
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 5423.2310 | lim_sup: 20785.4750
Límites IQR -> lim_inf: 968.2943 | lim_sup: 3716.7648
Límites IQR -> lim_inf: 11353.0716 | lim_sup: 35570.5706
Límites IQR -> lim_inf: -501.1950 | lim_sup: 1117.7250
Límites IQR -> lim_inf: 412.6750 | lim_sup: 1117.6750
Límites IQR -> lim_inf: 1120.8275 | lim_sup: 1778.8075
Límites IQR -> lim_inf: 85.4125 | lim_sup: 221.0325
Límites IQR -> lim_inf: -669.5300 | lim_sup: 1739.3500
Límites IQR -> lim_inf: -470.3225 | lim_sup: 1600.3975
Límites IQR -> lim_inf: -521.6475 | lim_sup: 2370.5525
Límites IQR -> lim_inf: -579.5350 | lim_sup: 1074.5650
Límites IQR -> lim_inf: -1072.9650 | lim_sup: 3292.1350
Límites IQR -> lim_inf: -238.7400 | lim_sup: 1524.7000
Límites IQR -> lim_inf: -359.8488 | lim_sup: 2071.6213
Límites IQR -> lim_inf: -144.2308 | lim_sup: 240.3846
Límites IQR -> lim_inf: -334.5577 | lim_sup: 557.5962
Límites IQR -> lim_inf: -973.0104 | lim_sup: 2082.3847
Límites IQR -> lim_inf: 8.4433 | lim_sup: 408.5388
Límites IQR -> lim_inf: 279.6693 | lim_sup: 2214.6784
Límites IQR -> lim_inf: -437.9550 | lim_sup: 898.0850
Límites IQR -> lim_inf: 87.0925 | lim_sup: 2016.5925
Límites IQR -> lim_inf: -9.1500 | lim_sup: 1905.4100
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -175.6912 | lim_sup: 292.8188
Límites IQR -> lim_inf: -251.3950 | lim_sup: 564.4850
Límites IQR -> lim_inf: -77.2500 | lim_sup: 128.7500
Límites IQR -> lim_inf: -209.3587 | lim_sup: 348.9312
Límites IQR -> lim_inf: 42.4250 | lim_sup: 2253.9050
Límites IQR -> lim_inf: -876.9875 | lim_sup: 3034.5525
Límites IQR -> lim_inf: -128.6014 | lim_sup: 460.4895
Límites IQR -> lim_inf: -3376.6744 | lim_sup: 14092.5618
Límites IQR -> lim_inf: 238.4615 | lim_sup: 546.1538
Límites IQR -> lim_inf: -138.4615 | lim_sup: 230.7692
Límites IQR -> lim_inf: 20142.5679 | lim_sup: 40164.9422
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -216.0187 | lim_sup: 360.0312
Límites IQR -> lim_inf: -461.2500 | lim_sup: 768.7500
Límites IQR -> lim_inf: -116.7300 | lim_sup: 194.5500
Límites IQR -> lim_inf: -25.5000 | lim_sup: 42.5000
Límites IQR -> lim_inf: -62.4938 | lim_sup: 1401.5363
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -82.5000 | lim_sup: 137.5000
Límites IQR -> lim_inf: -260.6250 | lim_sup: 434.3750
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 121.4300 | lim_sup: 121.4300
Límites IQR -> lim_inf: -433.6237 | lim_sup: 2189.8262
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -329.6925 | lim_sup: 549.4875
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 61.2600 | lim_sup: 61.2600
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -242.8200 | lim_sup: 404.7000
Límites IQR -> lim_inf: -759.3750 | lim_sup: 1265.6250
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -192.0000 | lim_sup: 320.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -138.9900 | lim_sup: 2677.1300
Límites IQR -> lim_inf: -221.8650 | lim_sup: 369.7750
Límites IQR -> lim_inf: -265.2300 | lim_sup: 442.0500
Límites IQR -> lim_inf: 250.0000 | lim_sup: 250.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 95.7200 | lim_sup: 95.7200
Límites IQR -> lim_inf: -112.5000 | lim_sup: 187.5000
Límites IQR -> lim_inf: -263.3100 | lim_sup: 438.8500
Límites IQR -> lim_inf: -386.7225 | lim_sup: 644.5375
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -38.6250 | lim_sup: 64.3750
Límites IQR -> lim_inf: -75.0000 | lim_sup: 125.0000
Límites IQR -> lim_inf: -342.2775 | lim_sup: 4116.2425
Límites IQR -> lim_inf: 47.9000 | lim_sup: 47.9000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -203.6025 | lim_sup: 339.3375
Límites IQR -> lim_inf: -91.0875 | lim_sup: 151.8125
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -19.3125 | lim_sup: 32.1875
Límites IQR -> lim_inf: 203.0000 | lim_sup: 203.0000
Límites IQR -> lim_inf: -166.1962 | lim_sup: 2169.1737
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -177.4350 | lim_sup: 295.7250
Límites IQR -> lim_inf: -246.7575 | lim_sup: 411.2625
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -475.6612 | lim_sup: 792.7687
Límites IQR -> lim_inf: -148.5262 | lim_sup: 1443.8437
Límites IQR -> lim_inf: -523.5000 | lim_sup: 872.5000
Límites IQR -> lim_inf: -163.7175 | lim_sup: 272.8625
Límites IQR -> lim_inf: -499.2525 | lim_sup: 832.0875
Límites IQR -> lim_inf: -38.6250 | lim_sup: 64.3750
Límites IQR -> lim_inf: -75.0000 | lim_sup: 125.0000
Límites IQR -> lim_inf: -407.6850 | lim_sup: 2344.7150
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -304.5000 | lim_sup: 507.5000
Límites IQR -> lim_inf: -268.7400 | lim_sup: 447.9000
Límites IQR -> lim_inf: 44.6250 | lim_sup: 49.2250
Límites IQR -> lim_inf: 121.4300 | lim_sup: 121.4300
Límites IQR -> lim_inf: -392.8563 | lim_sup: 3381.7938
Límites IQR -> lim_inf: 96.1800 | lim_sup: 96.1800
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -155.5200 | lim_sup: 259.2000
Límites IQR -> lim_inf: -728.3963 | lim_sup: 1213.9938
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -674.6237 | lim_sup: 3877.3863
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -198.2700 | lim_sup: 330.4500
Límites IQR -> lim_inf: -243.8288 | lim_sup: 406.3813
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -38.6250 | lim_sup: 64.3750
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -279.2800 | lim_sup: 2125.1800
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -279.9075 | lim_sup: 466.5125
Límites IQR -> lim_inf: -209.9738 | lim_sup: 349.9563
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 52.9500 | lim_sup: 52.9500
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -141.1500 | lim_sup: 235.2500
Límites IQR -> lim_inf: -104.2612 | lim_sup: 173.7687
Límites IQR -> lim_inf: 394.4750 | lim_sup: 416.5750
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -165.7338 | lim_sup: 2307.7963
Límites IQR -> lim_inf: 342.6400 | lim_sup: 342.6400
Límites IQR -> lim_inf: -217.0500 | lim_sup: 361.7500
Límites IQR -> lim_inf: -29.6400 | lim_sup: 49.4000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -69.0563 | lim_sup: 115.0938
Límites IQR -> lim_inf: -954.3600 | lim_sup: 1590.6000
Límites IQR -> lim_inf: -232.0075 | lim_sup: 898.1725
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 1897.6050 | lim_sup: 2094.7650
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -540.1750 | lim_sup: 3358.7850
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -36.1125 | lim_sup: 60.1875
Límites IQR -> lim_inf: -479.4000 | lim_sup: 799.0000
Límites IQR -> lim_inf: -1012.6237 | lim_sup: 1687.7062
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 463.2200 | lim_sup: 773.5400
Límites IQR -> lim_inf: -75.0000 | lim_sup: 125.0000
Límites IQR -> lim_inf: -64.8400 | lim_sup: 5235.5200
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -180.0000 | lim_sup: 300.0000
Límites IQR -> lim_inf: -153.7500 | lim_sup: 256.2500
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -50.9250 | lim_sup: 84.8750
Límites IQR -> lim_inf: -150.1212 | lim_sup: 1783.9887
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -231.4575 | lim_sup: 385.7625
Límites IQR -> lim_inf: -404.8763 | lim_sup: 674.7938
Límites IQR -> lim_inf: -91.5000 | lim_sup: 152.5000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 28.0375 | lim_sup: 1415.2175
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -202.5000 | lim_sup: 337.5000
Límites IQR -> lim_inf: -436.5750 | lim_sup: 727.6250
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -146.7000 | lim_sup: 244.5000
Límites IQR -> lim_inf: -223.8500 | lim_sup: 1822.3100
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -207.7687 | lim_sup: 346.2812
Límites IQR -> lim_inf: -464.6025 | lim_sup: 774.3375
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -116.7300 | lim_sup: 194.5500
Límites IQR -> lim_inf: -12.7313 | lim_sup: 21.2188
Límites IQR -> lim_inf: -88.9125 | lim_sup: 1733.1475
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -202.7738 | lim_sup: 337.9563
Límites IQR -> lim_inf: -348.1688 | lim_sup: 580.2812
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -152.5750 | lim_sup: 1647.1850
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -207.7687 | lim_sup: 346.2812
Límites IQR -> lim_inf: -936.4275 | lim_sup: 1560.7125
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -91.5000 | lim_sup: 152.5000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 77.6775 | lim_sup: 999.5175
Límites IQR -> lim_inf: 46.0500 | lim_sup: 46.0500
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -333.4500 | lim_sup: 555.7500
Límites IQR -> lim_inf: -389.0150 | lim_sup: 702.8650
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -96.0000 | lim_sup: 160.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -328.3100 | lim_sup: 2264.9100
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -153.7500 | lim_sup: 256.2500
Límites IQR -> lim_inf: -538.5000 | lim_sup: 897.5000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -223.2975 | lim_sup: 1462.6425
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -857.1675 | lim_sup: 1497.2125
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -154.5000 | lim_sup: 257.5000
Límites IQR -> lim_inf: -102.0000 | lim_sup: 170.0000
Límites IQR -> lim_inf: 137.6600 | lim_sup: 1459.3400
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -286.5000 | lim_sup: 477.5000
Límites IQR -> lim_inf: -402.2663 | lim_sup: 670.4438
Límites IQR -> lim_inf: -91.5000 | lim_sup: 152.5000
Límites IQR -> lim_inf: -63.6938 | lim_sup: 106.1563
Límites IQR -> lim_inf: 0.5300 | lim_sup: 1222.1100
Límites IQR -> lim_inf: 53.7300 | lim_sup: 53.7300
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -208.3988 | lim_sup: 347.3313
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -50.9250 | lim_sup: 84.8750
Límites IQR -> lim_inf: -119.6775 | lim_sup: 2073.3225
Límites IQR -> lim_inf: 78.2900 | lim_sup: 78.2900
Límites IQR -> lim_inf: -37.5000 | lim_sup: 62.5000
Límites IQR -> lim_inf: -192.0938 | lim_sup: 320.1562
Límites IQR -> lim_inf: -198.7500 | lim_sup: 331.2500
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -969.0687 | lim_sup: 3997.3412
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -269.3850 | lim_sup: 448.9750
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -316.8962 | lim_sup: 2043.4537
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -161.2500 | lim_sup: 268.7500
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -223.1362 | lim_sup: 371.8937
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -182.1450 | lim_sup: 303.5750
Límites IQR -> lim_inf: 32.1113 | lim_sup: 1386.5013
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -153.7500 | lim_sup: 256.2500
Límites IQR -> lim_inf: -191.6250 | lim_sup: 319.3750
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -102.0000 | lim_sup: 170.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -287.6025 | lim_sup: 1878.5775
Límites IQR -> lim_inf: -203.2500 | lim_sup: 338.7500
Límites IQR -> lim_inf: -219.9375 | lim_sup: 366.5625
Límites IQR -> lim_inf: -311.3700 | lim_sup: 518.9500
Límites IQR -> lim_inf: -91.5000 | lim_sup: 152.5000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -169.2275 | lim_sup: 1479.2725
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -198.7500 | lim_sup: 331.2500
Límites IQR -> lim_inf: -121.1400 | lim_sup: 201.9000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -306.8425 | lim_sup: 1864.9375
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -217.6687 | lim_sup: 362.7812
Límites IQR -> lim_inf: -271.0013 | lim_sup: 451.6688
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -252.3025 | lim_sup: 2394.8975
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -195.0000 | lim_sup: 325.0000
Límites IQR -> lim_inf: -424.5000 | lim_sup: 707.5000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -79.9125 | lim_sup: 1730.3275
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -209.6737 | lim_sup: 349.4562
Límites IQR -> lim_inf: -187.3237 | lim_sup: 312.2062
Límites IQR -> lim_inf: -116.7300 | lim_sup: 194.5500
Límites IQR -> lim_inf: -51.0000 | lim_sup: 85.0000
Límites IQR -> lim_inf: -145.6550 | lim_sup: 1760.1650
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -186.7500 | lim_sup: 311.2500
Límites IQR -> lim_inf: -329.9513 | lim_sup: 549.9188
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -116.7300 | lim_sup: 194.5500
Límites IQR -> lim_inf: -25.5000 | lim_sup: 42.5000
Límites IQR -> lim_inf: -98.8212 | lim_sup: 1352.0887
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -112.5000 | lim_sup: 187.5000
Límites IQR -> lim_inf: -347.5163 | lim_sup: 579.1938
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 40.6988 | lim_sup: 1252.6287
Límites IQR -> lim_inf: 55.5000 | lim_sup: 55.5000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -231.0450 | lim_sup: 385.0750
Límites IQR -> lim_inf: -572.7750 | lim_sup: 954.6250
Límites IQR -> lim_inf: -116.7300 | lim_sup: 194.5500
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -182.1212 | lim_sup: 2320.8687
Límites IQR -> lim_inf: 530.0000 | lim_sup: 530.0000
Límites IQR -> lim_inf: -181.3500 | lim_sup: 302.2500
Límites IQR -> lim_inf: -668.1750 | lim_sup: 1113.6250
Límites IQR -> lim_inf: -91.5000 | lim_sup: 152.5000
Límites IQR -> lim_inf: -118.6500 | lim_sup: 197.7500
Límites IQR -> lim_inf: -279.9125 | lim_sup: 2171.0875
Límites IQR -> lim_inf: 97.3900 | lim_sup: 97.3900
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -732.3900 | lim_sup: 1220.6500
Límites IQR -> lim_inf: -511.5225 | lim_sup: 852.5375
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 1056.2700 | lim_sup: 1815.6300
Límites IQR -> lim_inf: -163.9275 | lim_sup: 273.2125
Límites IQR -> lim_inf: -819.6162 | lim_sup: 4777.9737
Límites IQR -> lim_inf: 133.3100 | lim_sup: 133.3100
Límites IQR -> lim_inf: -92.3625 | lim_sup: 153.9375
Límites IQR -> lim_inf: -891.1163 | lim_sup: 1485.1937
Límites IQR -> lim_inf: -632.1000 | lim_sup: 1117.1000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 1336.4450 | lim_sup: 2297.2050
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -955.3075 | lim_sup: 6041.9325
Límites IQR -> lim_inf: -286.2488 | lim_sup: 477.0813
Límites IQR -> lim_inf: -1293.1500 | lim_sup: 2155.2500
Límites IQR -> lim_inf: -1203.9050 | lim_sup: 2548.8750
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 3465.9088 | lim_sup: 4290.8988
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -1263.3350 | lim_sup: 8156.1850
Límites IQR -> lim_inf: 233.4000 | lim_sup: 233.4000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -1143.9225 | lim_sup: 1906.5375
Límites IQR -> lim_inf: -634.4662 | lim_sup: 1569.5437
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 2941.1763 | lim_sup: 3895.9862
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -1044.2863 | lim_sup: 8980.4637
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -680.4000 | lim_sup: 1134.0000
Límites IQR -> lim_inf: -494.4100 | lim_sup: 871.5300
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -537.9262 | lim_sup: 896.5438
Límites IQR -> lim_inf: 156.0525 | lim_sup: 280.3525
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -138.4000 | lim_sup: 3386.4000
Límites IQR -> lim_inf: -1351.6238 | lim_sup: 4551.9662
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -60.9000 | lim_sup: 686.5400
Límites IQR -> lim_inf: -827.4375 | lim_sup: 1379.0625
Límites IQR -> lim_inf: -555.8713 | lim_sup: 39980.6588
Límites IQR -> lim_inf: -47.5500 | lim_sup: 1402.2300
Límites IQR -> lim_inf: 264.0450 | lim_sup: 1175.2050
Límites IQR -> lim_inf: -407.4125 | lim_sup: 1044.8275
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -30.0000 | lim_sup: 50.0000
Límites IQR -> lim_inf: -409.0350 | lim_sup: 681.7250
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -847.2463 | lim_sup: 3671.3438
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -314.3550 | lim_sup: 523.9250
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 121.4300 | lim_sup: 121.4300
Límites IQR -> lim_inf: -279.7962 | lim_sup: 1706.0938
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -120.0000 | lim_sup: 200.0000
Límites IQR -> lim_inf: -311.2500 | lim_sup: 518.7500
Límites IQR -> lim_inf: 121.4300 | lim_sup: 121.4300
Límites IQR -> lim_inf: -106.1262 | lim_sup: 1578.0237
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -120.0000 | lim_sup: 200.0000
Límites IQR -> lim_inf: -394.6575 | lim_sup: 953.4025
Límites IQR -> lim_inf: 103.6250 | lim_sup: 142.6250
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -59.3943 | lim_sup: 261.3966
Límites IQR -> lim_inf: 31.8182 | lim_sup: 68.1818
Límites IQR -> lim_inf: -1317.5352 | lim_sup: 3559.8557
Límites IQR -> lim_inf: 4731.7000 | lim_sup: 4942.6091
Límites IQR -> lim_inf: -317.5568 | lim_sup: 529.2614
Límites IQR -> lim_inf: 6306.0432 | lim_sup: 11320.8614
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -470.9863 | lim_sup: 1195.6438
Límites IQR -> lim_inf: -1103.0075 | lim_sup: 1932.1725
Límites IQR -> lim_inf: -1692.9825 | lim_sup: 3049.6375
Límites IQR -> lim_inf: 1549.8525 | lim_sup: 3937.0725
Límites IQR -> lim_inf: -33.7500 | lim_sup: 56.2500
Límites IQR -> lim_inf: -1422.3050 | lim_sup: 5857.5150
Límites IQR -> lim_inf: 80.0000 | lim_sup: 80.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -29.3063 | lim_sup: 48.8438
Límites IQR -> lim_inf: -376.9425 | lim_sup: 628.2375
Límites IQR -> lim_inf: -870.0300 | lim_sup: 1450.0500
Límites IQR -> lim_inf: 527.9950 | lim_sup: 1071.0750
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -202.1563 | lim_sup: 1633.6738
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 373.6150 | lim_sup: 662.4230
Límites IQR -> lim_inf: 55.9464 | lim_sup: 2575.9044
Límites IQR -> lim_inf: 1778.2577 | lim_sup: 4351.7958
Límites IQR -> lim_inf: -1287.9386 | lim_sup: 2730.8514
Límites IQR -> lim_inf: 4137.6129 | lim_sup: 11539.6599
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 163.4475 | lim_sup: 478.8475
Límites IQR -> lim_inf: -442.8429 | lim_sup: 2561.8646
Límites IQR -> lim_inf: 1560.5034 | lim_sup: 9097.6004
Límites IQR -> lim_inf: -330.4948 | lim_sup: 1491.8807
Límites IQR -> lim_inf: 4610.7330 | lim_sup: 13665.9050
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 216.1610 | lim_sup: 807.9570
Límites IQR -> lim_inf: -891.1878 | lim_sup: 2625.6462
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 2587.4108 | lim_sup: 5519.3153
Límites IQR -> lim_inf: -365.7247 | lim_sup: 1750.4628
Límites IQR -> lim_inf: 7524.9766 | lim_sup: 26941.7836
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 191.5165 | lim_sup: 191.5165
Límites IQR -> lim_inf: -1626.8031 | lim_sup: 4392.7384
Límites IQR -> lim_inf: -574.0577 | lim_sup: 9054.9243
Límites IQR -> lim_inf: -476.9712 | lim_sup: 1225.2037
Límites IQR -> lim_inf: 6154.9976 | lim_sup: 12488.9406
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 626.6308 | lim_sup: 759.3847
Límites IQR -> lim_inf: -1684.5631 | lim_sup: 6869.7239
Límites IQR -> lim_inf: 4871.3888 | lim_sup: 9239.3867
Límites IQR -> lim_inf: -3742.1814 | lim_sup: 8670.7161
Límites IQR -> lim_inf: 3740.8239 | lim_sup: 31939.5059
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 732.7387 | lim_sup: 746.0408
Límites IQR -> lim_inf: -1458.3326 | lim_sup: 3721.6544
Límites IQR -> lim_inf: 2009.0582 | lim_sup: 9160.6083
Límites IQR -> lim_inf: -877.6787 | lim_sup: 2200.0978
Límites IQR -> lim_inf: 562.1781 | lim_sup: 30958.1466
Límites IQR -> lim_inf: 3541.9845 | lim_sup: 3541.9845
Límites IQR -> lim_inf: -16.6444 | lim_sup: 27.7406
Límites IQR -> lim_inf: 1264.6299 | lim_sup: 2666.8217
Límites IQR -> lim_inf: -329.3858 | lim_sup: 5485.3096
Límites IQR -> lim_inf: 3672.2908 | lim_sup: 10816.5527
Límites IQR -> lim_inf: -523.2810 | lim_sup: 3589.6010
Límites IQR -> lim_inf: 3065.1500 | lim_sup: 36986.2720
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -66.0000 | lim_sup: 110.0000
Límites IQR -> lim_inf: -231.0863 | lim_sup: 385.1438
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 68.0000 | lim_sup: 68.0000
Límites IQR -> lim_inf: -58.2925 | lim_sup: 1334.9675
Límites IQR -> lim_inf: 1130.0000 | lim_sup: 1130.0000
Límites IQR -> lim_inf: 20.3238 | lim_sup: 60.0337
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -248.8050 | lim_sup: 599.9950
Límites IQR -> lim_inf: -227.7650 | lim_sup: 349.7350
Límites IQR -> lim_inf: 80.0000 | lim_sup: 80.0000
Límites IQR -> lim_inf: -38.3400 | lim_sup: 63.9000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -278.2575 | lim_sup: 463.7625
Límites IQR -> lim_inf: -646.5225 | lim_sup: 1077.5375
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 68.0000 | lim_sup: 68.0000
Límites IQR -> lim_inf: -116.5425 | lim_sup: 1878.3175
Límites IQR -> lim_inf: 200.0000 | lim_sup: 200.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -372.7650 | lim_sup: 1168.4750
Límites IQR -> lim_inf: -915.8225 | lim_sup: 1998.7575
Límites IQR -> lim_inf: -3438.8862 | lim_sup: 9103.5437
Límites IQR -> lim_inf: 1988.6250 | lim_sup: 2746.8250
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -23.9975 | lim_sup: 2044.6425
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -120.0000 | lim_sup: 200.0000
Límites IQR -> lim_inf: -744.0775 | lim_sup: 1456.3425
Límites IQR -> lim_inf: 64.0000 | lim_sup: 64.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -335.4412 | lim_sup: 2730.5087
Límites IQR -> lim_inf: 746.3500 | lim_sup: 746.3500
Límites IQR -> lim_inf: -0.9350 | lim_sup: 49.2050
Límites IQR -> lim_inf: 276.9231 | lim_sup: 276.9231
Límites IQR -> lim_inf: -1194.7328 | lim_sup: 1991.2213
Límites IQR -> lim_inf: -122.0250 | lim_sup: 337.0350
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -570.7012 | lim_sup: 951.1687
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 337.5400 | lim_sup: 587.5400
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -72.9727 | lim_sup: 572.4818
Límites IQR -> lim_inf: 1699.8818 | lim_sup: 1699.8818
Límites IQR -> lim_inf: -935.3148 | lim_sup: 2813.4034
Límites IQR -> lim_inf: 2310.2636 | lim_sup: 6404.4455
Límites IQR -> lim_inf: -833.8091 | lim_sup: 1389.6818
Límites IQR -> lim_inf: 7055.0205 | lim_sup: 14600.0932
Límites IQR -> lim_inf: 80.9091 | lim_sup: 80.9091
Límites IQR -> lim_inf: -657.3670 | lim_sup: 1735.8784
Límites IQR -> lim_inf: -6923.0000 | lim_sup: 11992.6364
Límites IQR -> lim_inf: -6176.7727 | lim_sup: 11490.5000
Límites IQR -> lim_inf: 15134.8886 | lim_sup: 18107.8159
Límites IQR -> lim_inf: 2336.5273 | lim_sup: 2336.5273
Límites IQR -> lim_inf: 54.5455 | lim_sup: 54.5455
Límites IQR -> lim_inf: -1154.5000 | lim_sup: 2354.5182
Límites IQR -> lim_inf: 3726.9045 | lim_sup: 5692.2500
Límites IQR -> lim_inf: -262.9659 | lim_sup: 1078.4886
Límites IQR -> lim_inf: 10317.9318 | lim_sup: 14226.1864
Límites IQR -> lim_inf: -170.6250 | lim_sup: 284.3750
Límites IQR -> lim_inf: 23.4400 | lim_sup: 23.4400
Límites IQR -> lim_inf: -104.2963 | lim_sup: 462.7138
Límites IQR -> lim_inf: -79.8135 | lim_sup: 558.6945
Límites IQR -> lim_inf: -371.2627 | lim_sup: 933.6511
Límites IQR -> lim_inf: -46.7610 | lim_sup: 77.9350
Límites IQR -> lim_inf: 714.3247 | lim_sup: 9609.3202
Límites IQR -> lim_inf: -136.6691 | lim_sup: 956.6834
Límites IQR -> lim_inf: -782.4791 | lim_sup: 1984.9775
Límites IQR -> lim_inf: -137.2500 | lim_sup: 228.7500
Límites IQR -> lim_inf: 1498.2337 | lim_sup: 2790.8538
Límites IQR -> lim_inf: 74.6806 | lim_sup: 146.9456
Límites IQR -> lim_inf: -729.0134 | lim_sup: 2243.5208
Límites IQR -> lim_inf: -214.1972 | lim_sup: 435.0220
Límites IQR -> lim_inf: 2336.8250 | lim_sup: 6512.4889
Límites IQR -> lim_inf: 504.8856 | lim_sup: 504.8856
Límites IQR -> lim_inf: -709.4508 | lim_sup: 1949.0580
Límites IQR -> lim_inf: -47.9850 | lim_sup: 79.9750
Límites IQR -> lim_inf: 1019.7511 | lim_sup: 4206.2760
Límites IQR -> lim_inf: -109.4315 | lim_sup: 766.0202
Límites IQR -> lim_inf: -969.3422 | lim_sup: 3936.2725
Límites IQR -> lim_inf: -252.0500 | lim_sup: 478.7500
Límites IQR -> lim_inf: 429.7858 | lim_sup: 17359.4473
Límites IQR -> lim_inf: -60.0000 | lim_sup: 420.0000
Límites IQR -> lim_inf: -484.4056 | lim_sup: 1166.9061
Límites IQR -> lim_inf: -7.1350 | lim_sup: 65.2250
Límites IQR -> lim_inf: 1034.5318 | lim_sup: 5363.7200
Límites IQR -> lim_inf: -27.4909 | lim_sup: 405.4545
Límites IQR -> lim_inf: -159.8045 | lim_sup: 266.3409
Límites IQR -> lim_inf: -783.1920 | lim_sup: 1827.4716
Límites IQR -> lim_inf: -468.8614 | lim_sup: 8004.0295
Límites IQR -> lim_inf: -587.6864 | lim_sup: 979.4773
Límites IQR -> lim_inf: 4837.6250 | lim_sup: 9372.3705
Límites IQR -> lim_inf: 140.4102 | lim_sup: 532.4557
Límites IQR -> lim_inf: -1697.4364 | lim_sup: 3155.7273
Límites IQR -> lim_inf: -1397.9886 | lim_sup: 4959.9023
Límites IQR -> lim_inf: 6586.5341 | lim_sup: 6751.3523
Límites IQR -> lim_inf: -415.9091 | lim_sup: 693.1818
Límites IQR -> lim_inf: 15239.2864 | lim_sup: 20850.2682
Límites IQR -> lim_inf: -63.6226 | lim_sup: 106.0377
Límites IQR -> lim_inf: 195.4615 | lim_sup: 195.4615
Límites IQR -> lim_inf: -137.5442 | lim_sup: 229.2403
Límites IQR -> lim_inf: -1.3135 | lim_sup: 543.5431
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -737.3250 | lim_sup: 1228.8750
Límites IQR -> lim_inf: 148.7350 | lim_sup: 1086.7750
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -80.4238 | lim_sup: 469.0263
Límites IQR -> lim_inf: -653.6800 | lim_sup: 3786.8000
Límites IQR -> lim_inf: -36.2800 | lim_sup: 958.2800
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 299.0200 | lim_sup: 299.0200
Límites IQR -> lim_inf: -1133.9713 | lim_sup: 2465.6188
Límites IQR -> lim_inf: 150.9000 | lim_sup: 580.1000
Límites IQR -> lim_inf: -308.2537 | lim_sup: 513.7562
Límites IQR -> lim_inf: -177.1113 | lim_sup: 7657.1188
Límites IQR -> lim_inf: 50.8838 | lim_sup: 1130.0738
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -51.3675 | lim_sup: 85.6125
Límites IQR -> lim_inf: -550.2562 | lim_sup: 917.0937
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 26.8175 | lim_sup: 104.6375
Límites IQR -> lim_inf: -57.0725 | lim_sup: 185.7875
Límites IQR -> lim_inf: 80.9091 | lim_sup: 80.9091
Límites IQR -> lim_inf: -1781.5057 | lim_sup: 3468.2670
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 4199.0000 | lim_sup: 4508.0909
Límites IQR -> lim_inf: -151.4500 | lim_sup: 1001.3864
Límites IQR -> lim_inf: 10226.9182 | lim_sup: 14955.1364
Límites IQR -> lim_inf: 242.1125 | lim_sup: 351.5725
Límites IQR -> lim_inf: 64.0900 | lim_sup: 94.2500
Límites IQR -> lim_inf: -209.0624 | lim_sup: 1767.4978
Límites IQR -> lim_inf: -58.2750 | lim_sup: 97.1250
Límites IQR -> lim_inf: 539.7303 | lim_sup: 14699.0616
Límites IQR -> lim_inf: -177.8413 | lim_sup: 633.6888
Límites IQR -> lim_inf: -45.3000 | lim_sup: 454.0600
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -66.0000 | lim_sup: 110.0000
Límites IQR -> lim_inf: -511.0425 | lim_sup: 851.7375
Límites IQR -> lim_inf: 64.0000 | lim_sup: 64.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 163.5250 | lim_sup: 2081.8050
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -345.6900 | lim_sup: 576.1500
Límites IQR -> lim_inf: -644.2650 | lim_sup: 1073.7750
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -116.7300 | lim_sup: 194.5500
Límites IQR -> lim_inf: -45.5363 | lim_sup: 75.8938
Límites IQR -> lim_inf: 269.0000 | lim_sup: 269.0000
Límites IQR -> lim_inf: -153.0675 | lim_sup: 255.1125
Límites IQR -> lim_inf: -384.9825 | lim_sup: 641.6375
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -120.0000 | lim_sup: 200.0000
Límites IQR -> lim_inf: -283.1250 | lim_sup: 471.8750
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 121.4300 | lim_sup: 121.4300
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 80.0000 | lim_sup: 80.0000
Límites IQR -> lim_inf: -817.1250 | lim_sup: 1361.8750
Límites IQR -> lim_inf: 4551.4000 | lim_sup: 4551.4000
Límites IQR -> lim_inf: 389.8700 | lim_sup: 561.8700
Límites IQR -> lim_inf: -93.7500 | lim_sup: 156.2500
Límites IQR -> lim_inf: 364.0000 | lim_sup: 364.0000
Límites IQR -> lim_inf: -90.0000 | lim_sup: 150.0000
Límites IQR -> lim_inf: -772.5375 | lim_sup: 1287.5625
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -120.0000 | lim_sup: 200.0000
Límites IQR -> lim_inf: -453.0000 | lim_sup: 755.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 284.0000 | lim_sup: 284.0000
Límites IQR -> lim_inf: -138.3750 | lim_sup: 230.6250
Límites IQR -> lim_inf: -333.7500 | lim_sup: 556.2500
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -98.2500 | lim_sup: 163.7500
Límites IQR -> lim_inf: -392.1188 | lim_sup: 653.5312
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -197.7650 | lim_sup: 2551.7950
Límites IQR -> lim_inf: 461.2455 | lim_sup: 461.2455
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -122.7273 | lim_sup: 204.5455
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -51.3675 | lim_sup: 85.6125
Límites IQR -> lim_inf: -17.5725 | lim_sup: 29.2875
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -217.5000 | lim_sup: 362.5000
Límites IQR -> lim_inf: -102.0000 | lim_sup: 170.0000
Límites IQR -> lim_inf: -386.7375 | lim_sup: 2033.0625
Límites IQR -> lim_inf: -586.4700 | lim_sup: 2083.6900
Límites IQR -> lim_inf: -89.7750 | lim_sup: 149.6250
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 53.0300 | lim_sup: 191.4300
Límites IQR -> lim_inf: -159.3750 | lim_sup: 265.6250
Límites IQR -> lim_inf: 1100.4500 | lim_sup: 1100.4500
Límites IQR -> lim_inf: -221.2500 | lim_sup: 368.7500
Límites IQR -> lim_inf: 240.0000 | lim_sup: 240.0000
Límites IQR -> lim_inf: 40.2875 | lim_sup: 425.1075
Límites IQR -> lim_inf: 162.4750 | lim_sup: 332.4150
Límites IQR -> lim_inf: 348.4550 | lim_sup: 1325.3750
Límites IQR -> lim_inf: 141.7000 | lim_sup: 1191.7800
Límites IQR -> lim_inf: 130.6100 | lim_sup: 549.6700
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 763.4030 | lim_sup: 763.4030
Límites IQR -> lim_inf: 209.6341 | lim_sup: 4191.8366
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 6782.3375 | lim_sup: 8589.3175
Límites IQR -> lim_inf: -1763.2312 | lim_sup: 9886.8187
Límites IQR -> lim_inf: 232.5183 | lim_sup: 22010.4973
Límites IQR -> lim_inf: 241.9100 | lim_sup: 241.9100
Límites IQR -> lim_inf: -1287.9850 | lim_sup: 2777.6750
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 477.5000 | lim_sup: 477.5000
Límites IQR -> lim_inf: -756.5288 | lim_sup: 1260.8813
Límites IQR -> lim_inf: -32.6125 | lim_sup: 494.8475
Límites IQR -> lim_inf: -90.4500 | lim_sup: 150.7500
Límites IQR -> lim_inf: -74.2768 | lim_sup: 123.7946
Límites IQR -> lim_inf: 1500.0000 | lim_sup: 1500.0000
Límites IQR -> lim_inf: -2526.9722 | lim_sup: 4991.3537
Límites IQR -> lim_inf: -1619.3764 | lim_sup: 5001.8196
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 8634.8700 | lim_sup: 8634.8700
Límites IQR -> lim_inf: -826.6700 | lim_sup: 1602.8500
Límites IQR -> lim_inf: -20995.3531 | lim_sup: 38589.6464
Límites IQR -> lim_inf: -260.9250 | lim_sup: 1453.8550
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -175.3575 | lim_sup: 292.2625
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -356.2500 | lim_sup: 893.7500
Límites IQR -> lim_inf: -190.7025 | lim_sup: 317.8375
Límites IQR -> lim_inf: -262.7225 | lim_sup: 1019.6175
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 77.8200 | lim_sup: 77.8200
Límites IQR -> lim_inf: 121.4300 | lim_sup: 121.4300
Límites IQR -> lim_inf: -317.9725 | lim_sup: 1556.6075
Límites IQR -> lim_inf: -217.6950 | lim_sup: 872.5650
Límites IQR -> lim_inf: -215.4825 | lim_sup: 359.1375
Límites IQR -> lim_inf: -667.2300 | lim_sup: 1112.0500
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -58.3650 | lim_sup: 97.2750
Límites IQR -> lim_inf: 121.4300 | lim_sup: 121.4300
Límites IQR -> lim_inf: -274.0913 | lim_sup: 1435.3588
Límites IQR -> lim_inf: -206.4938 | lim_sup: 453.8963
Límites IQR -> lim_inf: -217.7325 | lim_sup: 362.8875
Límites IQR -> lim_inf: -766.9088 | lim_sup: 1622.1813
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -58.3650 | lim_sup: 97.2750
Límites IQR -> lim_inf: -122.0362 | lim_sup: 203.3938
Límites IQR -> lim_inf: -78.4938 | lim_sup: 1447.7163
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -120.0000 | lim_sup: 200.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 152.0000 | lim_sup: 152.0000
Límites IQR -> lim_inf: 154.7000 | lim_sup: 154.7000
Límites IQR -> lim_inf: -34.1250 | lim_sup: 146.2350
Límites IQR -> lim_inf: -17.6625 | lim_sup: 188.5775
Límites IQR -> lim_inf: 22.7237 | lim_sup: 36.0938
Límites IQR -> lim_inf: 428.0000 | lim_sup: 428.0000
Límites IQR -> lim_inf: 168.6660 | lim_sup: 168.6660
Límites IQR -> lim_inf: 291.6067 | lim_sup: 1601.9888
Límites IQR -> lim_inf: -7623.2005 | lim_sup: 21970.8795
Límites IQR -> lim_inf: -465.7650 | lim_sup: 776.2750
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 24.3337 | lim_sup: 54.1638
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: -139.1250 | lim_sup: 231.8750
Límites IQR -> lim_inf: -761.6400 | lim_sup: 1878.2000
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 51.5000 | lim_sup: 51.5000
Límites IQR -> lim_inf: -462.0300 | lim_sup: 1690.0900
Límites IQR -> lim_inf: -651.0875 | lim_sup: 2074.6525
Límites IQR -> lim_inf: -186.7500 | lim_sup: 311.2500
Límites IQR -> lim_inf: -750.0100 | lim_sup: 1309.4300
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 51.5000 | lim_sup: 51.5000
Límites IQR -> lim_inf: 49.3000 | lim_sup: 49.3000
Límites IQR -> lim_inf: -334.7350 | lim_sup: 1727.6450
Límites IQR -> lim_inf: -510.0000 | lim_sup: 1990.0000
Límites IQR -> lim_inf: 109.2100 | lim_sup: 109.2100
Límites IQR -> lim_inf: -720.7975 | lim_sup: 1290.5825
Límites IQR -> lim_inf: 0.0000 | lim_sup: 0.0000
Límites IQR -> lim_inf: 80.1500 | lim_sup: 80.1500
Límites IQR -> lim_inf: -156.7725 | lim_sup: 1028.8475
Límites IQR -> lim_inf: -378.9900 | lim_sup: 1706.1140
Límites IQR -> lim_inf: -39.4003 | lim_sup: 84.3674
Límites IQR -> lim_inf: -14807.2421 | lim_sup: 35168.2489
Límites IQR -> lim_inf: -161.2000 | lim_sup: 486.0000
Límites IQR -> lim_inf: -152.6970 | lim_sup: 310.1030
Límites IQR -> lim_inf: -3601.5086 | lim_sup: 11128.6305
Límites IQR -> lim_inf: -603.1233 | lim_sup: 1336.7940
Límites IQR -> lim_inf: -32.3760 | lim_sup: 53.9600
Límites IQR -> lim_inf: -6060.5760 | lim_sup: 16436.3939
Límites IQR -> lim_inf: -322.8650 | lim_sup: 700.0950
Límites IQR -> lim_inf: -32.3730 | lim_sup: 53.9550
Límites IQR -> lim_inf: -4356.6329 | lim_sup: 11773.4665
Límites IQR -> lim_inf: -636.0932 | lim_sup: 2336.5490
Límites IQR -> lim_inf: -48.4530 | lim_sup: 147.4217
Límites IQR -> lim_inf: -9819.0923 | lim_sup: 24961.2518
Límites IQR -> lim_inf: -866.5000 | lim_sup: 2029.5000
Límites IQR -> lim_inf: -47.4062 | lim_sup: 154.0497
Límites IQR -> lim_inf: -6537.4124 | lim_sup: 17202.5328
Límites IQR -> lim_inf: -291.1000 | lim_sup: 634.5000
Límites IQR -> lim_inf: -188.0070 | lim_sup: 368.9530
Límites IQR -> lim_inf: -8637.9536 | lim_sup: 19038.8377
Límites IQR -> lim_inf: 80.0000 | lim_sup: 80.0000
Límites IQR -> lim_inf: -1363.6110 | lim_sup: 2741.3666
Límites IQR -> lim_inf: -108.0000 | lim_sup: 180.0000
Límites IQR -> lim_inf: -10229.8078 | lim_sup: 24243.8797
Límites IQR -> lim_inf: 18.7700 | lim_sup: 25.5700
Límites IQR -> lim_inf: -260.8513 | lim_sup: 837.9388
Límites IQR -> lim_inf: 71.6650 | lim_sup: 71.6650
Límites IQR -> lim_inf: -4658.8991 | lim_sup: 14887.7319
Límites IQR -> lim_inf: 5204.1303 | lim_sup: 6657.8362
Límites IQR -> lim_inf: -1708.4500 | lim_sup: 5454.5500
Límites IQR -> lim_inf: 94.9400 | lim_sup: 94.9400
Límites IQR -> lim_inf: 260.0000 | lim_sup: 260.0000
Límites IQR -> lim_inf: 240.0000 | lim_sup: 240.0000
Límites IQR -> lim_inf: -44.6375 | lim_sup: 1213.0225
Límites IQR -> lim_inf: 80.1500 | lim_sup: 80.1500
Límites IQR -> lim_inf: 260.0000 | lim_sup: 260.0000
Límites IQR -> lim_inf: 240.0000 | lim_sup: 240.0000
Límites IQR -> lim_inf: -144.2250 | lim_sup: 1059.6350
Límites IQR -> lim_inf: 80.1500 | lim_sup: 80.1500
Límites IQR -> lim_inf: 260.0000 | lim_sup: 260.0000
Límites IQR -> lim_inf: 4.0000 | lim_sup: 308.0000
Límites IQR -> lim_inf: -214.3825 | lim_sup: 692.5975
Límites IQR -> lim_inf: 80.1500 | lim_sup: 80.1500
Límites IQR -> lim_inf: -11115.9455 | lim_sup: 72335.4364
Límites IQR -> lim_inf: 13323.2636 | lim_sup: 13323.2636
ITE2 train enriquecido: (72492, 9)
=== RESUMEN % ITE2 (2022–2024 vs 2025 YTD) ===
clasificacion_estad_global_train_vs_test  CAMBIO  ESTABLE
variable         variable
cost_float_mod   cost_float_mod            84.51    15.49
cost_float_mod_2 cost_float_mod_2          84.11    15.89
cost_float_mod_3 cost_float_mod_3          83.10    16.90

=== Muestra detalle ITE2 ===
   ID_BUILDING     FM_COST_TYPE          variable  ratio_mediana  ratio_mad  \
0            2        Licencias    cost_float_mod            inf        1.0
1            2        Licencias  cost_float_mod_2       0.183432        0.0
2            2        Licencias  cost_float_mod_3            inf        1.0
3            2  Mtto. Contratos    cost_float_mod       1.000000        0.0
4            2  Mtto. Contratos  cost_float_mod_2       1.000000        0.0

    p_value  WAPE_naive_anual  ratio_anual  delta_prop_zeros  flag_mediana  \
0  1.000000          1.000000     0.969119          0.885714             1
1  1.000000          0.155000     0.969119          0.342857             0
2  1.000000          1.000000     0.877921          0.885714             1
3  0.037097          1.135495     0.805223          0.611111             1
4  0.037097          1.135495     0.805223          0.611111             1

   flag_mad  flag_pvalue  flag_wape  flag_anual  flag_sparsidad  n_flags  \
0         0            0          1           1               1        4
1         1            0          0           1               1        3
2         0            0          1           1               1        4
3         1            1          1           1               1        6
4         1            1          1           1               1        6

  clasificacion_estad_global_train_vs_test
0                                   CAMBIO
1                                   CAMBIO
2                                   CAMBIO
3                                   CAMBIO
4                                   CAMBIO

# =========================================
# 2) Clasificación 2022–2024 vs 2025 YTD
#     - TEST: siempre cost_float_mod real
#     - TRAIN: probamos mod, mod_2, mod_3
# =========================================
vars_coste = ["cost_float_mod", "cost_float_mod_2", "cost_float_mod_3"]
resultados = []

pairs_train = df_ite2_train_enr[["ID_BUILDING","FM_COST_TYPE"]].drop_duplicates()
pairs_test  = df_fd1_v5_ITE2_test[["ID_BUILDING","FM_COST_TYPE"]].drop_duplicates()
pairs_both  = pairs_train.merge(pairs_test, on=["ID_BUILDING","FM_COST_TYPE"])

for _, row in pairs_both.iterrows():
    bid, ctype = row["ID_BUILDING"], row["FM_COST_TYPE"]

    g_train = df_ite2_train_enr[(df_ite2_train_enr["ID_BUILDING"]==bid) & (df_ite2_train_enr["FM_COST_TYPE"]==ctype)]
    g_test  = df_fd1_v5_ITE2_test[(df_fd1_v5_ITE2_test["ID_BUILDING"]==bid) & (df_fd1_v5_ITE2_test["FM_COST_TYPE"]==ctype)]
    if g_train.empty or g_test.empty:
        continue

    s_test = _to_ms(g_test, "cost_float_mod")
    s_test = s_test[s_test.index.year == 2025]  # 2025 YTD

    for var in vars_coste:
        s_train = _to_ms(g_train, var)
        s_train = s_train[(s_train.index.year >= 2022) & (s_train.index.year <= 2024)]
        if len(s_train)==0 or len(s_test)==0:
            continue

        res = evaluar_estabilidad_robusta(
            s_train, s_test,
            alpha_pvalue=0.05,
            umbral_mediana=0.30,
            umbral_mad_hi=1.75,
            umbral_wape=0.50,
            umbral_anual=0.30,
            umbral_delta_zeros=0.30,
            regla_mayoria_k=2
        )
        resultados.append({
            "ID_BUILDING": bid,
            "FM_COST_TYPE": ctype,
            "variable": var,
            **res
        })

df_estabilidad_ite2 = pd.DataFrame(resultados)

resumen_ite2 = (
    df_estabilidad_ite2.groupby(["variable","clasificacion_estad_global_train_vs_test"])
    .size().groupby(level=0).apply(lambda x: 100*x/x.sum())
    .unstack(fill_value=0)
).round(2)

print("=== RESUMEN % ITE2 (2022–2024 vs 2025 YTD) ===")
print(resumen_ite2)

print("\n=== Muestra detalle ITE2 ===")
print(df_estabilidad_ite2.head())

=== RESUMEN % ITE2 (2022–2024 vs 2025 YTD) ===
clasificacion_estad_global_train_vs_test  CAMBIO  ESTABLE
variable         variable
cost_float_mod   cost_float_mod            84.51    15.49
cost_float_mod_2 cost_float_mod_2          84.11    15.89
cost_float_mod_3 cost_float_mod_3          83.10    16.90

=== Muestra detalle ITE2 ===
   ID_BUILDING     FM_COST_TYPE          variable  ratio_mediana  ratio_mad  \
0            2        Licencias    cost_float_mod            inf        1.0
1            2        Licencias  cost_float_mod_2       0.183432        0.0
2            2        Licencias  cost_float_mod_3            inf        1.0
3            2  Mtto. Contratos    cost_float_mod       1.000000        0.0
4            2  Mtto. Contratos  cost_float_mod_2       1.000000        0.0

    p_value  WAPE_naive_anual  ratio_anual  delta_prop_zeros  flag_mediana  \
0  1.000000          1.000000     0.969119          0.885714             1
1  1.000000          0.155000     0.969119          0.342857             0
2  1.000000          1.000000     0.877921          0.885714             1
3  0.037097          1.135495     0.805223          0.611111             1
4  0.037097          1.135495     0.805223          0.611111             1

   flag_mad  flag_pvalue  flag_wape  flag_anual  flag_sparsidad  n_flags  \
0         0            0          1           1               1        4
1         1            0          0           1               1        3
2         0            0          1           1               1        4
3         1            1          1           1               1        6
4         1            1          1           1               1        6

  clasificacion_estad_global_train_vs_test
0                                   CAMBIO
1                                   CAMBIO
2                                   CAMBIO
3                                   CAMBIO
4                                   CAMBIO

# ==========================================================
# 1) Función para enriquecer clasificación con variables de contexto
# ==========================================================
def enriquecer_con_contexto(df_estabilidad, df_train, nombre_iteracion="ITE"):
    # Nos quedamos con una única fila por pareja (ID_BUILDING, FM_COST_TYPE)
    base = df_train.drop_duplicates(subset=["ID_BUILDING","FM_COST_TYPE"])

    # Unimos
    df_enr = df_estabilidad.merge(
        base[
            [
                "ID_BUILDING","FM_COST_TYPE",
                "SUPPLIER_TYPE_MOD_2","FM_RESPONSIBLE_MOD",
                "TIPO_USO","COUNTRY_DEF","ID_REGION_GRUPO"
            ]
        ],
        on=["ID_BUILDING","FM_COST_TYPE"],
        how="left"
    )
    print(f"[OK] Enriquecido {nombre_iteracion}: {df_enr.shape}")
    return df_enr

# Ejemplo con ITE1 (si usas el panel 2022–2023 como referencia)
df_estab_ite1_enr = enriquecer_con_contexto(df_estabilidad_rb_22_23, df_fd1_v5_ITE1_train, "ITE1")

# Ejemplo con ITE2
df_estab_ite2_enr = enriquecer_con_contexto(df_estabilidad_ite2, df_fd1_v5_ITE2_train, "ITE2")

[OK] Enriquecido ITE1: (6600, 22)
[OK] Enriquecido ITE2: (6798, 22)

# ==========================================================
# 2) Función de análisis de distribución por variable de contexto
# ==========================================================
def distribucion_por_contexto(df_enr, variable, nombre_iteracion="ITE"):
    tabla = (
        df_enr.groupby([variable,"clasificacion_estad_global_train_vs_test"])
        .size()
        .groupby(level=0)
        .apply(lambda x: 100 * x / x.sum())
        .unstack(fill_value=0)
        .sort_values("ESTABLE", ascending=False)
    )
    print(f"\n=== Distribución {nombre_iteracion} por {variable} ===")
    return tabla.round(2)

# Ejemplo para ITE1 enriquecido
dist_tipo_gasto_ite1 = distribucion_por_contexto(df_estab_ite1_enr, "FM_COST_TYPE", "ITE1")
dist_proveedor_ite1  = distribucion_por_contexto(df_estab_ite1_enr, "SUPPLIER_TYPE_MOD_2", "ITE1")
dist_uso_ite1        = distribucion_por_contexto(df_estab_ite1_enr, "TIPO_USO", "ITE1")
dist_pais_ite1       = distribucion_por_contexto(df_estab_ite1_enr, "COUNTRY_DEF", "ITE1")
dist_region_ite1     = distribucion_por_contexto(df_estab_ite1_enr, "ID_REGION_GRUPO", "ITE1")
dist_responsable_ite1= distribucion_por_contexto(df_estab_ite1_enr, "FM_RESPONSIBLE_MOD", "ITE1")

# Ejemplo para ITE2 enriquecido
dist_tipo_gasto_ite2 = distribucion_por_contexto(df_estab_ite2_enr, "FM_COST_TYPE", "ITE2")
dist_proveedor_ite2  = distribucion_por_contexto(df_estab_ite2_enr, "SUPPLIER_TYPE_MOD_2", "ITE2")
dist_uso_ite2        = distribucion_por_contexto(df_estab_ite2_enr, "TIPO_USO", "ITE2")
dist_pais_ite2       = distribucion_por_contexto(df_estab_ite2_enr, "COUNTRY_DEF", "ITE2")
dist_region_ite2     = distribucion_por_contexto(df_estab_ite2_enr, "ID_REGION_GRUPO", "ITE2")
dist_responsable_ite2= distribucion_por_contexto(df_estab_ite2_enr, "FM_RESPONSIBLE_MOD", "ITE2")

=== Distribución ITE1 por FM_COST_TYPE ===

=== Distribución ITE1 por SUPPLIER_TYPE_MOD_2 ===

=== Distribución ITE1 por TIPO_USO ===

=== Distribución ITE1 por COUNTRY_DEF ===

=== Distribución ITE1 por ID_REGION_GRUPO ===

=== Distribución ITE1 por FM_RESPONSIBLE_MOD ===

=== Distribución ITE2 por FM_COST_TYPE ===

=== Distribución ITE2 por SUPPLIER_TYPE_MOD_2 ===

=== Distribución ITE2 por TIPO_USO ===

=== Distribución ITE2 por COUNTRY_DEF ===

=== Distribución ITE2 por ID_REGION_GRUPO ===

=== Distribución ITE2 por FM_RESPONSIBLE_MOD ===

def plot_distribucion(df_tabla, variable, nombre_iteracion="ITE"):
    plt.figure(figsize=(8,5))
    df_tabla["ESTABLE"].plot(kind="barh", color="steelblue")
    plt.xlabel("% ESTABLE")
    plt.title(f"{nombre_iteracion} - % ESTABLE por {variable}")
    plt.gca().invert_yaxis()
    plt.tight_layout()
    plt.show()

# Ejemplos con FM_COST_TYPE
plot_distribucion(dist_tipo_gasto_ite1, "FM_COST_TYPE", "ITE1")
plot_distribucion(dist_tipo_gasto_ite2, "FM_COST_TYPE", "ITE2")

# Ejemplos con SUPPLIER_TYPE_MOD_2
plot_distribucion(dist_proveedor_ite1, "SUPPLIER_TYPE_MOD_2", "ITE1")
plot_distribucion(dist_proveedor_ite2, "SUPPLIER_TYPE_MOD_2", "ITE2")

# Ejemplos con TIPO_USO
plot_distribucion(dist_uso_ite1, "TIPO_USO", "ITE1")
plot_distribucion(dist_uso_ite2, "TIPO_USO", "ITE2")

# Puedes repetir lo mismo con COUNTRY_DEF, ID_REGION_GRUPO y FM_RESPONSIBLE_MOD

def distribucion_contexto_por_variable(df_estabilidad, df_train, contexto, nombre_iteracion="ITE"):
    """
    Calcula % ESTABLE/CAMBIO por categoría de una variable de contexto,
    separado por 'variable' (cost_float_mod, cost_float_mod_2, cost_float_mod_3).
    Evitamos columnas duplicadas cuando el contexto coincide con una clave del join.
    """
    # Claves de pareja
    keys = ["ID_BUILDING", "FM_COST_TYPE"]

    # Base de contexto: una fila por pareja
    base = df_train.drop_duplicates(subset=keys)

    # Armamos la lista de columnas a traer del train
    cols_merge = keys.copy()
    if contexto not in keys:
        cols_merge.append(contexto)

    # Hacemos el merge (solo añadimos 'contexto' si no es key)
    df_enr = df_estabilidad.merge(
        base[cols_merge],
        on=keys,
        how="left",
        suffixes=("", "_ctx")
    )

    # Si el contexto es una key (p.ej. "FM_COST_TYPE"), usamos la propia columna ya presente;
    # si no lo es, usamos la columna traída del train.
    if contexto in keys:
        ctx_col = contexto  # ya existe en df_estabilidad
    else:
        ctx_col = contexto

    # Por seguridad: eliminamos posibles columnas duplicadas de nombre
    df_enr = df_enr.loc[:, ~df_enr.columns.duplicated()]

    # Tabla de distribución (%)
    tabla = (
        df_enr.groupby(["variable", ctx_col, "clasificacion_estad_global_train_vs_test"])
              .size()
              .groupby(level=[0, 1])
              .apply(lambda x: 100 * x / x.sum())
              .unstack(fill_value=0)
              .round(2)
    )

    print(f"\n=== {nombre_iteracion} — Distribución por {contexto} ===")
    return tabla

# ====== ITE2 ======
tablas_contexto_ite2 = {}
for contexto in ["FM_COST_TYPE", "TIPO_USO", "COUNTRY_DEF", "ID_REGION_GRUPO", "FM_RESPONSIBLE_MOD"]:
    tablas_contexto_ite2[contexto] = distribucion_contexto_por_variable(
        df_estabilidad_ite2, df_fd1_v5_ITE2_train, contexto, "ITE2"
    )
    print(tablas_contexto_ite2[contexto].to_string())

# ====== ITE1 (ajusta el DF de estabilidad que estés usando: df_estabilidad_rb_22_23 o similar) ======
tablas_contexto_ite1 = {}
for contexto in ["FM_COST_TYPE", "TIPO_USO", "COUNTRY_DEF", "ID_REGION_GRUPO", "FM_RESPONSIBLE_MOD"]:
    tablas_contexto_ite1[contexto] = distribucion_contexto_por_variable(
        df_estabilidad_rb_22_23, df_fd1_v5_ITE1_train, contexto, "ITE1"
    )
    print(tablas_contexto_ite1[contexto].to_string())

=== ITE2 — Distribución por FM_COST_TYPE ===
clasificacion_estad_global_train_vs_test                                       CAMBIO  ESTABLE
variable         FM_COST_TYPE          variable         FM_COST_TYPE
cost_float_mod   Eficiencia Energética cost_float_mod   Eficiencia Energética   80.00    20.00
                 Licencias             cost_float_mod   Licencias              100.00     0.00
                 Mtto. Contratos       cost_float_mod   Mtto. Contratos         86.96    13.04
                 Mtto. Correctivo      cost_float_mod   Mtto. Correctivo        85.11    14.89
                 Obras                 cost_float_mod   Obras                   95.00     5.00
                 Servicios Ctto.       cost_float_mod   Servicios Ctto.         81.28    18.72
                 Servicios Extra       cost_float_mod   Servicios Extra         95.48     4.52
                 Suministros           cost_float_mod   Suministros             71.96    28.04
cost_float_mod_2 Eficiencia Energética cost_float_mod_2 Eficiencia Energética   82.22    17.78
                 Licencias             cost_float_mod_2 Licencias               99.39     0.61
                 Mtto. Contratos       cost_float_mod_2 Mtto. Contratos         83.70    16.30
                 Mtto. Correctivo      cost_float_mod_2 Mtto. Correctivo        81.56    18.44
                 Obras                 cost_float_mod_2 Obras                   95.00     5.00
                 Servicios Ctto.       cost_float_mod_2 Servicios Ctto.         80.73    19.27
                 Servicios Extra       cost_float_mod_2 Servicios Extra         95.48     4.52
                 Suministros           cost_float_mod_2 Suministros             76.08    23.92
cost_float_mod_3 Eficiencia Energética cost_float_mod_3 Eficiencia Energética   68.89    31.11
                 Licencias             cost_float_mod_3 Licencias              100.00     0.00
                 Mtto. Contratos       cost_float_mod_3 Mtto. Contratos         86.41    13.59
                 Mtto. Correctivo      cost_float_mod_3 Mtto. Correctivo        83.11    16.89
                 Obras                 cost_float_mod_3 Obras                   95.00     5.00
                 Servicios Ctto.       cost_float_mod_3 Servicios Ctto.         82.40    17.60
                 Servicios Extra       cost_float_mod_3 Servicios Extra         94.58     5.42
                 Suministros           cost_float_mod_3 Suministros             68.63    31.37

=== ITE2 — Distribución por TIPO_USO ===
clasificacion_estad_global_train_vs_test                                 CAMBIO  ESTABLE
variable         TIPO_USO           variable         TIPO_USO
cost_float_mod   Almacén/Bodega     cost_float_mod   Almacén/Bodega       86.96    13.04
                 Bar/Restaurante    cost_float_mod   Bar/Restaurante      80.00    20.00
                 Bingo              cost_float_mod   Bingo                84.96    15.04
                 Casino Electrónico cost_float_mod   Casino Electrónico   71.48    28.52
                 Casino Tradicional cost_float_mod   Casino Tradicional   72.66    27.34
                 Delegación         cost_float_mod   Delegación           82.51    17.49
                 Hotel              cost_float_mod   Hotel                50.00    50.00
                 Industrial         cost_float_mod   Industrial           85.71    14.29
                 Local Apuestas     cost_float_mod   Local Apuestas       88.24    11.76
                 Oficinas           cost_float_mod   Oficinas             78.79    21.21
                 Parking            cost_float_mod   Parking             100.00     0.00
                 Salón de Juego     cost_float_mod   Salón de Juego       96.40     3.60
                 Sin actividad      cost_float_mod   Sin actividad        90.00    10.00
                 Subarrendado       cost_float_mod   Subarrendado        100.00     0.00
cost_float_mod_2 Almacén/Bodega     cost_float_mod_2 Almacén/Bodega       80.43    19.57
                 Bar/Restaurante    cost_float_mod_2 Bar/Restaurante      73.33    26.67
                 Bingo              cost_float_mod_2 Bingo                84.07    15.93
                 Casino Electrónico cost_float_mod_2 Casino Electrónico   73.89    26.11
                 Casino Tradicional cost_float_mod_2 Casino Tradicional   76.17    23.83
                 Delegación         cost_float_mod_2 Delegación           76.50    23.50
                 Hotel              cost_float_mod_2 Hotel                50.00    50.00
                 Industrial         cost_float_mod_2 Industrial           85.71    14.29
                 Local Apuestas     cost_float_mod_2 Local Apuestas       82.35    17.65
                 Oficinas           cost_float_mod_2 Oficinas             74.24    25.76
                 Parking            cost_float_mod_2 Parking             100.00     0.00
                 Salón de Juego     cost_float_mod_2 Salón de Juego       95.28     4.72
                 Sin actividad      cost_float_mod_2 Sin actividad        90.00    10.00
                 Subarrendado       cost_float_mod_2 Subarrendado        100.00     0.00
cost_float_mod_3 Almacén/Bodega     cost_float_mod_3 Almacén/Bodega       86.96    13.04
                 Bar/Restaurante    cost_float_mod_3 Bar/Restaurante      80.00    20.00
                 Bingo              cost_float_mod_3 Bingo                85.84    14.16
                 Casino Electrónico cost_float_mod_3 Casino Electrónico   67.78    32.22
                 Casino Tradicional cost_float_mod_3 Casino Tradicional   71.09    28.91
                 Delegación         cost_float_mod_3 Delegación           81.97    18.03
                 Hotel              cost_float_mod_3 Hotel                50.00    50.00
                 Industrial         cost_float_mod_3 Industrial           71.43    28.57
                 Local Apuestas     cost_float_mod_3 Local Apuestas       88.24    11.76
                 Oficinas           cost_float_mod_3 Oficinas             78.79    21.21
                 Parking            cost_float_mod_3 Parking             100.00     0.00
                 Salón de Juego     cost_float_mod_3 Salón de Juego       95.51     4.49
                 Sin actividad      cost_float_mod_3 Sin actividad        90.00    10.00
                 Subarrendado       cost_float_mod_3 Subarrendado        100.00     0.00

=== ITE2 — Distribución por COUNTRY_DEF ===
clasificacion_estad_global_train_vs_test                                     CAMBIO  ESTABLE
variable         COUNTRY_DEF          variable         COUNTRY_DEF
cost_float_mod   Colombia             cost_float_mod   Colombia               75.70    24.30
                 Costa Rica           cost_float_mod   Costa Rica             76.60    23.40
                 España               cost_float_mod   España                 91.61     8.39
                 México               cost_float_mod   México                 79.69    20.31
                 Panamá               cost_float_mod   Panamá                 63.11    36.89
                 Perú                 cost_float_mod   Perú                   70.30    29.70
                 República Dominicana cost_float_mod   República Dominicana   63.16    36.84
cost_float_mod_2 Colombia             cost_float_mod_2 Colombia               87.65    12.35
                 Costa Rica           cost_float_mod_2 Costa Rica             70.21    29.79
                 España               cost_float_mod_2 España                 89.38    10.62
                 México               cost_float_mod_2 México                 78.65    21.35
                 Panamá               cost_float_mod_2 Panamá                 61.33    38.67
                 Perú                 cost_float_mod_2 Perú                   73.27    26.73
                 República Dominicana cost_float_mod_2 República Dominicana   57.89    42.11
cost_float_mod_3 Colombia             cost_float_mod_3 Colombia               70.92    29.08
                 Costa Rica           cost_float_mod_3 Costa Rica             76.60    23.40
                 España               cost_float_mod_3 España                 91.13     8.87
                 México               cost_float_mod_3 México                 78.12    21.88
                 Panamá               cost_float_mod_3 Panamá                 59.11    40.89
                 Perú                 cost_float_mod_3 Perú                   69.31    30.69
                 República Dominicana cost_float_mod_3 República Dominicana   63.16    36.84

=== ITE2 — Distribución por ID_REGION_GRUPO ===
clasificacion_estad_global_train_vs_test                           CAMBIO  ESTABLE
variable         ID_REGION_GRUPO variable         ID_REGION_GRUPO
cost_float_mod   100             cost_float_mod   100               66.67    33.33
                 1000005         cost_float_mod   1000005           66.67    33.33
                 1000010         cost_float_mod   1000010           80.00    20.00
                 1000015         cost_float_mod   1000015           83.33    16.67
                 1000016         cost_float_mod   1000016           93.75     6.25
                 1000017         cost_float_mod   1000017           83.33    16.67
                 1000018         cost_float_mod   1000018           83.33    16.67
                 1000019         cost_float_mod   1000019           61.54    38.46
                 1000020         cost_float_mod   1000020           73.33    26.67
                 1000021         cost_float_mod   1000021           83.33    16.67
                 1000022         cost_float_mod   1000022           66.10    33.90
                 1000023         cost_float_mod   1000023           76.19    23.81
                 1000030         cost_float_mod   1000030           88.89    11.11
                 1000031         cost_float_mod   1000031          100.00     0.00
                 1000038         cost_float_mod   1000038          100.00     0.00
                 1000039         cost_float_mod   1000039           63.16    36.84
                 1000054         cost_float_mod   1000054           81.82    18.18
                 1000058         cost_float_mod   1000058           63.64    36.36
                 1000096         cost_float_mod   1000096           76.92    23.08
                 1000118         cost_float_mod   1000118           85.71    14.29
                 103             cost_float_mod   103               74.67    25.33
                 104             cost_float_mod   104               66.67    33.33
                 105             cost_float_mod   105               86.67    13.33
                 106             cost_float_mod   106               83.33    16.67
                 107             cost_float_mod   107               33.33    66.67
                 108             cost_float_mod   108               66.67    33.33
                 109             cost_float_mod   109              100.00     0.00
                 11              cost_float_mod   11                94.44     5.56
                 112             cost_float_mod   112               75.00    25.00
                 113             cost_float_mod   113               77.78    22.22
                 115             cost_float_mod   115              100.00     0.00
                 117             cost_float_mod   117               51.61    48.39
                 12              cost_float_mod   12                83.33    16.67
                 15              cost_float_mod   15                97.37     2.63
                 17              cost_float_mod   17                84.51    15.49
                 18              cost_float_mod   18                95.00     5.00
                 2               cost_float_mod   2                 90.52     9.48
                 21              cost_float_mod   21                92.45     7.55
                 27              cost_float_mod   27                96.97     3.03
                 3               cost_float_mod   3                 91.43     8.57
                 32              cost_float_mod   32                94.24     5.76
                 36              cost_float_mod   36                92.31     7.69
                 4               cost_float_mod   4                 89.29    10.71
                 48              cost_float_mod   48                84.44    15.56
                 5               cost_float_mod   5                 93.88     6.12
                 6               cost_float_mod   6                 98.59     1.41
                 7               cost_float_mod   7                 96.55     3.45
                 8               cost_float_mod   8                 93.60     6.40
                 94              cost_float_mod   94                61.88    38.12
                 96              cost_float_mod   96                72.22    27.78
                 Otros_Colombia  cost_float_mod   Otros_Colombia    73.33    26.67
                 Otros_España    cost_float_mod   Otros_España      89.24    10.76
                 Otros_México    cost_float_mod   Otros_México      77.14    22.86
                 Otros_Panamá    cost_float_mod   Otros_Panamá      54.17    45.83
                 Otros_Perú      cost_float_mod   Otros_Perú        76.19    23.81
cost_float_mod_2 100             cost_float_mod_2 100               50.00    50.00
                 1000005         cost_float_mod_2 1000005           66.67    33.33
                 1000010         cost_float_mod_2 1000010           70.00    30.00
                 1000015         cost_float_mod_2 1000015           83.33    16.67
                 1000016         cost_float_mod_2 1000016           75.00    25.00
                 1000017         cost_float_mod_2 1000017           83.33    16.67
                 1000018         cost_float_mod_2 1000018           83.33    16.67
                 1000019         cost_float_mod_2 1000019           69.23    30.77
                 1000020         cost_float_mod_2 1000020           73.33    26.67
                 1000021         cost_float_mod_2 1000021           83.33    16.67
                 1000022         cost_float_mod_2 1000022           67.80    32.20
                 1000023         cost_float_mod_2 1000023           76.19    23.81
                 1000030         cost_float_mod_2 1000030           88.89    11.11
                 1000031         cost_float_mod_2 1000031          100.00     0.00
                 1000038         cost_float_mod_2 1000038           80.00    20.00
                 1000039         cost_float_mod_2 1000039           57.89    42.11
                 1000054         cost_float_mod_2 1000054           81.82    18.18
                 1000058         cost_float_mod_2 1000058           63.64    36.36
                 1000096         cost_float_mod_2 1000096           84.62    15.38
                 1000118         cost_float_mod_2 1000118           85.71    14.29
                 103             cost_float_mod_2 103               86.67    13.33
                 104             cost_float_mod_2 104               83.33    16.67
                 105             cost_float_mod_2 105               95.00     5.00
                 106             cost_float_mod_2 106              100.00     0.00
                 107             cost_float_mod_2 107               66.67    33.33
                 108             cost_float_mod_2 108              100.00     0.00
                 109             cost_float_mod_2 109              100.00     0.00
                 11              cost_float_mod_2 11                94.44     5.56
                 112             cost_float_mod_2 112               91.67     8.33
                 113             cost_float_mod_2 113               88.89    11.11
                 115             cost_float_mod_2 115              100.00     0.00
                 117             cost_float_mod_2 117               74.19    25.81
                 12              cost_float_mod_2 12                80.00    20.00
                 15              cost_float_mod_2 15                94.74     5.26
                 17              cost_float_mod_2 17                85.92    14.08
                 18              cost_float_mod_2 18                92.50     7.50
                 2               cost_float_mod_2 2                 86.64    13.36
                 21              cost_float_mod_2 21                94.34     5.66
                 27              cost_float_mod_2 27                93.94     6.06
                 3               cost_float_mod_2 3                 94.29     5.71
                 32              cost_float_mod_2 32                92.81     7.19
                 36              cost_float_mod_2 36                88.46    11.54
                 4               cost_float_mod_2 4                 89.29    10.71
                 48              cost_float_mod_2 48                75.56    24.44
                 5               cost_float_mod_2 5                 93.88     6.12
                 6               cost_float_mod_2 6                 95.77     4.23
                 7               cost_float_mod_2 7                100.00     0.00
                 8               cost_float_mod_2 8                 91.20     8.80
                 94              cost_float_mod_2 94                61.25    38.75
                 96              cost_float_mod_2 96                66.67    33.33
                 Otros_Colombia  cost_float_mod_2 Otros_Colombia    80.00    20.00
                 Otros_España    cost_float_mod_2 Otros_España      86.05    13.95
                 Otros_México    cost_float_mod_2 Otros_México      77.14    22.86
                 Otros_Panamá    cost_float_mod_2 Otros_Panamá      54.17    45.83
                 Otros_Perú      cost_float_mod_2 Otros_Perú        85.71    14.29
cost_float_mod_3 100             cost_float_mod_3 100               66.67    33.33
                 1000005         cost_float_mod_3 1000005           66.67    33.33
                 1000010         cost_float_mod_3 1000010           80.00    20.00
                 1000015         cost_float_mod_3 1000015           83.33    16.67
                 1000016         cost_float_mod_3 1000016           81.25    18.75
                 1000017         cost_float_mod_3 1000017           83.33    16.67
                 1000018         cost_float_mod_3 1000018           83.33    16.67
                 1000019         cost_float_mod_3 1000019           69.23    30.77
                 1000020         cost_float_mod_3 1000020           66.67    33.33
                 1000021         cost_float_mod_3 1000021           66.67    33.33
                 1000022         cost_float_mod_3 1000022           64.41    35.59
                 1000023         cost_float_mod_3 1000023           76.19    23.81
                 1000030         cost_float_mod_3 1000030           88.89    11.11
                 1000031         cost_float_mod_3 1000031          100.00     0.00
                 1000038         cost_float_mod_3 1000038          100.00     0.00
                 1000039         cost_float_mod_3 1000039           63.16    36.84
                 1000054         cost_float_mod_3 1000054           81.82    18.18
                 1000058         cost_float_mod_3 1000058           72.73    27.27
                 1000096         cost_float_mod_3 1000096           69.23    30.77
                 1000118         cost_float_mod_3 1000118           85.71    14.29
                 103             cost_float_mod_3 103               70.67    29.33
                 104             cost_float_mod_3 104               66.67    33.33
                 105             cost_float_mod_3 105               78.33    21.67
                 106             cost_float_mod_3 106               83.33    16.67
                 107             cost_float_mod_3 107               33.33    66.67
                 108             cost_float_mod_3 108               66.67    33.33
                 109             cost_float_mod_3 109               83.33    16.67
                 11              cost_float_mod_3 11                94.44     5.56
                 112             cost_float_mod_3 112               66.67    33.33
                 113             cost_float_mod_3 113               77.78    22.22
                 115             cost_float_mod_3 115               83.33    16.67
                 117             cost_float_mod_3 117               48.39    51.61
                 12              cost_float_mod_3 12                86.67    13.33
                 15              cost_float_mod_3 15                97.37     2.63
                 17              cost_float_mod_3 17                83.10    16.90
                 18              cost_float_mod_3 18                95.00     5.00
                 2               cost_float_mod_3 2                 90.52     9.48
                 21              cost_float_mod_3 21                90.57     9.43
                 27              cost_float_mod_3 27                96.97     3.03
                 3               cost_float_mod_3 3                 94.29     5.71
                 32              cost_float_mod_3 32                92.81     7.19
                 36              cost_float_mod_3 36                92.31     7.69
                 4               cost_float_mod_3 4                 89.29    10.71
                 48              cost_float_mod_3 48                84.44    15.56
                 5               cost_float_mod_3 5                 93.88     6.12
                 6               cost_float_mod_3 6                 98.59     1.41
                 7               cost_float_mod_3 7                 96.55     3.45
                 8               cost_float_mod_3 8                 92.80     7.20
                 94              cost_float_mod_3 94                56.88    43.12
                 96              cost_float_mod_3 96                72.22    27.78
                 Otros_Colombia  cost_float_mod_3 Otros_Colombia    80.00    20.00
                 Otros_España    cost_float_mod_3 Otros_España      88.08    11.92
                 Otros_México    cost_float_mod_3 Otros_México      75.71    24.29
                 Otros_Panamá    cost_float_mod_3 Otros_Panamá      50.00    50.00
                 Otros_Perú      cost_float_mod_3 Otros_Perú        76.19    23.81

=== ITE2 — Distribución por FM_RESPONSIBLE_MOD ===
clasificacion_estad_global_train_vs_test                                       CAMBIO  ESTABLE
variable         FM_RESPONSIBLE_MOD    variable         FM_RESPONSIBLE_MOD
cost_float_mod   Eficiencia Energética cost_float_mod   Eficiencia Energética   72.61    27.39
                 Gestión Espacios      cost_float_mod   Gestión Espacios       100.00     0.00
                 Licencias             cost_float_mod   Licencias              100.00     0.00
                 Mantenimiento         cost_float_mod   Mantenimiento           86.94    13.06
                 Obras Proyectos       cost_float_mod   Obras Proyectos         93.55     6.45
cost_float_mod_2 Eficiencia Energética cost_float_mod_2 Eficiencia Energética   76.58    23.42
                 Gestión Espacios      cost_float_mod_2 Gestión Espacios       100.00     0.00
                 Licencias             cost_float_mod_2 Licencias               99.39     0.61
                 Mantenimiento         cost_float_mod_2 Mantenimiento           84.95    15.05
                 Obras Proyectos       cost_float_mod_2 Obras Proyectos         93.55     6.45
cost_float_mod_3 Eficiencia Energética cost_float_mod_3 Eficiencia Energética   68.65    31.35
                 Gestión Espacios      cost_float_mod_3 Gestión Espacios       100.00     0.00
                 Licencias             cost_float_mod_3 Licencias              100.00     0.00
                 Mantenimiento         cost_float_mod_3 Mantenimiento           86.27    13.73
                 Obras Proyectos       cost_float_mod_3 Obras Proyectos         93.55     6.45

=== ITE1 — Distribución por FM_COST_TYPE ===
clasificacion_estad_global_train_vs_test                                       CAMBIO  ESTABLE
variable         FM_COST_TYPE          variable         FM_COST_TYPE
cost_float_mod   Eficiencia Energética cost_float_mod   Eficiencia Energética   56.25    43.75
                 Licencias             cost_float_mod   Licencias               98.31     1.69
                 Mtto. Contratos       cost_float_mod   Mtto. Contratos         84.44    15.56
                 Mtto. Correctivo      cost_float_mod   Mtto. Correctivo        78.67    21.33
                 Obras                 cost_float_mod   Obras                   93.55     6.45
                 Servicios Ctto.       cost_float_mod   Servicios Ctto.         83.18    16.82
                 Servicios Extra       cost_float_mod   Servicios Extra         90.54     9.46
                 Suministros           cost_float_mod   Suministros             82.37    17.63
cost_float_mod_2 Eficiencia Energética cost_float_mod_2 Eficiencia Energética   56.25    43.75
                 Licencias             cost_float_mod_2 Licencias               98.31     1.69
                 Mtto. Contratos       cost_float_mod_2 Mtto. Contratos         84.17    15.83
                 Mtto. Correctivo      cost_float_mod_2 Mtto. Correctivo        77.52    22.48
                 Obras                 cost_float_mod_2 Obras                   93.55     6.45
                 Servicios Ctto.       cost_float_mod_2 Servicios Ctto.         81.38    18.62
                 Servicios Extra       cost_float_mod_2 Servicios Extra         91.40     8.60
                 Suministros           cost_float_mod_2 Suministros             82.80    17.20
cost_float_mod_3 Eficiencia Energética cost_float_mod_3 Eficiencia Energética   56.25    43.75
                 Licencias             cost_float_mod_3 Licencias               98.31     1.69
                 Mtto. Contratos       cost_float_mod_3 Mtto. Contratos         85.56    14.44
                 Mtto. Correctivo      cost_float_mod_3 Mtto. Correctivo        80.05    19.95
                 Obras                 cost_float_mod_3 Obras                   93.55     6.45
                 Servicios Ctto.       cost_float_mod_3 Servicios Ctto.         84.08    15.92
                 Servicios Extra       cost_float_mod_3 Servicios Extra         90.83     9.17
                 Suministros           cost_float_mod_3 Suministros             81.72    18.28

=== ITE1 — Distribución por TIPO_USO ===
clasificacion_estad_global_train_vs_test                                 CAMBIO  ESTABLE
variable         TIPO_USO           variable         TIPO_USO
cost_float_mod   Almacén/Bodega     cost_float_mod   Almacén/Bodega       88.89    11.11
                 Bar/Restaurante    cost_float_mod   Bar/Restaurante      75.00    25.00
                 Bingo              cost_float_mod   Bingo                85.06    14.94
                 Casino Electrónico cost_float_mod   Casino Electrónico   70.83    29.17
                 Casino Tradicional cost_float_mod   Casino Tradicional   62.34    37.66
                 Delegación         cost_float_mod   Delegación           91.53     8.47
                 Hotel              cost_float_mod   Hotel                42.86    57.14
                 Industrial         cost_float_mod   Industrial           87.50    12.50
                 Local Apuestas     cost_float_mod   Local Apuestas       85.71    14.29
                 Oficinas           cost_float_mod   Oficinas             83.87    16.13
                 Parking            cost_float_mod   Parking              66.67    33.33
                 Salón de Juego     cost_float_mod   Salón de Juego       96.86     3.14
                 Sin actividad      cost_float_mod   Sin actividad        92.31     7.69
                 Subarrendado       cost_float_mod   Subarrendado        100.00     0.00
cost_float_mod_2 Almacén/Bodega     cost_float_mod_2 Almacén/Bodega       88.89    11.11
                 Bar/Restaurante    cost_float_mod_2 Bar/Restaurante      75.00    25.00
                 Bingo              cost_float_mod_2 Bingo                84.23    15.77
                 Casino Electrónico cost_float_mod_2 Casino Electrónico   71.98    28.02
                 Casino Tradicional cost_float_mod_2 Casino Tradicional   60.67    39.33
                 Delegación         cost_float_mod_2 Delegación           90.40     9.60
                 Hotel              cost_float_mod_2 Hotel                57.14    42.86
                 Industrial         cost_float_mod_2 Industrial           75.00    25.00
                 Local Apuestas     cost_float_mod_2 Local Apuestas       85.71    14.29
                 Oficinas           cost_float_mod_2 Oficinas             79.03    20.97
                 Parking            cost_float_mod_2 Parking              66.67    33.33
                 Salón de Juego     cost_float_mod_2 Salón de Juego       96.63     3.37
                 Sin actividad      cost_float_mod_2 Sin actividad        92.31     7.69
                 Subarrendado       cost_float_mod_2 Subarrendado        100.00     0.00
cost_float_mod_3 Almacén/Bodega     cost_float_mod_3 Almacén/Bodega       88.89    11.11
                 Bar/Restaurante    cost_float_mod_3 Bar/Restaurante      75.00    25.00
                 Bingo              cost_float_mod_3 Bingo                86.72    13.28
                 Casino Electrónico cost_float_mod_3 Casino Electrónico   71.59    28.41
                 Casino Tradicional cost_float_mod_3 Casino Tradicional   63.18    36.82
                 Delegación         cost_float_mod_3 Delegación           91.53     8.47
                 Hotel              cost_float_mod_3 Hotel                42.86    57.14
                 Industrial         cost_float_mod_3 Industrial          100.00     0.00
                 Local Apuestas     cost_float_mod_3 Local Apuestas       78.57    21.43
                 Oficinas           cost_float_mod_3 Oficinas             85.48    14.52
                 Parking            cost_float_mod_3 Parking              66.67    33.33
                 Salón de Juego     cost_float_mod_3 Salón de Juego       96.98     3.02
                 Sin actividad      cost_float_mod_3 Sin actividad        84.62    15.38
                 Subarrendado       cost_float_mod_3 Subarrendado        100.00     0.00

=== ITE1 — Distribución por COUNTRY_DEF ===
clasificacion_estad_global_train_vs_test                                     CAMBIO  ESTABLE
variable         COUNTRY_DEF          variable         COUNTRY_DEF
cost_float_mod   Colombia             cost_float_mod   Colombia               81.25    18.75
                 Costa Rica           cost_float_mod   Costa Rica             53.85    46.15
                 España               cost_float_mod   España                 93.09     6.91
                 México               cost_float_mod   México                 74.86    25.14
                 Panamá               cost_float_mod   Panamá                 52.49    47.51
                 Perú                 cost_float_mod   Perú                   78.30    21.70
                 República Dominicana cost_float_mod   República Dominicana   34.78    65.22
cost_float_mod_2 Colombia             cost_float_mod_2 Colombia               83.93    16.07
                 Costa Rica           cost_float_mod_2 Costa Rica             51.28    48.72
                 España               cost_float_mod_2 España                 92.38     7.62
                 México               cost_float_mod_2 México                 74.32    25.68
                 Panamá               cost_float_mod_2 Panamá                 52.04    47.96
                 Perú                 cost_float_mod_2 Perú                   76.42    23.58
                 República Dominicana cost_float_mod_2 República Dominicana   43.48    56.52
cost_float_mod_3 Colombia             cost_float_mod_3 Colombia               82.14    17.86
                 Costa Rica           cost_float_mod_3 Costa Rica             53.85    46.15
                 España               cost_float_mod_3 España                 93.45     6.55
                 México               cost_float_mod_3 México                 74.86    25.14
                 Panamá               cost_float_mod_3 Panamá                 52.94    47.06
                 Perú                 cost_float_mod_3 Perú                   80.19    19.81
                 República Dominicana cost_float_mod_3 República Dominicana   39.13    60.87

=== ITE1 — Distribución por ID_REGION_GRUPO ===
clasificacion_estad_global_train_vs_test                           CAMBIO  ESTABLE
variable         ID_REGION_GRUPO variable         ID_REGION_GRUPO
cost_float_mod   100             cost_float_mod   100               50.00    50.00
                 1000005         cost_float_mod   1000005           54.55    45.45
                 1000010         cost_float_mod   1000010           66.67    33.33
                 1000015         cost_float_mod   1000015           66.67    33.33
                 1000016         cost_float_mod   1000016          100.00     0.00
                 1000017         cost_float_mod   1000017           83.33    16.67
                 1000018         cost_float_mod   1000018           83.33    16.67
                 1000019         cost_float_mod   1000019           50.00    50.00
                 1000020         cost_float_mod   1000020           72.22    27.78
                 1000021         cost_float_mod   1000021           66.67    33.33
                 1000022         cost_float_mod   1000022           76.19    23.81
                 1000023         cost_float_mod   1000023           77.27    22.73
                 1000030         cost_float_mod   1000030           77.78    22.22
                 1000031         cost_float_mod   1000031           66.67    33.33
                 1000038         cost_float_mod   1000038           37.50    62.50
                 1000039         cost_float_mod   1000039           34.78    65.22
                 1000054         cost_float_mod   1000054           50.00    50.00
                 1000058         cost_float_mod   1000058           83.33    16.67
                 1000096         cost_float_mod   1000096           61.54    38.46
                 1000118         cost_float_mod   1000118           42.86    57.14
                 103             cost_float_mod   103               81.16    18.84
                 104             cost_float_mod   104               66.67    33.33
                 105             cost_float_mod   105               88.33    11.67
                 106             cost_float_mod   106               66.67    33.33
                 107             cost_float_mod   107               33.33    66.67
                 108             cost_float_mod   108               66.67    33.33
                 109             cost_float_mod   109               83.33    16.67
                 11              cost_float_mod   11                95.31     4.69
                 112             cost_float_mod   112               88.89    11.11
                 113             cost_float_mod   113               88.89    11.11
                 115             cost_float_mod   115               83.33    16.67
                 117             cost_float_mod   117               72.41    27.59
                 12              cost_float_mod   12                77.78    22.22
                 15              cost_float_mod   15                95.12     4.88
                 17              cost_float_mod   17                85.07    14.93
                 18              cost_float_mod   18                97.67     2.33
                 2               cost_float_mod   2                 88.99    11.01
                 21              cost_float_mod   21                95.38     4.62
                 27              cost_float_mod   27               100.00     0.00
                 3               cost_float_mod   3                 93.94     6.06
                 32              cost_float_mod   32                92.80     7.20
                 36              cost_float_mod   36               100.00     0.00
                 4               cost_float_mod   4                 91.30     8.70
                 48              cost_float_mod   48                88.57    11.43
                 5               cost_float_mod   5                 98.15     1.85
                 6               cost_float_mod   6                 95.00     5.00
                 7               cost_float_mod   7                100.00     0.00
                 8               cost_float_mod   8                 96.38     3.62
                 94              cost_float_mod   94                54.14    45.86
                 96              cost_float_mod   96                50.00    50.00
                 Otros_Colombia  cost_float_mod   Otros_Colombia   100.00     0.00
                 Otros_España    cost_float_mod   Otros_España      93.56     6.44
                 Otros_México    cost_float_mod   Otros_México      74.60    25.40
                 Otros_Panamá    cost_float_mod   Otros_Panamá      45.83    54.17
                 Otros_Perú      cost_float_mod   Otros_Perú        85.71    14.29
cost_float_mod_2 100             cost_float_mod_2 100               58.33    41.67
                 1000005         cost_float_mod_2 1000005           50.00    50.00
                 1000010         cost_float_mod_2 1000010           66.67    33.33
                 1000015         cost_float_mod_2 1000015           66.67    33.33
                 1000016         cost_float_mod_2 1000016          100.00     0.00
                 1000017         cost_float_mod_2 1000017           83.33    16.67
                 1000018         cost_float_mod_2 1000018           83.33    16.67
                 1000019         cost_float_mod_2 1000019           50.00    50.00
                 1000020         cost_float_mod_2 1000020           66.67    33.33
                 1000021         cost_float_mod_2 1000021           66.67    33.33
                 1000022         cost_float_mod_2 1000022           76.19    23.81
                 1000023         cost_float_mod_2 1000023           72.73    27.27
                 1000030         cost_float_mod_2 1000030           77.78    22.22
                 1000031         cost_float_mod_2 1000031           66.67    33.33
                 1000038         cost_float_mod_2 1000038           37.50    62.50
                 1000039         cost_float_mod_2 1000039           43.48    56.52
                 1000054         cost_float_mod_2 1000054           50.00    50.00
                 1000058         cost_float_mod_2 1000058           83.33    16.67
                 1000096         cost_float_mod_2 1000096           84.62    15.38
                 1000118         cost_float_mod_2 1000118           42.86    57.14
                 103             cost_float_mod_2 103               84.06    15.94
                 104             cost_float_mod_2 104               66.67    33.33
                 105             cost_float_mod_2 105               88.33    11.67
                 106             cost_float_mod_2 106               66.67    33.33
                 107             cost_float_mod_2 107               33.33    66.67
                 108             cost_float_mod_2 108               66.67    33.33
                 109             cost_float_mod_2 109               83.33    16.67
                 11              cost_float_mod_2 11                95.31     4.69
                 112             cost_float_mod_2 112               88.89    11.11
                 113             cost_float_mod_2 113              100.00     0.00
                 115             cost_float_mod_2 115               83.33    16.67
                 117             cost_float_mod_2 117               72.41    27.59
                 12              cost_float_mod_2 12                77.78    22.22
                 15              cost_float_mod_2 15                95.12     4.88
                 17              cost_float_mod_2 17                79.10    20.90
                 18              cost_float_mod_2 18                97.67     2.33
                 2               cost_float_mod_2 2                 87.67    12.33
                 21              cost_float_mod_2 21                98.46     1.54
                 27              cost_float_mod_2 27               100.00     0.00
                 3               cost_float_mod_2 3                 93.94     6.06
                 32              cost_float_mod_2 32                91.20     8.80
                 36              cost_float_mod_2 36               100.00     0.00
                 4               cost_float_mod_2 4                 91.30     8.70
                 48              cost_float_mod_2 48                88.57    11.43
                 5               cost_float_mod_2 5                 98.15     1.85
                 6               cost_float_mod_2 6                 95.00     5.00
                 7               cost_float_mod_2 7                 96.97     3.03
                 8               cost_float_mod_2 8                 93.48     6.52
                 94              cost_float_mod_2 94                52.23    47.77
                 96              cost_float_mod_2 96                50.00    50.00
                 Otros_Colombia  cost_float_mod_2 Otros_Colombia   100.00     0.00
                 Otros_España    cost_float_mod_2 Otros_España      93.87     6.13
                 Otros_México    cost_float_mod_2 Otros_México      76.19    23.81
                 Otros_Panamá    cost_float_mod_2 Otros_Panamá      50.00    50.00
                 Otros_Perú      cost_float_mod_2 Otros_Perú        80.95    19.05
cost_float_mod_3 100             cost_float_mod_3 100               50.00    50.00
                 1000005         cost_float_mod_3 1000005           54.55    45.45
                 1000010         cost_float_mod_3 1000010           66.67    33.33
                 1000015         cost_float_mod_3 1000015           66.67    33.33
                 1000016         cost_float_mod_3 1000016          100.00     0.00
                 1000017         cost_float_mod_3 1000017           83.33    16.67
                 1000018         cost_float_mod_3 1000018           83.33    16.67
                 1000019         cost_float_mod_3 1000019           50.00    50.00
                 1000020         cost_float_mod_3 1000020           72.22    27.78
                 1000021         cost_float_mod_3 1000021           66.67    33.33
                 1000022         cost_float_mod_3 1000022           74.60    25.40
                 1000023         cost_float_mod_3 1000023           86.36    13.64
                 1000030         cost_float_mod_3 1000030           77.78    22.22
                 1000031         cost_float_mod_3 1000031           66.67    33.33
                 1000038         cost_float_mod_3 1000038           37.50    62.50
                 1000039         cost_float_mod_3 1000039           39.13    60.87
                 1000054         cost_float_mod_3 1000054           60.00    40.00
                 1000058         cost_float_mod_3 1000058           83.33    16.67
                 1000096         cost_float_mod_3 1000096           84.62    15.38
                 1000118         cost_float_mod_3 1000118           42.86    57.14
                 103             cost_float_mod_3 103               82.61    17.39
                 104             cost_float_mod_3 104               66.67    33.33
                 105             cost_float_mod_3 105               83.33    16.67
                 106             cost_float_mod_3 106               66.67    33.33
                 107             cost_float_mod_3 107               33.33    66.67
                 108             cost_float_mod_3 108               66.67    33.33
                 109             cost_float_mod_3 109               83.33    16.67
                 11              cost_float_mod_3 11                93.75     6.25
                 112             cost_float_mod_3 112               88.89    11.11
                 113             cost_float_mod_3 113              100.00     0.00
                 115             cost_float_mod_3 115               83.33    16.67
                 117             cost_float_mod_3 117               72.41    27.59
                 12              cost_float_mod_3 12                81.48    18.52
                 15              cost_float_mod_3 15                97.56     2.44
                 17              cost_float_mod_3 17                85.07    14.93
                 18              cost_float_mod_3 18                97.67     2.33
                 2               cost_float_mod_3 2                 90.75     9.25
                 21              cost_float_mod_3 21                95.38     4.62
                 27              cost_float_mod_3 27               100.00     0.00
                 3               cost_float_mod_3 3                 93.94     6.06
                 32              cost_float_mod_3 32                94.40     5.60
                 36              cost_float_mod_3 36               100.00     0.00
                 4               cost_float_mod_3 4                 95.65     4.35
                 48              cost_float_mod_3 48                88.57    11.43
                 5               cost_float_mod_3 5                 96.30     3.70
                 6               cost_float_mod_3 6                 96.67     3.33
                 7               cost_float_mod_3 7                100.00     0.00
                 8               cost_float_mod_3 8                 97.10     2.90
                 94              cost_float_mod_3 94                53.50    46.50
                 96              cost_float_mod_3 96                50.00    50.00
                 Otros_Colombia  cost_float_mod_3 Otros_Colombia   100.00     0.00
                 Otros_España    cost_float_mod_3 Otros_España      92.33     7.67
                 Otros_México    cost_float_mod_3 Otros_México      74.60    25.40
                 Otros_Panamá    cost_float_mod_3 Otros_Panamá      50.00    50.00
                 Otros_Perú      cost_float_mod_3 Otros_Perú        90.48     9.52

=== ITE1 — Distribución por FM_RESPONSIBLE_MOD ===
clasificacion_estad_global_train_vs_test                                       CAMBIO  ESTABLE
variable         FM_RESPONSIBLE_MOD    variable         FM_RESPONSIBLE_MOD
cost_float_mod   Eficiencia Energética cost_float_mod   Eficiencia Energética   79.92    20.08
                 Gestión Espacios      cost_float_mod   Gestión Espacios       100.00     0.00
                 Licencias             cost_float_mod   Licencias               98.31     1.69
                 Mantenimiento         cost_float_mod   Mantenimiento           83.90    16.10
                 Obras Proyectos       cost_float_mod   Obras Proyectos         91.30     8.70
cost_float_mod_2 Eficiencia Energética cost_float_mod_2 Eficiencia Energética   80.31    19.69
                 Gestión Espacios      cost_float_mod_2 Gestión Espacios       100.00     0.00
                 Licencias             cost_float_mod_2 Licencias               98.31     1.69
                 Mantenimiento         cost_float_mod_2 Mantenimiento           83.29    16.71
                 Obras Proyectos       cost_float_mod_2 Obras Proyectos         91.30     8.70
cost_float_mod_3 Eficiencia Energética cost_float_mod_3 Eficiencia Energética   79.34    20.66
                 Gestión Espacios      cost_float_mod_3 Gestión Espacios       100.00     0.00
                 Licencias             cost_float_mod_3 Licencias               98.31     1.69
                 Mantenimiento         cost_float_mod_3 Mantenimiento           84.84    15.16
                 Obras Proyectos       cost_float_mod_3 Obras Proyectos         91.30     8.70

# Ruta base 2 - para guardar los archivos resultado de la Estrategia 2.

ruta_base_2 = "/content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_2"

def _to_excel(df, filename, sheet_name="Sheet1"):
    if df is not None and len(df) > 0:
        df.to_excel(os.path.join(ruta_base_2, filename),
                    index=False, sheet_name=sheet_name, engine="openpyxl")

def _to_csv(df, filename):
    if df is not None and len(df) > 0:
        df.to_csv(os.path.join(ruta_base_2, filename),
                  index=False, sep=";")

# --- ITE1 ---
try: _to_excel(df_fd1_v5_ITE1_train, "df_fd1_v5_ITE1_train.xlsx"); _to_csv(df_fd1_v5_ITE1_train, "df_fd1_v5_ITE1_train.csv")
except: pass
try: _to_excel(df_fd1_v5_ITE1_test, "df_fd1_v5_ITE1_test.xlsx"); _to_csv(df_fd1_v5_ITE1_test, "df_fd1_v5_ITE1_test.csv")
except: pass
try: _to_excel(df_train_enr, "df_train_enr_ITE1.xlsx"); _to_csv(df_train_enr, "df_train_enr_ITE1.csv")
except: pass
try: _to_excel(df_estabilidad_rb_22_23, "df_estabilidad_rb_22_23_ITE1.xlsx"); _to_csv(df_estabilidad_rb_22_23, "df_estabilidad_rb_22_23_ITE1.csv")
except: pass
try: _to_excel(df_estabilidad, "df_estabilidad_ITE1.xlsx"); _to_csv(df_estabilidad, "df_estabilidad_ITE1.csv")
except: pass
try: _to_excel(df_estabilidad_rb, "df_estabilidad_rb_ITE1.xlsx"); _to_csv(df_estabilidad_rb, "df_estabilidad_rb_ITE1.csv")
except: pass

# --- ITE2 ---
try: _to_excel(df_fd1_v5_ITE2_train, "df_fd1_v5_ITE2_train.xlsx"); _to_csv(df_fd1_v5_ITE2_train, "df_fd1_v5_ITE2_train.csv")
except: pass
try: _to_excel(df_fd1_v5_ITE2_test, "df_fd1_v5_ITE2_test.xlsx"); _to_csv(df_fd1_v5_ITE2_test, "df_fd1_v5_ITE2_test.csv")
except: pass
try: _to_excel(df_ite2_train_enr, "df_ite2_train_enr.xlsx"); _to_csv(df_ite2_train_enr, "df_ite2_train_enr.csv")
except: pass
try: _to_excel(df_estabilidad_ite2, "df_estabilidad_ite2.xlsx"); _to_csv(df_estabilidad_ite2, "df_estabilidad_ite2.csv")
except: pass

# --- Resúmenes ITE1 ---
for name, df in {
    "dist_tipo_gasto_ite1":      globals().get("dist_tipo_gasto_ite1"),
    "dist_uso_ite1":             globals().get("dist_uso_ite1"),
    "dist_pais_ite1":            globals().get("dist_pais_ite1"),
    "dist_region_ite1":          globals().get("dist_region_ite1"),
    "dist_responsable_ite1":     globals().get("dist_responsable_ite1"),
}.items():
    try:
        _to_excel(df, f"{name}.xlsx")
        _to_csv(df, f"{name}.csv")
    except: pass

# --- Resúmenes ITE2 ---
for name, df in {
    "dist_tipo_gasto_ite2":      globals().get("dist_tipo_gasto_ite2"),
    "dist_uso_ite2":             globals().get("dist_uso_ite2"),
    "dist_pais_ite2":            globals().get("dist_pais_ite2"),
    "dist_region_ite2":          globals().get("dist_region_ite2"),
    "dist_responsable_ite2":     globals().get("dist_responsable_ite2"),
}.items():
    try:
        _to_excel(df, f"{name}.xlsx")
        _to_csv(df, f"{name}.csv")
    except: pass

print("Guardado completado en Excel y CSV (sep=';') en carpeta ESTRATEGIA_2.")

Guardado completado en Excel y CSV (sep=';') en carpeta ESTRATEGIA_2.

# Ruta del notebook .ipynb que vamos a convertir

RUTA_NOTEBOOK = "/content/drive/MyDrive/PROYECTO_NEITH/DOCUMENTOS_PROYECTOS/Neith.ipynb"

# Carpeta de salida y nombre del HTML
OUTPUT_DIR = "/content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO"
OUTPUT_HTML = "Neith_notebook_v5_Analisis_serie_serie_estrategia_2.html"

# Aseguramos carpeta de salida (ya en cabecera)
# import os
os.makedirs(OUTPUT_DIR, exist_ok=True)

# Instalamos nbconvert (ya en cabecera)
# !pip install -q "nbconvert>=7.0.0"

# Convertimos notebook a HTML y lo guardamos con el nombre indicado en la carpeta de salida
!jupyter nbconvert --to html "$RUTA_NOTEBOOK" --output "$OUTPUT_HTML" --output-dir "$OUTPUT_DIR"

# Verificamos
salida = os.path.join(OUTPUT_DIR, OUTPUT_HTML)
print("Archivo HTML generado en:", salida, "\nExiste:", os.path.exists(salida))

[NbConvertApp] Converting notebook /content/drive/MyDrive/PROYECTO_NEITH/DOCUMENTOS_PROYECTOS/Neith.ipynb to html
[NbConvertApp] WARNING | Alternative text is missing on 17 image(s).
[NbConvertApp] Writing 3648609 bytes to /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/Neith_notebook_v5_Analisis_serie_serie_estrategia_2.html
Archivo HTML generado en: /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/Neith_notebook_v5_Analisis_serie_serie_estrategia_2.html
Existe: True

# =========================
# Bloque 1) Constantes y parámetros
# =========================



# Definimos rutas base de Estrategia 2 y Estrategia 3
ruta_base_2 = "/content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_2"
ruta_base_3 = "/content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3"

# Definimos subcarpetas de salida
SUBFOLDERS = ["RESULTADOS", "METRICAS", "LOGS", "FIGS"]

# Definimos parámetros canónicos de pares clave y fechas
PAIR_COLS = ["ID_BUILDING","FM_COST_TYPE"]
DATECOL   = "FECHA"

# Definimos rangos de meses canónicos
TRAIN_START, TRAIN_END = "2021-01-01", "2023-12-01"
TEST_START,  TEST_END  = "2024-01-01", "2024-12-01"

# Nombres esperados de columnas de valor
VALUE_TRAINTEST = "cost_float_mod"
VALUE_ENR_MOD2  = "cost_float_mod2"   # aceptaremos variantes con guion bajo
VALUE_ENR_MOD3  = "cost_float_mod3"

# Definimos separador estándar de CSV
CSV_SEP = ";"


# =========================
# Bloque 2) Funciones auxiliares
# =========================

def ensure_dirs(base_path: str, subfolders: list):
    """Creamos la estructura de carpetas de salida para Estrategia 3."""
    os.makedirs(base_path, exist_ok=True)
    for sf in subfolders:
        os.makedirs(os.path.join(base_path, sf), exist_ok=True)

def log(msg: str):
    """Escribimos mensajes de trazabilidad en el log de preparación."""
    print(msg)
    with open(os.path.join(ruta_base_3, "LOGS", "estrategia3_setup.log"), "a", encoding="utf-8") as f:
        f.write(f"[{datetime.now().isoformat(timespec='seconds')}] {msg}\n")

def coerce_fecha(df: pd.DataFrame) -> pd.DataFrame:
    """Aseguramos la columna FECHA. Si no existe, la construimos desde YEAR y MONTH."""
    cols = set(df.columns.str.upper())
    # Aceptamos distintas capitalizaciones
    if "FECHA" in df.columns:
        df["FECHA"] = pd.to_datetime(df["FECHA"]).dt.to_period("M").dt.to_timestamp()
        return df
    # Si no existe FECHA, intentamos con YEAR + MONTH
    year_col  = "YEAR"  if "YEAR"  in df.columns else None
    month_col = "MONTH" if "MONTH" in df.columns else None
    if year_col and month_col:
        df["FECHA"] = pd.to_datetime(
            dict(year=df[year_col].astype(int), month=df[month_col].astype(int), day=1)
        ).dt.to_period("M").dt.to_timestamp()
        return df
    raise ValueError("No encontramos FECHA ni YEAR+MONTH para construir la fecha.")

def canonical_month_range(start_str, end_str):
    """Generamos el rango mensual canónico."""
    return pd.date_range(start=start_str, end=end_str, freq="MS")

def agg_remove_dups_and_reindex(df: pd.DataFrame, datecol: str, valuecol: str,
                                start: str, end: str,
                                pair_cols=PAIR_COLS, agg="sum") -> pd.DataFrame:
    """Agregamos duplicados por (pair, mes) y reindexamos mensualmente rellenando huecos a 0."""
    # Agregamos por mes canónico
    df_agg = (df.groupby(pair_cols + [pd.Grouper(key=datecol, freq="MS")], as_index=False)[valuecol]
                .agg(agg))
    # Reindexamos por pareja a malla mensual
    all_pairs = df_agg[pair_cols].drop_duplicates()
    months = canonical_month_range(start, end)
    out = []
    for bid, ctype in all_pairs.values:
        sub = df_agg[(df_agg[pair_cols[0]]==bid) & (df_agg[pair_cols[1]]==ctype)].set_index(datecol)
        sub = sub.reindex(months).rename_axis(datecol).reset_index()
        sub[pair_cols[0]] = bid
        sub[pair_cols[1]] = ctype
        sub[valuecol] = sub[valuecol].fillna(0.0)
        out.append(sub)
    if not out:
        # Devolvemos vacío con columnas esperadas si no hay pares
        cols = pair_cols + [datecol, valuecol]
        return pd.DataFrame(columns=cols)
    df_full = pd.concat(out, ignore_index=True)
    return df_full[[pair_cols[0], pair_cols[1], datecol, valuecol]]

def save_csv(df: pd.DataFrame, path: str):
    """Guardamos CSV con separador estándar del proyecto."""
    df.to_csv(path, index=False, sep=CSV_SEP)

def normalize_enr_column_names(df: pd.DataFrame) -> pd.DataFrame:
    """Normalizamos nombres de columnas de outliers si vinieran como cost_float_mod_2/_3."""
    rename_map = {}
    if "cost_float_mod_2" in df.columns and VALUE_ENR_MOD2 not in df.columns:
        rename_map["cost_float_mod_2"] = VALUE_ENR_MOD2
    if "cost_float_mod_3" in df.columns and VALUE_ENR_MOD3 not in df.columns:
        rename_map["cost_float_mod_3"] = VALUE_ENR_MOD3
    if rename_map:
        df = df.rename(columns=rename_map)
    return df


# =========================
# Bloque 3) Ejecución
# =========================

# 3.1) Preparamos carpetas y log
ensure_dirs(ruta_base_3, SUBFOLDERS)
with open(os.path.join(ruta_base_3, "LOGS", "estrategia3_setup.log"), "w", encoding="utf-8") as f:
    f.write("== LOG Preparación Estrategia 3 - Punto 0 ==\n")

log("Iniciamos el Punto 0 — Preparación.")

# 3.2) Cargamos ficheros de entrada
train_path = os.path.join(ruta_base_2, "df_fd1_v5_ITE1_train.xlsx")
test_path  = os.path.join(ruta_base_2, "df_fd1_v5_ITE1_test.xlsx")
enr_path   = os.path.join(ruta_base_2, "df_train_enr_ITE1.xlsx")

df_fd1_v5_ITE1_train = pd.read_excel(train_path, engine="openpyxl")
df_fd1_v5_ITE1_test  = pd.read_excel(test_path,  engine="openpyxl")
df_train_enr_ITE1    = pd.read_excel(enr_path,   engine="openpyxl")

log("Ficheros cargados desde Estrategia 2.")

# 3.3) Aseguramos columna FECHA mensual canónica en los tres dataframes
df_fd1_v5_ITE1_train = coerce_fecha(df_fd1_v5_ITE1_train.copy())
df_fd1_v5_ITE1_test  = coerce_fecha(df_fd1_v5_ITE1_test.copy())
df_train_enr_ITE1    = coerce_fecha(df_train_enr_ITE1.copy())

# 3.4) Validamos/normalizamos columnas de outliers en df_train_enr_ITE1
df_train_enr_ITE1 = normalize_enr_column_names(df_train_enr_ITE1)

# 3.5) Comprobamos presencia de columnas clave
missing_train = [c for c in [*PAIR_COLS, DATECOL, VALUE_TRAINTEST] if c not in df_fd1_v5_ITE1_train.columns]
missing_test  = [c for c in [*PAIR_COLS, DATECOL, VALUE_TRAINTEST] if c not in df_fd1_v5_ITE1_test.columns]
missing_enr   = [c for c in [*PAIR_COLS, DATECOL, VALUE_TRAINTEST, "is_outlier", VALUE_ENR_MOD2, VALUE_ENR_MOD3] if c not in df_train_enr_ITE1.columns]

if missing_train: log(f"AVISO: faltan columnas en TRAIN: {missing_train}")
if missing_test:  log(f"AVISO: faltan columnas en TEST : {missing_test}")
if missing_enr:   log(f"AVISO: faltan columnas en ENR  : {missing_enr}")

# 3.6) Reindexamos mensualmente (agregando duplicados por suma y rellenando huecos a 0)
train_full = agg_remove_dups_and_reindex(df_fd1_v5_ITE1_train, DATECOL, VALUE_TRAINTEST,
                                         start=TRAIN_START, end=TRAIN_END, pair_cols=PAIR_COLS, agg="sum")
test_full  = agg_remove_dups_and_reindex(df_fd1_v5_ITE1_test,  DATECOL, VALUE_TRAINTEST,
                                         start=TEST_START, end=TEST_END, pair_cols=PAIR_COLS, agg="sum")

# 3.7) Construimos versión enriquecida para train (merge de mod2, mod3, is_outlier)
cols_enr_keep = [*PAIR_COLS, DATECOL, VALUE_TRAINTEST, "is_outlier", VALUE_ENR_MOD2, VALUE_ENR_MOD3]
cols_enr_keep = [c for c in cols_enr_keep if c in df_train_enr_ITE1.columns]  # filtramos por si faltan
enr_dedup = df_train_enr_ITE1[cols_enr_keep].drop_duplicates()

# Nos quedamos con train_full y le añadimos las columnas enriquecidas por clave
train_full_enr = train_full.merge(enr_dedup, on=[*PAIR_COLS, DATECOL, VALUE_TRAINTEST], how="left")

# 3.8) Verificación de longitudes por pareja
verif = (train_full.groupby(PAIR_COLS)[DATECOL].nunique().rename("n_meses_train").reset_index()
         .merge(test_full.groupby(PAIR_COLS)[DATECOL].nunique().rename("n_meses_test").reset_index(),
                on=PAIR_COLS, how="outer").fillna(0))
verif["n_meses_train"] = verif["n_meses_train"].astype(int)
verif["n_meses_test"]  = verif["n_meses_test"].astype(int)
verif["ok_train_36"]   = (verif["n_meses_train"]==36)
verif["ok_test_12"]    = (verif["n_meses_test"]==12)

save_csv(verif, os.path.join(ruta_base_3, "METRICAS", "verificacion_longitudes_train_test.csv"))
log(f"Parejas con train != 36: {(~verif['ok_train_36']).sum()} | con test != 12: {(~verif['ok_test_12']).sum()}")

# 3.9) Punto de control A: conteos y listas de pares
pairs_train = set(map(tuple, train_full[PAIR_COLS].drop_duplicates().values.tolist()))
pairs_test  = set(map(tuple, test_full[PAIR_COLS].drop_duplicates().values.tolist()))
only_test   = sorted(list(pairs_test - pairs_train))
only_train  = sorted(list(pairs_train - pairs_test))

save_csv(pd.DataFrame(only_test,  columns=PAIR_COLS), os.path.join(ruta_base_3, "METRICAS", "pairs_solo_test.csv"))
save_csv(pd.DataFrame(only_train, columns=PAIR_COLS), os.path.join(ruta_base_3, "METRICAS", "pairs_solo_train.csv"))

print("=== Punto de control A ===")
print("Parejas en train:", len(pairs_train))
print("Parejas en test :", len(pairs_test))
print("Parejas SOLO TEST (fallback SNaive12, sin histórico, cobertura=0):", len(only_test))
print("Parejas SOLO TRAIN:", len(only_train))

# 3.10) Marcado de SOLO TEST como sin histórico para trazabilidad de modelo_final
if only_test:
    df_only_test = pd.DataFrame(only_test, columns=PAIR_COLS)
    df_only_test["modelo_final"] = "fallback_snaive12"
    df_only_test["cobertura"]    = 0
    save_csv(df_only_test, os.path.join(ruta_base_3, "METRICAS", "solo_test_modelo_final.csv"))

# 3.11) Guardamos salidas canónicas
save_csv(train_full,      os.path.join(ruta_base_3, "RESULTADOS", "train_full_2021_2023.csv"))
save_csv(train_full_enr,  os.path.join(ruta_base_3, "RESULTADOS", "train_full_enr_2021_2023.csv"))
save_csv(test_full,       os.path.join(ruta_base_3, "RESULTADOS", "test_full_2024.csv"))
log("Guardados canónicos generados.")

# 3.12) Vista preliminar (diagnóstico visual mínimo)
print("\n== df_fd1_v5_ITE1_train.head() ==")
display(df_fd1_v5_ITE1_train.head(5))
print("\n== df_fd1_v5_ITE1_test.head() ==")
display(df_fd1_v5_ITE1_test.head(5))
print("\n== df_train_enr_ITE1.head() ==")
display(df_train_enr_ITE1.head(5))

# Confirmamos columnas de outliers presentes en el enriquecido
present_outlier_cols = [c for c in [VALUE_ENR_MOD2, VALUE_ENR_MOD3] if c in df_train_enr_ITE1.columns]
print("\nColumnas de outliers presentes en df_train_enr_ITE1:", present_outlier_cols)

log("Punto 0 finalizado.")

Iniciamos el Punto 0 — Preparación.
Ficheros cargados desde Estrategia 2.
Parejas con train != 36: 461 | con test != 12: 198
=== Punto de control A ===
Parejas en train: 2429
Parejas en test : 2692
Parejas SOLO TEST (fallback SNaive12, sin histórico, cobertura=0): 461
Parejas SOLO TRAIN: 198
Guardados canónicos generados.

== df_fd1_v5_ITE1_train.head() ==

== df_fd1_v5_ITE1_test.head() ==

== df_train_enr_ITE1.head() ==

Columnas de outliers presentes en df_train_enr_ITE1: ['cost_float_mod2', 'cost_float_mod3']
Punto 0 finalizado.

# ===========================
# Paso 1 — Parámetros y rutas
# ===========================
# Asumimos que ya hemos ejecutado el Paso 0 y existen:
# - ruta_base_3, SUBFOLDERS, CSV_SEP
# - PAIR_COLS, DATECOL, VALUE_TRAINTEST
# - train_full (2021–2023) y test_full (2024)
# - funciones ensure_dirs() y log()


# Aseguramos carpetas por si se invoca el Paso 1 de forma aislada
ensure_dirs(ruta_base_3, SUBFOLDERS)

ruta_metricas = os.path.join(ruta_base_3, "METRICAS")
out_csv = os.path.join(ruta_metricas, "metrics_baselines_cv.csv")
cols_csv = ["ID_BUILDING","FM_COST_TYPE","baseline","n_obs","MAE","WAPE","SMAPE","MASE1","MASE12"]

# Creamos CSV con cabecera si no existe
if not os.path.exists(out_csv):
    pd.DataFrame(columns=cols_csv).to_csv(out_csv, sep=CSV_SEP, index=False)
    log(f"Inicializamos {out_csv} con cabecera.")

# Verificamos disponibilidad de dataframes canónicos del Paso 0
assert "train_full" in globals(), "Falta train_full (generado en el Paso 0)."
assert "test_full"  in globals(), "Falta test_full (generado en el Paso 0)."

# Ordenamos y tipamos por seguridad
train_df = train_full.sort_values([*PAIR_COLS, DATECOL]).reset_index(drop=True).copy()
test_df  = test_full.sort_values([*PAIR_COLS, DATECOL]).reset_index(drop=True).copy()

# ==========================================
# Paso 1 — Funciones auxiliares y métricas
# ==========================================
# Implementamos utilitarias robustas siguiendo el estándar del proyecto.

def _safe_div(num, den, eps=1e-8):
    # Implementamos división segura para evitar divisiones por cero.
    return num / (den + eps)

def _mae(y_true, y_pred):
    # Calculamos el MAE de forma directa.
    y_true = np.asarray(y_true, dtype=float)
    y_pred = np.asarray(y_pred, dtype=float)
    return np.mean(np.abs(y_pred - y_true))

def _wape_safe(y_true, y_pred, eps=1e-8):
    # Calculamos el WAPE con denominador seguro sum(|y_true|)+eps.
    y_true = np.asarray(y_true, dtype=float)
    y_pred = np.asarray(y_pred, dtype=float)
    num = np.sum(np.abs(y_pred - y_true))
    den = np.sum(np.abs(y_true))
    return _safe_div(num, den, eps)

def _smape_safe(y_true, y_pred, eps=1e-8):
    # Calculamos el SMAPE con denominador simétrico y eps para estabilidad.
    y_true = np.asarray(y_true, dtype=float)
    y_pred = np.asarray(y_pred, dtype=float)
    num = np.abs(y_pred - y_true)
    den = (np.abs(y_true) + np.abs(y_pred)) / 2.0
    return np.mean(_safe_div(num, den, eps)) * 100.0

def _scale_diff(series, lag):
    # Calculamos la escala como media de |y_t - y_{t-lag}| sobre train.
    arr = np.asarray(series, dtype=float)
    if len(arr) <= lag:
        return np.nan
    diffs = np.abs(arr[lag:] - arr[:-lag])
    if len(diffs) == 0:
        return np.nan
    return np.mean(diffs)

def mase_from_scale(errors, scale):
    # Calculamos el MASE a partir de un vector de errores y una escala.
    if scale is None or np.isnan(scale) or scale == 0:
        return np.nan
    errors = np.asarray(errors, dtype=float)
    return np.mean(np.abs(errors)) / float(scale)

def _eval_series(y_true, y_pred, escalas_train: dict):
    # Calculamos las métricas comunes con las escalas proporcionadas del train.
    y_true = np.asarray(y_true, dtype=float)
    y_pred = np.asarray(y_pred, dtype=float)
    err = y_pred - y_true
    mae   = _mae(y_true, y_pred)
    wape  = _wape_safe(y_true, y_pred)
    smape = _smape_safe(y_true, y_pred)
    mase1  = mase_from_scale(err,  escalas_train.get("scale1"))
    mase12 = mase_from_scale(err, escalas_train.get("scale12"))
    return {"MAE": mae, "WAPE": wape, "SMAPE": smape, "MASE1": mase1, "MASE12": mase12}

# ==========================================
# Paso 1 — Baselines por pareja y guardado
# ==========================================
# Generamos Naive(1), SNaive(12) y Mediana estacional, evaluamos y persistimos.

# Construimos sets para saber qué pares tienen test y train
pairs_train = set(map(tuple, train_df[PAIR_COLS].drop_duplicates().values.tolist()))
pairs_test  = set(map(tuple,  test_df[PAIR_COLS].drop_duplicates().values.tolist()))
pairs_commons = sorted(list(pairs_train & pairs_test))
pairs_only_test = sorted(list(pairs_test - pairs_train))

if pairs_only_test:
    log(f"Advertencia: {len(pairs_only_test)} parejas SOLO TEST, se omiten en baselines por falta de histórico.")

rows_out = []

for bid, ctype in pairs_commons:
    # Filtramos serie train y test
    tr = train_df[(train_df[PAIR_COLS[0]]==bid) & (train_df[PAIR_COLS[1]]==ctype)].copy()
    te = test_df [(test_df [PAIR_COLS[0]]==bid) & (test_df [PAIR_COLS[1]]==ctype)].copy()
    if te.empty or tr.empty:
        continue  # nos aseguramos de tener ambos lados

    # Alineamos por fecha y extraemos vectores
    tr = tr.sort_values(DATECOL)
    te = te.sort_values(DATECOL)
    y_train = tr[VALUE_TRAINTEST].astype(float).values
    y_true  = te[VALUE_TRAINTEST].astype(float).values
    n_obs   = len(y_true)

    # Calculamos escalas MASE sobre train
    scale1  = _scale_diff(y_train, lag=1)
    scale12 = _scale_diff(y_train, lag=12)
    escalas = {"scale1": scale1, "scale12": scale12}

    # Baseline Naive(1): último valor del train replicado
    last_train = y_train[-1]
    y_pred_naive1 = np.repeat(last_train, n_obs)
    m = _eval_series(y_true, y_pred_naive1, escalas)
    rows_out.append({
        "ID_BUILDING": bid, "FM_COST_TYPE": ctype,
        "baseline": "Naive1", "n_obs": n_obs, **m
    })

    # Baseline SNaive(12): mismo mes del año anterior; anclamos 2024 contra 2023
    tr["MONTH_NUM"] = pd.to_datetime(tr[DATECOL]).dt.month
    te["MONTH_NUM"] = pd.to_datetime(te[DATECOL]).dt.month
    tr_2023 = tr[pd.to_datetime(tr[DATECOL]).dt.year == 2023].copy()

    y_pred_snaive12 = np.full(n_obs, np.nan, dtype=float)
    if not tr_2023.empty:
        map_2023 = tr_2023.set_index("MONTH_NUM")[VALUE_TRAINTEST].to_dict()
        for i, mth in enumerate(te["MONTH_NUM"].values):
            if mth in map_2023:
                y_pred_snaive12[i] = map_2023[mth]

    # Fallback: mediana mensual de 2021–2023 si faltan meses de 2023
    if np.isnan(y_pred_snaive12).any():
        mediana_mensual = tr.groupby("MONTH_NUM")[VALUE_TRAINTEST].median().to_dict()
        for i, mth in enumerate(te["MONTH_NUM"].values):
            if np.isnan(y_pred_snaive12[i]) and mth in mediana_mensual:
                y_pred_snaive12[i] = mediana_mensual[mth]

    # Último fallback: último valor de train si aún quedan NaN
    if np.isnan(y_pred_snaive12).any():
        y_pred_snaive12 = np.where(np.isnan(y_pred_snaive12), last_train, y_pred_snaive12)

    m = _eval_series(y_true, y_pred_snaive12, escalas)
    rows_out.append({
        "ID_BUILDING": bid, "FM_COST_TYPE": ctype,
        "baseline": "SNaive12", "n_obs": n_obs, **m
    })

    # Baseline Mediana estacional por mes (robusto a outliers)
    mediana_mensual = tr.groupby("MONTH_NUM")[VALUE_TRAINTEST].median()
    y_pred_med = te["MONTH_NUM"].map(mediana_mensual).astype(float).values
    if np.isnan(y_pred_med).any():
        y_pred_med = np.where(np.isnan(y_pred_med), last_train, y_pred_med)

    m = _eval_series(y_true, y_pred_med, escalas)
    rows_out.append({
        "ID_BUILDING": bid, "FM_COST_TYPE": ctype,
        "baseline": "MedianaEstacional", "n_obs": n_obs, **m
    })

# Persistimos resultados en modo append
if rows_out:
    df_out = pd.DataFrame(rows_out)[cols_csv]
    df_out.to_csv(out_csv, sep=CSV_SEP, index=False, mode="a", header=False)
    log(f"Guardadas {len(rows_out)} filas de métricas de baselines en {out_csv}.")
else:
    log("No se generaron filas de métricas (revisar disponibilidad de pares comunes).")

Advertencia: 461 parejas SOLO TEST, se omiten en baselines por falta de histórico.
Guardadas 6693 filas de métricas de baselines en /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3/METRICAS/metrics_baselines_cv.csv.

# ==========================================
# Paso 1 — Punto de control B: MASE12 ≈ 1
# ==========================================
# Validamos que SNaive(12) tenga MASE12 cercano a 1 en series con suficiente histórico.

# Construimos resumen de disponibilidad
hist = (train_df.groupby(PAIR_COLS)
        .agg(n_train=(DATECOL, "nunique"))
        .reset_index())
has_test = (test_df.groupby(PAIR_COLS)
            .agg(n_test=(DATECOL, "nunique"))
            .reset_index())
panel = hist.merge(has_test, on=PAIR_COLS, how="inner")
candidatos = panel[(panel["n_train"] >= 24) & (panel["n_test"] >= 12)]

# Cargamos métricas y filtramos SNaive12 sobre las parejas candidatas
check_df = pd.read_csv(out_csv, sep=CSV_SEP)
check_df = check_df[check_df["baseline"] == "SNaive12"]

if not candidatos.empty and not check_df.empty:
    mask = pd.merge(check_df[PAIR_COLS].drop_duplicates(), candidatos[PAIR_COLS], on=PAIR_COLS, how="inner")
    check_pairs = pd.merge(check_df, mask, on=PAIR_COLS, how="inner")
else:
    check_pairs = pd.DataFrame(columns=cols_csv)

fails = []
for _, r in check_pairs.iterrows():
    m12 = r["MASE12"]
    if not pd.isna(m12):
        try:
            assert abs(m12 - 1.0) < 0.10
        except AssertionError:
            fails.append((r["ID_BUILDING"], r["FM_COST_TYPE"], m12))

print(f"Series verificadas (SNaive12 con MASE12≈1): {len(check_pairs) - len(fails)} / {len(check_pairs)}")
if fails:
    print("Parejas fuera de tolerancia |MASE12-1|<0.10 (mostramos hasta 10):")
    for t in fails[:10]:
        print("  -", t)

# Si el número de fallos es alto, revisaremos que:
# 1) La alineación 2024-mes vs 2023-mes sea correcta.
# 2) La escala12 se haya calculado con suficientes lags (>= 13 puntos en train).
# 3) No estemos ante series sin estacionalidad clara o con outliers que alteren el patrón anual.

Series verificadas (SNaive12 con MASE12≈1): 242 / 2231
Parejas fuera de tolerancia |MASE12-1|<0.10 (mostramos hasta 10):
  - (2, 'Licencias', 0.0675809646949442)
  - (2, 'Mtto. Contratos', 2.0247738210077566)
  - (2, 'Mtto. Correctivo', 0.54078965972694)
  - (2, 'Obras', 0.0095601541180889)
  - (2, 'Servicios Ctto.', 13.675578361454166)
  - (2, 'Servicios Extra', 0.6451420029895367)
  - (2, 'Suministros', 0.6910076996515223)
  - (9, 'Licencias', 1.7526171050032666)
  - (9, 'Mtto. Contratos', 0.8979701546545737)
  - (9, 'Mtto. Correctivo', 1.4839386748570396)

# ==========================================
# Paso 1 — Bloque 5) Resumen agregado por portfolio
# ==========================================
# En este bloque realizamos resúmenes globales y por baseline, e identificamos el
# baseline "ganador" por pareja usando como métrica principal MASE12 (más próxima a 0 mejor).
# Guardamos varios CSVs de salida en METRICAS para análisis y reporting.


ruta_metricas = os.path.join(ruta_base_3, "METRICAS")
in_csv  = os.path.join(ruta_metricas, "metrics_baselines_cv.csv")
assert os.path.exists(in_csv), f"No existe {in_csv}. Debemos ejecutar antes los bloques de baselines."

dfm = pd.read_csv(in_csv, sep=CSV_SEP)

# Nos aseguramos de tipos numéricos
metric_cols = ["MAE","WAPE","SMAPE","MASE1","MASE12"]
for c in metric_cols + ["n_obs"]:
    dfm[c] = pd.to_numeric(dfm[c], errors="coerce")

# 5.1) Resumen global por baseline (media y mediana)
summary_by_baseline = (
    dfm.groupby("baseline")
       .agg(
           n_pairs=("ID_BUILDING","count"),
           MAE_mean=("MAE","mean"),       MAE_median=("MAE","median"),
           WAPE_mean=("WAPE","mean"),     WAPE_median=("WAPE","median"),
           SMAPE_mean=("SMAPE","mean"),   SMAPE_median=("SMAPE","median"),
           MASE1_mean=("MASE1","mean"),   MASE1_median=("MASE1","median"),
           MASE12_mean=("MASE12","mean"), MASE12_median=("MASE12","median")
       )
       .reset_index()
       .sort_values("MASE12_mean")
)

# 5.2) Baseline “ganador” por pareja según MASE12 (mínimo)
# Empatamos usando orden de preferencia: SNaive12 > MedianaEstacional > Naive1
prefer = {"SNaive12": 0, "MedianaEstacional": 1, "Naive1": 2}
dfm["_pref"] = dfm["baseline"].map(prefer).fillna(9)

df_sorted = dfm.sort_values(["ID_BUILDING","FM_COST_TYPE","MASE12","_pref"], ascending=[True,True,True,True])
best_by_pair = df_sorted.groupby(PAIR_COLS, as_index=False).first()[
    [*PAIR_COLS, "baseline", "MASE12", "MAE", "WAPE", "SMAPE", "MASE1"]
].rename(columns={"baseline":"baseline_winner", "MASE12":"MASE12_winner"})

# 5.3) Cuota de victorias por baseline
rankshare = (
    best_by_pair["baseline_winner"]
    .value_counts(dropna=False)
    .rename_axis("baseline_winner")
    .reset_index(name="wins")
    .assign(share=lambda d: d["wins"] / d["wins"].sum())
    .sort_values("wins", ascending=False)
)

# 5.4) Resumen global (todas las filas)
overall_summary = (
    dfm.agg({
        "MAE":"mean", "WAPE":"mean", "SMAPE":"mean", "MASE1":"mean", "MASE12":"mean"
    }).to_frame(name="mean").T
)
overall_summary["n_rows"]  = len(dfm)
overall_summary["n_pairs"] = dfm.groupby(PAIR_COLS).ngroups

# 5.5) Guardamos salidas
p1 = os.path.join(ruta_metricas, "metrics_baselines_summary_by_baseline.csv")
p2 = os.path.join(ruta_metricas, "metrics_baselines_best_by_pair.csv")
p3 = os.path.join(ruta_metricas, "metrics_baselines_rankshare.csv")
p4 = os.path.join(ruta_metricas, "metrics_baselines_overall_summary.csv")

summary_by_baseline.to_csv(p1, sep=CSV_SEP, index=False)
best_by_pair.to_csv(p2, sep=CSV_SEP, index=False)
rankshare.to_csv(p3, sep=CSV_SEP, index=False)
overall_summary.to_csv(p4, sep=CSV_SEP, index=False)

log(f"Resumen por baseline guardado en: {p1}")
log(f"Ganador por pareja guardado en: {p2}")
log(f"Cuota de victorias por baseline guardado en: {p3}")
log(f"Resumen global guardado en: {p4}")

# 5.6) Impresión rápida para revisión
print("\n== Resumen global (medias) ==")
display(overall_summary)

print("\n== Resumen por baseline (ordenado por MASE12_mean ascendente) ==")
display(summary_by_baseline)

print("\n== Cuota de victorias por baseline (conteo y proporción) ==")
display(rankshare.head(10))

# - Revisaremos qué baseline presenta menor MASE12_mean en el resumen por baseline.
# - Observaremos en rankshare qué baseline gana más parejas (baseline_winner).
# - Si SNaive12 domina en victorias y en MASE12_mean, entenderemos que existe estacionalidad anual fuerte.
# - Si MedianaEstacional gana en series intermitentes o ruidosas, lo tendremos en cuenta para los fallbacks.
# - Si Naive1 ganara en muchos casos, revisaremos tendencia/estacionalidad ausente o problemas de datos.

Resumen por baseline guardado en: /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3/METRICAS/metrics_baselines_summary_by_baseline.csv
Ganador por pareja guardado en: /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3/METRICAS/metrics_baselines_best_by_pair.csv
Cuota de victorias por baseline guardado en: /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3/METRICAS/metrics_baselines_rankshare.csv
Resumen global guardado en: /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3/METRICAS/metrics_baselines_overall_summary.csv

== Resumen global (medias) ==

== Resumen por baseline (ordenado por MASE12_mean ascendente) ==

== Cuota de victorias por baseline (conteo y proporción) ==

# ============================================
# Paso 2 — Segmentación por intermitencia
# Bloque 1) Parámetros, rutas y precondiciones
# ============================================

# Asumimos que vienen de pasos previos:
# - ruta_base_3, SUBFOLDERS, CSV_SEP
# - PAIR_COLS, DATECOL, VALUE_TRAINTEST
# - train_full (2021–2023, malla mensual con huecos a 0) y test_full (2024)
# - funciones ensure_dirs() y log()

# 1.1) Aseguramos subcarpetas y log dedicado del paso
ensure_dirs(ruta_base_3, SUBFOLDERS)
ruta_metricas = os.path.join(ruta_base_3, "METRICAS")
ruta_logs     = os.path.join(ruta_base_3, "LOGS")

log_path_step2 = os.path.join(ruta_logs, "estrategia3_step2.log")
if not os.path.exists(log_path_step2):
    with open(log_path_step2, "w", encoding="utf-8") as f:
        f.write("== LOG Estrategia 3 - Paso 2 ==\n")

def log2(msg: str):
    # Centralizamos el log específico del Paso 2 y también imprimimos en consola
    print(msg)
    with open(log_path_step2, "a", encoding="utf-8") as f:
        f.write(f"[{datetime.now().isoformat(timespec='seconds')}] {msg}\n")

log2("Iniciamos Paso 2 — Segmentación por intermitencia.")

# 1.2) Validamos que los dataframes canónicos están disponibles
assert "train_full" in globals(), "Falta train_full (generado en el Paso 0)."
assert "test_full"  in globals(), "Falta test_full (generado en el Paso 0)."

# 1.3) Ordenamos copias de trabajo sin alterar originales
train_df = train_full.sort_values([*PAIR_COLS, DATECOL]).reset_index(drop=True).copy()
test_df  = test_full .sort_values([*PAIR_COLS, DATECOL]).reset_index(drop=True).copy()

Iniciamos Paso 2 — Segmentación por intermitencia.

# ===================================================
# Paso 2 — Bloque 2) Funciones auxiliares de KPIs
# ===================================================

def _rachas_de_ceros(y: np.ndarray):
    # Detectamos rachas consecutivas de ceros en un vector 1D
    # y devolvemos las longitudes de cada racha.
    y = np.asarray(y, dtype=float)
    is_zero = (y == 0)
    if is_zero.size == 0:
        return []
    # Identificamos inicios y finales de rachas
    dif = np.diff(is_zero.astype(int), prepend=0, append=0)
    starts = np.where(dif == 1)[0]
    ends   = np.where(dif == -1)[0]
    return list((ends - starts).astype(int))

def _coef_var_no_cero(y: np.ndarray):
    # Calculamos el coeficiente de variación solo sobre valores > 0.
    # Si hay menos de 2 valores no cero, devolvemos NaN.
    y = np.asarray(y, dtype=float)
    nz = y[y > 0]
    if nz.size < 2:
        return np.nan
    mu = nz.mean()
    if mu == 0:
        return np.nan
    return float(nz.std(ddof=1) / mu)

def _kpis_intermitencia(tr: pd.DataFrame, datecol: str, valcol: str):
    # Calculamos los KPIs solicitados sobre un sub-dataframe de una pareja.
    y = tr[valcol].astype(float).values
    n_meses = len(y)
    n_ceros = int((y == 0).sum())
    prop_zeros = n_ceros / n_meses if n_meses > 0 else np.nan

    rachas = _rachas_de_ceros(y)
    max_racha_0 = int(max(rachas)) if len(rachas) > 0 else 0
    adi_simple  = float(np.mean(rachas)) if len(rachas) > 0 else 0.0

    cv_y = _coef_var_no_cero(y)
    return {
        "n_meses_train": n_meses,
        "n_ceros": n_ceros,
        "prop_zeros": prop_zeros,
        "max_racha_0": max_racha_0,
        "adi_simple": adi_simple,
        "cv_y": cv_y
    }

# ===================================================
# Paso 2 — Bloque 3) Cálculo de KPIs y segmentación
# ===================================================

# 3.1) Panel de parejas en train
pairs_train = train_df[PAIR_COLS].drop_duplicates().values.tolist()

# 3.2) Calculamos KPIs por pareja
rows = []
for bid, ctype in pairs_train:
    sub = train_df[(train_df[PAIR_COLS[0]] == bid) & (train_df[PAIR_COLS[1]] == ctype)].copy()
    sub = sub.sort_values(DATECOL)
    k = _kpis_intermitencia(sub, DATECOL, VALUE_TRAINTEST)

    # 3.3) Segmentamos con reglas pragmáticas
    # - Si n_meses_train < 12, forzamos INTERMITENTE por cautela y lo dejamos registrado.
    low_history = int(k["n_meses_train"] < 12)
    if low_history:
        segment = "INTERMITENTE"
    else:
        segment = "INTERMITENTE" if (k["prop_zeros"] >= 0.35 or k["max_racha_0"] >= 4) else "GENERAL"

    rows.append({
        "ID_BUILDING": bid,
        "FM_COST_TYPE": ctype,
        **k,
        "segment": segment,
        "low_history": low_history
    })

seg_df = pd.DataFrame(rows)

# 3.4) Guardamos el CSV de segmentación
path_seg = os.path.join(ruta_metricas, "segmentation_intermitencia.csv")
seg_df.to_csv(path_seg, sep=CSV_SEP, index=False)
log2(f"Guardado: {path_seg}")

# 3.5) Construimos listas de trabajo
pairs_intermittentes = seg_df.loc[seg_df["segment"] == "INTERMITENTE", PAIR_COLS].drop_duplicates().copy()
pairs_generales     = seg_df.loc[seg_df["segment"] == "GENERAL", PAIR_COLS].drop_duplicates().copy()

# 3.6) Reutilizamos pairs_solo_test del Paso 0 si existe; si no, lo recomputamos
try:
    # Intentamos detectar si ya existe un CSV del Paso 0
    path_solo_test = os.path.join(ruta_metricas, "pairs_solo_test.csv")
    if os.path.exists(path_solo_test):
        pairs_solo_test = pd.read_csv(path_solo_test, sep=CSV_SEP)
    else:
        # Recomputamos a partir de train_df y test_df
        s_train = set(map(tuple, train_df[PAIR_COLS].drop_duplicates().values.tolist()))
        s_test  = set(map(tuple,  test_df [PAIR_COLS].drop_duplicates().values.tolist()))
        only_test = sorted(list(s_test - s_train))
        pairs_solo_test = pd.DataFrame(only_test, columns=PAIR_COLS)
        pairs_solo_test.to_csv(path_solo_test, sep=CSV_SEP, index=False)
        log2(f"Guardado (recomputado): {path_solo_test}")
except Exception as e:
    log2(f"AVISO: no fue posible cargar/recomputar pairs_solo_test ({e}).")
    pairs_solo_test = pd.DataFrame(columns=PAIR_COLS)

# 3.7) Guardamos listas de trabajo
path_interm = os.path.join(ruta_metricas, "pairs_intermittentes.csv")
path_gener  = os.path.join(ruta_metricas, "pairs_generales.csv")

pairs_intermittentes.to_csv(path_interm, sep=CSV_SEP, index=False)
pairs_generales.to_csv(path_gener, sep=CSV_SEP, index=False)

log2(f"Guardado: {path_interm}")
log2(f"Guardado: {path_gener}")

Guardado: /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3/METRICAS/segmentation_intermitencia.csv
Guardado: /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3/METRICAS/pairs_intermittentes.csv
Guardado: /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3/METRICAS/pairs_generales.csv

# ======================================================
# Paso 2 — Bloque 4) Punto de control C y diagnósticos
# ======================================================

print("== Paso 2 — Segmentación por intermitencia ==")

total_train_pairs = seg_df[PAIR_COLS].drop_duplicates().shape[0]
n_intermittentes  = pairs_intermittentes.shape[0]
n_generales       = pairs_generales.shape[0]
n_solo_test       = pairs_solo_test.shape[0] if isinstance(pairs_solo_test, pd.DataFrame) else 0

print(f"Total_train_pairs: {total_train_pairs}")
print(f"n_intermittentes : {n_intermittentes}")
print(f"n_generales      : {n_generales}")
print(f"n_solo_test      : {n_solo_test}   # informativo (no suma con train)")

log2(f"Total_train_pairs: {total_train_pairs}")
log2(f"n_intermittentes : {n_intermittentes}")
log2(f"n_generales      : {n_generales}")
log2(f"n_solo_test      : {n_solo_test}")

# 4.1) Verificación identidad
assert (n_intermittentes + n_generales) == total_train_pairs, \
    "La suma de intermitentes y generales no cuadra con el total del train."

print("Comprobación -> n_intermittentes + n_generales == Total_train_pairs: OK")
log2("Comprobación -> n_intermittentes + n_generales == Total_train_pairs: OK")

# 4.2) Tops de casos límite
top_prop = (seg_df
            .sort_values("prop_zeros", ascending=False)
            [PAIR_COLS + ["prop_zeros","max_racha_0"]]
            .head(5))
top_racha = (seg_df
             .sort_values("max_racha_0", ascending=False)
             [PAIR_COLS + ["max_racha_0","prop_zeros"]]
             .head(5))

print("-- Top 5 por prop_zeros --")
for _, r in top_prop.iterrows():
    print((r["ID_BUILDING"], r["FM_COST_TYPE"], float(r["prop_zeros"]), int(r["max_racha_0"])))

print("-- Top 5 por max_racha_0 --")
for _, r in top_racha.iterrows():
    print((r["ID_BUILDING"], r["FM_COST_TYPE"], int(r["max_racha_0"]), float(r["prop_zeros"])))

# 4.3) Mensajes finales
print(f"Guardado: {path_seg}")
print(f"Guardado: {path_interm}")
print(f"Guardado: {path_gener}")

log2("Paso 2 finalizado.")

== Paso 2 — Segmentación por intermitencia ==
Total_train_pairs: 2429
n_intermittentes : 1397
n_generales      : 1032
n_solo_test      : 461   # informativo (no suma con train)
Total_train_pairs: 2429
n_intermittentes : 1397
n_generales      : 1032
n_solo_test      : 461
Comprobación -> n_intermittentes + n_generales == Total_train_pairs: OK
Comprobación -> n_intermittentes + n_generales == Total_train_pairs: OK
-- Top 5 por prop_zeros --
(9, 'Eficiencia Energética', 1.0, 36)
(18, 'Eficiencia Energética', 1.0, 36)
(2, 'Eficiencia Energética', 1.0, 36)
(1060, 'Servicios Extra', 1.0, 36)
(1065, 'Eficiencia Energética', 1.0, 36)
-- Top 5 por max_racha_0 --
(1306, 'Obras', 36, 1.0)
(62, 'Licencias', 36, 1.0)
(9, 'Eficiencia Energética', 36, 1.0)
(2, 'Eficiencia Energética', 36, 1.0)
(18, 'Eficiencia Energética', 36, 1.0)
Guardado: /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3/METRICAS/segmentation_intermitencia.csv
Guardado: /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3/METRICAS/pairs_intermittentes.csv
Guardado: /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3/METRICAS/pairs_generales.csv
Paso 2 finalizado.

# ============================================================
# Paso 3 — Conjunto de modelos candidato
# Bloque 1) Parámetros, rutas, artefactos y logging
# ============================================================

# Definimos valores por defecto por si no vienen del Paso 2
# Explicación: preferimos ser robustos a la hora de ejecutar el notebook de forma independiente.
ruta_base_3 = globals().get("ruta_base_3", "/content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3")
CSV_SEP     = globals().get("CSV_SEP", ";")
SUBFOLDERS  = globals().get("SUBFOLDERS", ["METRICAS", "LOGS", "CONFIG"])

# Rutas de trabajo
ruta_metricas = os.path.join(ruta_base_3, "METRICAS")
ruta_logs     = os.path.join(ruta_base_3, "LOGS")
ruta_config   = os.path.join(ruta_base_3, "CONFIG")

# Aseguramos subcarpetas de salida
def ensure_dirs(base_path: str, subfolders: list):
    # Creamos las carpetas de salida para mantener la trazabilidad del paso
    os.makedirs(base_path, exist_ok=True)
    for sf in subfolders:
        os.makedirs(os.path.join(base_path, sf), exist_ok=True)

ensure_dirs(ruta_base_3, SUBFOLDERS)

# Logging del paso 3
log_path_step3 = os.path.join(ruta_logs, "estrategia3_step3.log")
if not os.path.exists(log_path_step3):
    with open(log_path_step3, "w", encoding="utf-8") as f:
        f.write("== LOG Estrategia 3 - Paso 3 ==\n")

def log3(msg: str):
    # Centralizamos el log específico del Paso 3 y también imprimimos en consola
    print(msg)
    with open(log_path_step3, "a", encoding="utf-8") as f:
        f.write(f"[{datetime.now().isoformat(timespec='seconds')}] {msg}\n")

log3("Iniciamos Paso 3 — Conjunto de modelos candidato.")

# Artefactos esperados desde el Paso 2
path_seg   = os.path.join(ruta_metricas, "segmentation_intermitencia.csv")
path_interm= os.path.join(ruta_metricas, "pairs_intermittentes.csv")
path_gener = os.path.join(ruta_metricas, "pairs_generales.csv")

# Configuración de umbrales y métricas de decisión
# Explicación: fijamos los criterios de corte que utilizaremos en el Paso 4 para seleccionar y hacer fallback.
METRIC_PRIMARY   = "SMAPE"        # métrica principal para selección local por pareja
METRICS_TRACK    = ["SMAPE", "WAPE", "MASE12"]  # métricas que vamos a registrar
THRESHOLDS_FAIL  = {"MASE12": 1.20, "SMAPE": 25.0}  # disparadores de fallback
H                = 12  # horizonte mensual estándar

Iniciamos Paso 3 — Conjunto de modelos candidato.

# ============================================================
# Paso 3 — Conjunto de modelos candidato
# Bloque 2) Funciones auxiliares y definición del menú
# ============================================================

# Leemos listas de parejas y segmentación
def load_artifacts():
    # Cargamos los artefactos del Paso 2 y validamos su consistencia mínima
    seg_df = pd.read_csv(path_seg, sep=CSV_SEP)
    pairs_intermittentes = pd.read_csv(path_interm, sep=CSV_SEP)
    pairs_generales      = pd.read_csv(path_gener, sep=CSV_SEP)

    # Validamos columnas esenciales
    req_cols = ["ID_BUILDING", "FM_COST_TYPE"]
    for df_name, df_obj in [("segmentation_intermitencia.csv", seg_df),
                            ("pairs_intermittentes.csv", pairs_intermittentes),
                            ("pairs_generales.csv", pairs_generales)]:
        missing = set(req_cols) - set(df_obj.columns)
        assert not missing, f"Faltan columnas {missing} en {df_name}"

    return seg_df, pairs_intermittentes, pairs_generales

# Identificamos series todo-cero en train para baseline estructural 0
def detect_all_zero_series(seg_df: pd.DataFrame) -> pd.DataFrame:
    # Consideramos max_racha_0 == 36 como proxy de 36 meses seguidos a cero en 2021–2023
    # Guardamos bandera para orquestación posterior
    df = seg_df.copy()
    df["flag_all_zero_train"] = (df.get("max_racha_0", 0).astype(int) >= 36).astype(int)
    return df

# Definimos el menú de modelos por ruta
def build_menu_general():
    # Definimos el menú de candidatos para series generales
    modelos = {
        "candidatos": [
            {"name": "Theta",         "family": "theta", "params": {}},
            {"name": "ETS_auto",      "family": "ets",   "params": {"auto": True, "components": ["A/A", "A/M", "M/M"]}},
            {"name": "ARIMA_auto",    "family": "arima", "params": {"grid": "light"}},
            {"name": "Holt_damped",   "family": "holt",  "params": {"damped": True}},
        ],
        "baselines": [
            {"name": "Naive1",        "family": "baseline", "params": {"type": "naive1"}},
            {"name": "SNaive12",      "family": "baseline", "params": {"type": "snaive12"}},
            {"name": "MedianaEstac",  "family": "baseline", "params": {"type": "median_seasonal", "m": 12}},
        ],
        "rules": {
            "diagnostico_guiado": True,
            "fallback_if": THRESHOLDS_FAIL,   # usamos los umbrales definidos en Bloque 1
            "metric_primary": METRIC_PRIMARY,
            "metrics_track": METRICS_TRACK,
        }
    }
    return modelos

def build_menu_intermittent():
    # Definimos el menú de candidatos para series intermitentes
    modelos = {
        "candidatos": [
            {"name": "Croston", "family": "croston", "params": {"variant": "classic"}},
            {"name": "SBA",     "family": "croston", "params": {"variant": "sba"}},
            {"name": "TSB",     "family": "tsb",     "params": {"alpha_grid": [0.1, 0.3, 0.5]}},
            {"name": "ADIDA_MAPA", "family": "adida_mapa", "params": {
                "agg_level": "Q",    # trimestral
                "model_low_freq": ["ETS_auto", "Theta"],
                "disagg": "perfil_historico"
            }},
        ],
        "baselines": [
            {"name": "SNaive12",      "family": "baseline", "params": {"type": "snaive12"}},
            {"name": "PerfilEscaso",  "family": "baseline", "params": {"type": "perfil_escaso_uniforme", "m": 12}},
        ],
        "rules": {
            "baseline_cero_if_all_zero_train": True,
            "metric_primary": METRIC_PRIMARY,
            "metrics_track": METRICS_TRACK,
        }
    }
    return modelos

# Construimos un diccionario de configuración global del Paso 3
def build_step3_config(seg_df: pd.DataFrame):
    # Creamos un paquete de configuración serializable a JSON para el Paso 4
    config = {
        "horizon": H,
        "metrics": {
            "primary": METRIC_PRIMARY,
            "track": METRICS_TRACK,
            "thresholds_fail": THRESHOLDS_FAIL
        },
        "routes": {
            "GENERAL": build_menu_general(),
            "INTERMITENTE": build_menu_intermittent()
        },
        "governance": {
            "lists_control": {
                "pairs_intermittentes_csv": os.path.relpath(path_interm, ruta_base_3),
                "pairs_generales_csv": os.path.relpath(path_gener, ruta_base_3),
                "segmentation_csv": os.path.relpath(path_seg, ruta_base_3),
            },
            "logs": {
                "step3_log": os.path.relpath(log_path_step3, ruta_base_3)
            }
        },
        "flags": {
            "all_zero_flag_column": "flag_all_zero_train"
        }
    }

    # Añadimos resumen de poblaciones por segmento para trazabilidad
    total_pairs = seg_df[["ID_BUILDING","FM_COST_TYPE"]].drop_duplicates().shape[0]
    n_interm    = seg_df[seg_df["segment"] == "INTERMITENTE"][["ID_BUILDING","FM_COST_TYPE"]].drop_duplicates().shape[0]
    n_general   = seg_df[seg_df["segment"] == "GENERAL"][["ID_BUILDING","FM_COST_TYPE"]].drop_duplicates().shape[0]

    config["population"] = {
        "total_train_pairs": int(total_pairs),
        "n_intermittentes": int(n_interm),
        "n_generales": int(n_general)
    }
    return config

# ============================================================
# Paso 3 — Conjunto de modelos candidato
# Bloque 3) Ejecución y guardado de artefactos de configuración
# ============================================================

# Cargamos artefactos del Paso 2
seg_df, pairs_intermittentes, pairs_generales = load_artifacts()
log3("Artefactos del Paso 2 cargados correctamente.")

# Marcamos series todo-cero en train
seg_df = detect_all_zero_series(seg_df)
flag_count = int(seg_df["flag_all_zero_train"].sum())
log3(f"Series con flag_all_zero_train=1 detectadas: {flag_count}")

# Guardamos la segmentación enriquecida con la bandera
path_seg_enriched = os.path.join(ruta_metricas, "segmentation_intermitencia_step3.csv")
seg_df.to_csv(path_seg_enriched, sep=CSV_SEP, index=False)
log3(f"Guardado: {path_seg_enriched}")

# Construimos la configuración del Paso 3
step3_config = build_step3_config(seg_df)

# Guardamos la configuración como JSON para que el Paso 4 pueda orquestar entrenamientos
path_cfg_json = os.path.join(ruta_config, "step3_menu_modelos_config.json")
with open(path_cfg_json, "w", encoding="utf-8") as f:
    json.dump(step3_config, f, ensure_ascii=False, indent=2)
log3(f"Guardado: {path_cfg_json}")

# Resumen en consola para verificación rápida
print("== Resumen Paso 3 — Menú de modelos candidato ==")
print(f"Horizonte: {step3_config['horizon']}")
print("Métrica primaria:", step3_config["metrics"]["primary"])
print("Métricas a trazar:", step3_config["metrics"]["track"])
print("Umbrales de fallback:", step3_config["metrics"]["thresholds_fail"])

print("\nRutas definidas:")
for route in step3_config["routes"].keys():
    print(f" - {route}")

print("\nListas de control:")
print(step3_config["governance"]["lists_control"])

print("\nPoblación por segmento:")
print(step3_config["population"])

log3("Paso 3 finalizado. Menú de modelos candidato definido y configurado.")

Artefactos del Paso 2 cargados correctamente.
Series con flag_all_zero_train=1 detectadas: 104
Guardado: /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3/METRICAS/segmentation_intermitencia_step3.csv
Guardado: /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3/CONFIG/step3_menu_modelos_config.json
== Resumen Paso 3 — Menú de modelos candidato ==
Horizonte: 12
Métrica primaria: SMAPE
Métricas a trazar: ['SMAPE', 'WAPE', 'MASE12']
Umbrales de fallback: {'MASE12': 1.2, 'SMAPE': 25.0}

Rutas definidas:
 - GENERAL
 - INTERMITENTE

Listas de control:
{'pairs_intermittentes_csv': 'METRICAS/pairs_intermittentes.csv', 'pairs_generales_csv': 'METRICAS/pairs_generales.csv', 'segmentation_csv': 'METRICAS/segmentation_intermitencia.csv'}

Población por segmento:
{'total_train_pairs': 2429, 'n_intermittentes': 1397, 'n_generales': 1032}
Paso 3 finalizado. Menú de modelos candidato definido y configurado.

# ============================================================
# ESTRATEGIA 3 — PASO 4
# Validación “rolling origin” en 2023 y selección top-2 por serie
# ============================================================
# En este paso implementamos:
#  - Una validación rolling con orígenes en 2023-04-01, 2023-08-01 y 2023-11-01.
#  - Evaluación con h=1 (y opcional h=3).
#  - Cálculo de métricas MASE12 (primaria), WAPE y SMAPE.
#  - Agregación por serie-modelo, selección top-2 y comparación contra SNaive12 (2024).
#  - Persistencia de CSVs y logging de diagnóstico (Punto de control D).
# ============================================================

# =========================
# Bloque 1) Parámetros y rutas
# =========================
# Definimos rutas base y archivos esperados de pasos anteriores.
# Si ejecutamos el Paso 4 de forma aislada, tomamos valores por defecto.

# Fijamos semilla global para reproducibilidad de cualquier aleatoriedad interna.
np.random.seed(7)

ruta_base_3 = globals().get("ruta_base_3", "/content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3")
CSV_SEP     = globals().get("CSV_SEP", ";")

RUTA_METRICAS = os.path.join(ruta_base_3, "METRICAS")
RUTA_LOGS     = os.path.join(ruta_base_3, "LOGS")
RUTA_CONFIG   = os.path.join(ruta_base_3, "CONFIG")

os.makedirs(RUTA_METRICAS, exist_ok=True)
os.makedirs(RUTA_LOGS, exist_ok=True)
os.makedirs(RUTA_CONFIG, exist_ok=True)

LOG_PATH_STEP4 = os.path.join(RUTA_LOGS, "estrategia3_step4.log")

# Parámetros de rolling origin y horizontes
DO_H3   = False
ORIGINS = ["2023-04-01", "2023-08-01", "2023-11-01"]  # MS
H_LIST  = [1] + ([3] if DO_H3 else [])

# Columnas canónicas y frecuencia
PAIR_COLS = ["ID_BUILDING", "FM_COST_TYPE"]
DATECOL   = "FECHA"
VALCOL    = "cost_float_mod"
FREQ      = "MS"

# Archivos de entrada obligatorios (pasos previos)
PATH_SEG_ENR = os.path.join(RUTA_METRICAS, "segmentation_intermitencia_step3.csv")
PATH_PAIRS_INT = os.path.join(RUTA_METRICAS, "pairs_intermittentes.csv")
PATH_PAIRS_GEN = os.path.join(RUTA_METRICAS, "pairs_generales.csv")
PATH_CFG_JSON  = os.path.join(RUTA_CONFIG,  "step3_menu_modelos_config.json")
PATH_BASELINES_2024 = os.path.join(RUTA_METRICAS, "metrics_baselines_cv.csv")

# Archivos de salida de este paso
PATH_RAW = os.path.join(RUTA_METRICAS, "rolling_metrics_raw.csv")
PATH_AGG = os.path.join(RUTA_METRICAS, "rolling_metrics_agg.csv")
PATH_SEL = os.path.join(RUTA_METRICAS, "seleccion_modelos_cv.csv")

# =========================
# Bloque 2) Utilidades de IO y logging
# =========================
def log4(msg: str):
    # Registramos mensajes del paso 4 en archivo y consola para trazabilidad.
    stamp = datetime.now().isoformat(timespec="seconds")
    line = f"[{stamp}] {msg}"
    print(line)
    with open(LOG_PATH_STEP4, "a", encoding="utf-8") as f:
        f.write(line + "\n")

def _ensure_dtindex(df: pd.DataFrame, datecol: str):
    # Aseguramos que la columna fecha esté en tipo datetime64 y sea MS.
    df = df.copy()
    df[datecol] = pd.to_datetime(df[datecol]).dt.to_period("M").dt.to_timestamp()
    return df

def _concat_train_test(train_df: pd.DataFrame, test_df: pd.DataFrame) -> pd.DataFrame:
    # Construimos una serie continua 2021–2024 para poder evaluar cruces de frontera (nov->dic->ene).
    df = pd.concat([train_df, test_df], ignore_index=True)
    df = _ensure_dtindex(df, DATECOL)
    df = df.sort_values([*PAIR_COLS, DATECOL]).reset_index(drop=True)
    return df

# =========================
# Bloque 3) Métricas (reusamos criterio del Paso 1)
# =========================
def _safe_div(num, den, eps=1e-8):
    return num / (den + eps)

def _mae(y_true, y_pred):
    y_true = np.asarray(y_true, dtype=float)
    y_pred = np.asarray(y_pred, dtype=float)
    return float(np.mean(np.abs(y_pred - y_true)))

def _wape_safe(y_true, y_pred, eps=1e-8):
    y_true = np.asarray(y_true, dtype=float)
    y_pred = np.asarray(y_pred, dtype=float)
    num = np.sum(np.abs(y_pred - y_true))
    den = np.sum(np.abs(y_true))
    return float(_safe_div(num, den, eps))

def _smape_safe(y_true, y_pred, eps=1e-8):
    y_true = np.asarray(y_true, dtype=float)
    y_pred = np.asarray(y_pred, dtype=float)
    num = np.abs(y_pred - y_true)
    den = (np.abs(y_true) + np.abs(y_pred)) / 2.0
    return float(np.mean(_safe_div(num, den, eps)) * 100.0)

def _scale_diff(series, lag):
    arr = np.asarray(series, dtype=float)
    if len(arr) <= lag:
        return np.nan
    diffs = np.abs(arr[lag:] - arr[:-lag])
    if len(diffs) == 0:
        return np.nan
    return float(np.mean(diffs))

def mase_from_scale(errors, scale):
    if scale is None or np.isnan(scale) or scale == 0:
        return np.nan
    errors = np.asarray(errors, dtype=float)
    return float(np.mean(np.abs(errors)) / float(scale))

def _eval_series(y_true, y_pred, y_train_for_scales):
    # Calculamos métricas con escalas MASE sobre el train de la ventana.
    y_true = np.asarray(y_true, dtype=float)
    y_pred = np.asarray(y_pred, dtype=float)
    err = y_pred - y_true
    scale1  = _scale_diff(y_train_for_scales, lag=1)
    scale12 = _scale_diff(y_train_for_scales, lag=12)
    return {
        "MAE":   _mae(y_true, y_pred),
        "WAPE":  _wape_safe(y_true, y_pred),
        "SMAPE": _smape_safe(y_true, y_pred),
        "MASE1": mase_from_scale(err, scale1),
        "MASE12": mase_from_scale(err, scale12)
    }

# =========================
# Bloque 4) Wrappers de modelos
# =========================
# Definimos una API común: fit_predict(model_spec, y_train, dates_train, horizon, freq="MS")
# Implementamos baselines y modelos frecuentes. Dejamos ARIMA/ETS/Theta protegidos si faltan deps.

def _baseline_naive1(y_train, h):
    # Repetimos el último valor observado.
    return np.repeat(float(y_train[-1]) if len(y_train) else 0.0, h)

def _baseline_snaive12(y_train, dates_train, h):
    # Usamos el valor del mismo mes del año anterior si existe; si no, caemos al último valor.
    if len(y_train) == 0:
        return np.zeros(h, dtype=float)
    # Construimos un mapa mes->valor último año observado
    df = pd.DataFrame({"FECHA": dates_train, "y": y_train})
    df["MONTH_NUM"] = pd.to_datetime(df["FECHA"]).dt.month
    # Tomamos el último año completo disponible en el train
    df["YEAR"] = pd.to_datetime(df["FECHA"]).dt.year
    last_year = df["YEAR"].max()
    map_last_year = (df[df["YEAR"] == last_year].set_index("MONTH_NUM")["y"].to_dict())
    # Generamos meses futuros consecutivos
    last_date = pd.to_datetime(dates_train[-1]).to_period("M").to_timestamp()
    future = pd.date_range(last_date + pd.offsets.MonthBegin(1), periods=h, freq="MS")
    pred = []
    for d in future:
        m = d.month
        if m in map_last_year:
            pred.append(map_last_year[m])
        else:
            pred.append(float(y_train[-1]))
    return np.array(pred, dtype=float)

def _baseline_median_seasonal(y_train, dates_train, h, m=12):
    # Tomamos la mediana por mes sobre el train y proyectamos.
    if len(y_train) == 0:
        return np.zeros(h, dtype=float)
    df = pd.DataFrame({"FECHA": dates_train, "y": y_train})
    df["MONTH_NUM"] = pd.to_datetime(df["FECHA"]).dt.month
    med = df.groupby("MONTH_NUM")["y"].median().to_dict()
    last_date = pd.to_datetime(dates_train[-1]).to_period("M").to_timestamp()
    future = pd.date_range(last_date + pd.offsets.MonthBegin(1), periods=h, freq="MS")
    return np.array([med.get(d.month, float(y_train[-1])) for d in future], dtype=float)

# Croston/SBA/TSB (implementaciones directas sobre demanda intermitente)
def _croston_classic(y, h, alpha=0.1):
    # Implementamos Croston clásico sobre valores no cero y tiempos entre demandas.
    y = np.asarray(y, dtype=float)
    n = len(y)
    if n == 0:
        return np.zeros(h, dtype=float)
    # Extraemos demandas y intervalos
    q = []
    a = []
    for v in y:
        if v > 0:
            q.append(v)
    if len(q) == 0:
        return np.zeros(h, dtype=float)
    # Estimadores recursivos
    z = 0.0
    p = 0.0
    k = 0
    first = True
    for v in y:
        k += 1
        if v > 0:
            if first:
                z = v
                p = k
                first = False
            else:
                z = z + alpha * (v - z)
                p = p + alpha * (k - p)
            k = 0
    # Pronóstico Croston = z/p
    f = (z / p) if p > 0 else 0.0
    return np.repeat(f, h)

def _croston_sba(y, h, alpha=0.1):
    # SBA corrige el sesgo multiplicando por (1 - alpha/2)
    base = _croston_classic(y, h, alpha=alpha)
    return base * (1.0 - alpha / 2.0)

def _tsb(y, h, alpha=0.2, beta=0.1):
    # TSB modela probabilidad de demanda y tamaño por separado.
    y = np.asarray(y, dtype=float)
    n = len(y)
    if n == 0:
        return np.zeros(h, dtype=float)
    demand = (y > 0).astype(int)
    sizes = np.where(y > 0, y, 0.0)

    p = demand[0]
    z = sizes[0] if sizes[0] > 0 else (np.mean(sizes[sizes > 0]) if np.any(sizes > 0) else 0.0)

    for t in range(1, n):
        p = p + alpha * (demand[t] - p)
        if y[t] > 0:
            z = z + beta * (y[t] - z)

    f = p * z
    return np.repeat(float(f), h)

def _theta_predict(y_train, dates_train, h):
    # Implementamos Theta usando statsmodels con period=12 para satisfacer el requisito del modelo.
    if ThetaModel is None or len(y_train) == 0:
        return np.repeat(float(y_train[-1]) if len(y_train) else 0.0, h)
    try:
        y = np.asarray(y_train, dtype=float)
        # Forzamos period=12 para mensualidad; así no dependemos de la freq del índice
        tm = ThetaModel(y, period=12)
        res = tm.fit()
        return np.asarray(res.forecast(h), dtype=float)
    except Exception:
        # Si fallara por cualquier razón, caemos a una baseline robusta
        return _baseline_median_seasonal(y_train, dates_train, h, m=12)

def _ets_auto_predict(y_train, h, seasonal="add"):
    # Probamos una configuración ETS básica si está disponible; si no, caemos a Naive1.
    if ExponentialSmoothing is None or len(y_train) < 3:
        return _baseline_naive1(y_train, h)
    try:
        model = ExponentialSmoothing(
            np.asarray(y_train, dtype=float),
            trend="add",
            seasonal=seasonal,  # "add" o "mul"
            seasonal_periods=12,
            damped_trend=True,
            initialization_method="estimated"
        ).fit(optimized=True)
        return np.asarray(model.forecast(h), dtype=float)
    except Exception:
        return _baseline_naive1(y_train, h)

def _holt_damped_predict(y_train, h):
    # Implementamos Holt amortiguado vía ETS sin estacionalidad (si no, Naive1).
    if ExponentialSmoothing is None or len(y_train) < 3:
        return _baseline_naive1(y_train, h)
    try:
        model = ExponentialSmoothing(
            np.asarray(y_train, dtype=float),
            trend="add",
            seasonal=None,
            damped_trend=True,
            initialization_method="estimated"
        ).fit(optimized=True)
        return np.asarray(model.forecast(h), dtype=float)
    except Exception:
        return _baseline_naive1(y_train, h)

def _arima_auto_light(y_train, h):
    # Hacemos un grid muy ligero sobre (p,d,q) pequeños; si falla, Naive1.
    if SARIMAX is None or len(y_train) < 8:
        return _baseline_naive1(y_train, h)
    y = np.asarray(y_train, dtype=float)
    best_aic = np.inf
    best_forecast = None
    for p in [0,1]:
        for d in [0,1]:
            for q in [0,1]:
                try:
                    mod = SARIMAX(y, order=(p,d,q), seasonal_order=(0,0,0,0), trend="c", enforce_stationarity=False, enforce_invertibility=False)
                    res = mod.fit(disp=False)
                    if res.aic < best_aic:
                        best_aic = res.aic
                        best_forecast = res.forecast(h)
                except Exception:
                    continue
    if best_forecast is None:
        return _baseline_naive1(y_train, h)
    return np.asarray(best_forecast, dtype=float)

def _adida_mapa(y_train, h, low_freq="Q", low_models=("ETS_auto","Theta")):
    # Implementamos un ADIDA/MAPA simple:
    #  1) Agregamos la serie a baja frecuencia (trimestral).
    #  2) Pronosticamos a baja frecuencia con ETS o Theta.
    #  3) Desagregamos por perfil histórico proporcional.
    y = np.asarray(y_train, dtype=float)
    if len(y) < 6:
        return _baseline_median_seasonal(y_train, dates_train=np.arange(len(y)), h=h, m=12)
    # Agregación trimestral
    pad = (3 - (len(y) % 3)) % 3
    y_pad = np.hstack([y, np.zeros(pad)])
    yQ = y_pad.reshape(-1, 3).sum(axis=1)

    # Modelo a baja frecuencia
    if "ETS_auto" in low_models and ExponentialSmoothing is not None and len(yQ) >= 3:
        try:
            mQ = ExponentialSmoothing(yQ, trend="add", seasonal=None, damped_trend=True).fit(optimized=True)
            fq = np.asarray(mQ.forecast(int(np.ceil(h/3))), dtype=float)
        except Exception:
            fq = np.repeat(float(yQ[-1]) if len(yQ) else 0.0, int(np.ceil(h/3)))
    elif "Theta" in low_models and ThetaModel is not None:
        try:
            tm = ThetaModel(yQ).fit()
            fq = np.asarray(tm.forecast(int(np.ceil(h/3))), dtype=float)
        except Exception:
            fq = np.repeat(float(yQ[-1]) if len(yQ) else 0.0, int(np.ceil(h/3)))
    else:
        fq = np.repeat(float(yQ[-1]) if len(yQ) else 0.0, int(np.ceil(h/3)))

    # Perfil histórico mensual proporcional dentro de cada trimestre
    # Calculamos pesos medios por posición (mes1, mes2, mes3) en los trimestres históricos.
    if len(y) >= 6:
        blocks = y_pad.reshape(-1, 3)
        totals = blocks.sum(axis=1, keepdims=True)
        weights = np.divide(blocks, np.where(totals==0, 1.0, totals))
        w = np.nan_to_num(weights.mean(axis=0))
        if w.sum() == 0:
            w = np.array([1/3, 1/3, 1/3], dtype=float)
    else:
        w = np.array([1/3, 1/3, 1/3], dtype=float)

    # Desagregamos cada trimestre pronosticado en 3 meses con los pesos medios
    f_months = []
    for qv in fq:
        f_months.extend(list(w * qv))
    f_months = np.asarray(f_months, dtype=float)[:h]
    return f_months

def fit_predict(model_spec: dict, y_train, dates_train, horizon, freq="MS", exog=None) -> np.ndarray:
    # Enrutamos por familia y devolvemos un np.ndarray de tamaño h.
    fam = model_spec.get("family", "").lower()
    name = model_spec.get("name", "").lower()
    params = model_spec.get("params", {}) or {}

    if fam == "baseline":
        t = (params.get("type") or "").lower()
        if t == "naive1":
            return _baseline_naive1(y_train, horizon)
        if t == "snaive12":
            return _baseline_snaive12(y_train, dates_train, horizon)
        if t == "median_seasonal":
            return _baseline_median_seasonal(y_train, dates_train, horizon, m=params.get("m", 12))
        if t == "perfil_escaso_uniforme":
            return np.repeat(np.mean(np.asarray(y_train, dtype=float)), horizon) if len(y_train) else np.zeros(horizon)
        if t == "cero":
            return np.zeros(horizon, dtype=float)
        return _baseline_naive1(y_train, horizon)

    if fam == "theta":
        return _theta_predict(y_train, dates_train, horizon)

    if fam == "ets":
        comps = params.get("components", ["A/A", "A/M"])
        # Probamos una variante aditiva por simplicidad
        return _ets_auto_predict(y_train, horizon, seasonal="add")

    if fam == "holt":
        return _holt_damped_predict(y_train, horizon)

    if fam == "arima":
        return _arima_auto_light(y_train, horizon)

    if fam == "croston":
        variant = (params.get("variant") or "classic").lower()
        alpha = float(params.get("alpha", 0.1))
        if variant == "sba":
            return _croston_sba(y_train, horizon, alpha=alpha)
        return _croston_classic(y_train, horizon, alpha=alpha)

    if fam == "tsb":
        alpha = float(params.get("alpha", 0.2))
        beta  = float(params.get("beta", 0.1))
        return _tsb(y_train, horizon, alpha=alpha, beta=beta)

    if fam == "adida_mapa":
        return _adida_mapa(y_train, horizon,
                           low_freq=params.get("agg_level", "Q"),
                           low_models=tuple(params.get("model_low_freq", ["ETS_auto","Theta"])))
    # Si no reconocemos la familia, devolvemos Naive1 como red de seguridad.
    return _baseline_naive1(y_train, horizon)

# =========================
# Bloque 5) Carga de artefactos del Paso 2–3 y panel de trabajo
# =========================
# Cargamos segmentación, listas y menú de modelos. Validamos columnas mínimas.

assert os.path.exists(PATH_SEG_ENR), f"Falta {PATH_SEG_ENR}"
assert os.path.exists(PATH_PAIRS_INT), f"Falta {PATH_PAIRS_INT}"
assert os.path.exists(PATH_PAIRS_GEN), f"Falta {PATH_PAIRS_GEN}"
assert os.path.exists(PATH_CFG_JSON),  f"Falta {PATH_CFG_JSON}"
assert "train_full" in globals(), "Falta train_full del Paso 0."
assert "test_full"  in globals(), "Falta test_full del Paso 0."

seg_enr  = pd.read_csv(PATH_SEG_ENR, sep=CSV_SEP)
pairs_int = pd.read_csv(PATH_PAIRS_INT, sep=CSV_SEP)
pairs_gen = pd.read_csv(PATH_PAIRS_GEN, sep=CSV_SEP)

with open(PATH_CFG_JSON, "r", encoding="utf-8") as f:
    step3_cfg = json.load(f)

flag_col = step3_cfg.get("flags", {}).get("all_zero_flag_column", "flag_all_zero_train")

# Aseguramos dataframes canónicos
train_df = _ensure_dtindex(train_full, DATECOL)
test_df  = _ensure_dtindex(test_full, DATECOL)
all_df   = _concat_train_test(train_df, test_df)

# Construimos el panel de pares y rutas
pairs_int["route"] = "INTERMITENTE"
pairs_gen["route"] = "GENERAL"
panel_pairs = pd.concat([pairs_int, pairs_gen], ignore_index=True)
panel_pairs = panel_pairs.merge(seg_enr[PAIR_COLS + [flag_col, "segment"]].drop_duplicates(),
                                on=PAIR_COLS, how="left")
panel_pairs["route"] = panel_pairs["route"].fillna(panel_pairs["segment"])  # por si faltara

# Resumen poblacional para log
n_total = panel_pairs[PAIR_COLS].drop_duplicates().shape[0]
n_int   = (panel_pairs["route"] == "INTERMITENTE").sum()
n_gen   = (panel_pairs["route"] == "GENERAL").sum()
log4(f"Población Paso 4 — total={n_total} | INTERMITENTE={n_int} | GENERAL={n_gen}")

# =========================
# Bloque 6) Construcción de menú por ruta
# =========================
menu_routes = step3_cfg.get("routes", {})

def _get_models_for_route(route: str):
    # Obtenemos candidatos + baselines para una ruta dada.
    cfg = menu_routes.get(route.upper(), {})
    candidatos = cfg.get("candidatos", [])
    baselines  = cfg.get("baselines", [])
    return list(candidatos) + list(baselines)

# =========================
# Bloque 7) Bucle principal — Rolling origin
# =========================
# Generamos la tabla raw con filas (pair, route, model, origin, h, métricas).
raw_rows = []

# Preparamos índice por pareja para extraer series más rápido
all_df = all_df.sort_values([*PAIR_COLS, DATECOL])
grouped = all_df.groupby(PAIR_COLS, sort=False)

for _, row in panel_pairs.iterrows():
    bid, ctype, route = row["ID_BUILDING"], row["FM_COST_TYPE"], str(row["route"]).upper()
    is_all_zero = int(row.get(flag_col, 0)) == 1

    # Extraemos serie completa 2021–2024 de la pareja
    try:
        sub = grouped.get_group((bid, ctype)).sort_values(DATECOL)
    except KeyError:
        # Si no encontramos la pareja en el agregado, continuamos
        continue

    y_all   = sub[VALCOL].astype(float).values
    d_all   = sub[DATECOL].values

    # Si es todo-cero (bandera), registramos baseline_cero y pasamos a la siguiente pareja.
    if is_all_zero:
        for origin in ORIGINS:
            for h in H_LIST:
                raw_rows.append({
                    "ID_BUILDING": bid, "FM_COST_TYPE": ctype,
                    "route": route, "model_name": "baseline_cero",
                    "origin": origin, "horizon": h, "n_obs": 0,
                    "MAE": 0.0, "WAPE": 0.0, "SMAPE": 0.0, "MASE1": 0.0, "MASE12": 0.0
                })
        log4(f"[{bid}|{ctype}] flag_all_zero_train=1 -> baseline_cero registrado.")
        continue

    # Preparamos lista de modelos para la ruta
    models = _get_models_for_route(route)

    # Iteramos orígenes y horizontes
    for origin in ORIGINS:
        origin_ts = pd.to_datetime(origin).to_period("M").to_timestamp()

        # Definimos ventana de train (<= origin-1M) y ventana de evaluación (origin..origin+h-1)
        # Calculamos índices
        mask_train = d_all <= (origin_ts - pd.offsets.MonthBegin(1))
        if not np.any(mask_train):
            continue  # si no hay histórico, saltamos origen

        y_tr = y_all[mask_train]
        d_tr = d_all[mask_train]

        for h in H_LIST:
            # Fechas futuras a evaluar
            future = pd.date_range(origin_ts, periods=h, freq=FREQ)

            # Verdad-terreno: tomamos de d_all / y_all si existen
            mask_te = np.isin(d_all, future)
            if mask_te.sum() == 0:
                continue  # no hay cobertura en este origen-h, lo ignoramos

            y_true = y_all[mask_te]
            # Puede ocurrir que falten algunos meses en la intersección; tomamos la longitud efectiva
            eff_h = len(y_true)
            f_dates = future[:eff_h]

            # Ajustamos métricas usando escalas calculadas sobre y_tr
            for m in models:
                try:
                    y_pred_full = fit_predict(m, y_tr, d_tr, horizon=len(future), freq=FREQ)
                except Exception as e:
                    # Si el modelo falla, registramos NaN y continuamos
                    log4(f"[{bid}|{ctype}] Error en {m.get('name')} ({route}) @ {origin}, h={h}: {e}")
                    continue

                y_pred = np.asarray(y_pred_full, dtype=float)[:eff_h]
                metrics = _eval_series(y_true, y_pred, y_train_for_scales=y_tr)

                raw_rows.append({
                    "ID_BUILDING": bid, "FM_COST_TYPE": ctype,
                    "route": route, "model_name": m.get("name"),
                    "origin": origin, "horizon": h, "n_obs": eff_h,
                    **metrics
                })

# Construimos DataFrame raw y persistimos
if len(raw_rows) == 0:
    log4("No generamos filas en rolling_metrics_raw (revisar cobertura/orígenes).")
    rolling_raw = pd.DataFrame(columns=["ID_BUILDING","FM_COST_TYPE","route","model_name","origin","horizon","n_obs","MAE","WAPE","SMAPE","MASE1","MASE12"])
else:
    rolling_raw = pd.DataFrame(raw_rows)

rolling_raw.to_csv(PATH_RAW, sep=CSV_SEP, index=False)
log4(f"Guardado RAW: {PATH_RAW} con {len(rolling_raw)} filas")

# =========================
# Bloque 8) Agregación por serie y modelo
# =========================
if rolling_raw.empty:
    rolling_agg = pd.DataFrame(columns=["ID_BUILDING","FM_COST_TYPE","route","model_name","n_origins","MAE_mean","WAPE_mean","SMAPE_mean","MASE1_mean","MASE12_mean"])
else:
    rolling_agg = (
        rolling_raw
        .groupby(["ID_BUILDING","FM_COST_TYPE","route","model_name"], as_index=False)
        .agg(
            n_origins=("origin", "nunique"),
            MAE_mean=("MAE","mean"),
            WAPE_mean=("WAPE","mean"),
            SMAPE_mean=("SMAPE","mean"),
            MASE1_mean=("MASE1","mean"),
            MASE12_mean=("MASE12","mean")
        )
    )

rolling_agg.to_csv(PATH_AGG, sep=CSV_SEP, index=False)
log4(f"Guardado AGG: {PATH_AGG} con {len(rolling_agg)} filas")

# =========================
# Bloque 9) Selección top-2 y comparación contra SNaive12 (2024)
# =========================
# Leemos desempeño de SNaive12 en 2024 desde Paso 1
if os.path.exists(PATH_BASELINES_2024):
    baselines2024 = pd.read_csv(PATH_BASELINES_2024, sep=CSV_SEP)
    snaive24 = baselines2024[baselines2024["baseline"] == "SNaive12"][PAIR_COLS + ["MASE12"]].rename(columns={"MASE12":"MASE12_SNaive12_2024"})
else:
    snaive24 = pd.DataFrame(columns=PAIR_COLS + ["MASE12_SNaive12_2024"])
    log4(f"AVISO: no encontramos {PATH_BASELINES_2024}. La comparación contra SNaive12_2024 quedará vacía.")

# Seleccionamos top-2 por pareja según MASE12_mean (desempate WAPE_mean)
def _select_top2(df_pair):
    df_pair = df_pair.sort_values(["MASE12_mean","WAPE_mean"], ascending=[True, True])
    top = df_pair.head(2).reset_index(drop=True)
    out = {
        "modelo_top1": top.loc[0, "model_name"] if len(top) > 0 else None,
        "MASE12_top1": float(top.loc[0, "MASE12_mean"]) if len(top) > 0 else np.nan,
        "modelo_top2": top.loc[1, "model_name"] if len(top) > 1 else None,
        "MASE12_top2": float(top.loc[1, "MASE12_mean"]) if len(top) > 1 else np.nan,
        "route": df_pair["route"].iloc[0] if len(df_pair) else None
    }
    return pd.Series(out)

if rolling_agg.empty:
    sel = pd.DataFrame(columns=PAIR_COLS + ["route","modelo_top1","modelo_top2","MASE12_top1","MASE12_top2"])
else:
    sel = rolling_agg.groupby(PAIR_COLS, as_index=False).apply(_select_top2)
    if isinstance(sel.columns, pd.MultiIndex):
        sel.columns = sel.columns.get_level_values(-1)
    sel = sel.reset_index(drop=True)

# Comparamos contra SNaive12_2024
sel = sel.merge(snaive24, on=PAIR_COLS, how="left")
sel["delta_MASE12_vs_SNaive12"] = sel["MASE12_top1"] - sel["MASE12_SNaive12_2024"]
sel["winner_beats_snaive12"] = (sel["delta_MASE12_vs_SNaive12"] < 0).astype(int)

# Regla adicional: si no hay mejora vs SNaive12_2024, fijamos modelo_top1=SNaive12 y dejamos top2 como estaba
mask_no_improve = (~sel["MASE12_SNaive12_2024"].isna()) & (sel["winner_beats_snaive12"] == 0)
sel.loc[mask_no_improve, "modelo_top1"] = "SNaive12"

# Añadimos los casos flag_all_zero_train con baseline_cero si no aparecieran por ausencia de raw
zeros_df = panel_pairs.loc[panel_pairs[flag_col] == 1, PAIR_COLS].drop_duplicates()
if not zeros_df.empty:
    zeros_df = zeros_df.assign(route="INTERMITENTE",
                               modelo_top1="baseline_cero",
                               modelo_top2=None,
                               MASE12_top1=0.0,
                               MASE12_top2=np.nan,
                               MASE12_SNaive12_2024=np.nan,
                               delta_MASE12_vs_SNaive12=np.nan,
                               winner_beats_snaive12=1)
    sel = pd.concat([sel, zeros_df], ignore_index=True).drop_duplicates(subset=PAIR_COLS, keep="first")

# Persistimos selección
sel = sel[PAIR_COLS + ["route","modelo_top1","modelo_top2","MASE12_top1","MASE12_top2","winner_beats_snaive12","delta_MASE12_vs_SNaive12"]]
sel.to_csv(PATH_SEL, sep=CSV_SEP, index=False)
log4(f"Guardado SELECCIÓN: {PATH_SEL} con {len(sel)} parejas")

# =========================
# Bloque 10) Punto de control D (diagnóstico)
# =========================
if len(sel) > 0 and "winner_beats_snaive12" in sel.columns:
    pct_mejora = float(sel["winner_beats_snaive12"].mean()) if len(sel) else 0.0
    log4(f"Punto D — % de series donde top1 supera a SNaive12_2024: {pct_mejora:.2%}")

    # Si el porcentaje es bajo, registramos recomendación operativa
    if pct_mejora < 0.30:
        log4("Recomendación: reducir candidatos a Theta/ETS y/o activar tratamiento de outliers (mod_3) para las peores series y revalidar.")
    elif pct_mejora < 0.40:
        log4("Aviso: mejora moderada; considerar reducir menú a modelos más estables y revisar outliers selectivamente.")

    # Listado de series problemáticas por umbrales del JSON (si existen)
    thr = step3_cfg.get("metrics", {}).get("thresholds_fail", {"MASE12": 1.2, "SMAPE": 25.0})
    if not rolling_agg.empty:
        # Unimos top1 con su MASE12_mean y SMAPE_mean
        top1_perf = (rolling_agg.merge(sel[PAIR_COLS + ["modelo_top1"]], left_on=PAIR_COLS + ["model_name"], right_on=PAIR_COLS + ["modelo_top1"], how="inner"))
        prob = top1_perf[(top1_perf["MASE12_mean"] > float(thr.get("MASE12", 1.2))) | (top1_perf["SMAPE_mean"] > float(thr.get("SMAPE", 25.0)))]
        prob = prob.sort_values(["MASE12_mean","SMAPE_mean"], ascending=[False, False]).head(20)
        if not prob.empty:
            log4(f"Series por encima de umbrales (top 20): {len(prob)} casos")
            # Imprimimos algunas para inspección rápida
            display(prob[PAIR_COLS + ["model_name","MASE12_mean","SMAPE_mean","WAPE_mean"]].head(10))
else:
    log4("Punto D — No hay selección o columna de mejora, revisar ejecución del rolling.")

[2025-09-21T22:31:32] Población Paso 4 — total=2429 | INTERMITENTE=1397 | GENERAL=1032
[2025-09-21T22:31:32] [2|Eficiencia Energética] flag_all_zero_train=1 -> baseline_cero registrado.
[2025-09-21T22:31:33] [9|Eficiencia Energética] flag_all_zero_train=1 -> baseline_cero registrado.
[2025-09-21T22:31:33] [18|Eficiencia Energética] flag_all_zero_train=1 -> baseline_cero registrado.
[2025-09-21T22:31:34] [59|Obras] flag_all_zero_train=1 -> baseline_cero registrado.
[2025-09-21T22:31:34] [62|Licencias] flag_all_zero_train=1 -> baseline_cero registrado.
[2025-09-21T22:31:34] [74|Licencias] flag_all_zero_train=1 -> baseline_cero registrado.
[2025-09-21T22:31:37] [127|Servicios Extra] flag_all_zero_train=1 -> baseline_cero registrado.
[2025-09-21T22:31:37] [129|Eficiencia Energética] flag_all_zero_train=1 -> baseline_cero registrado.
[2025-09-21T22:31:41] [137|Obras] flag_all_zero_train=1 -> baseline_cero registrado.
[2025-09-21T22:31:43] [146|Eficiencia Energética] flag_all_zero_train=1 -> baseline_cero registrado.
[2025-09-21T22:31:43] [149|Eficiencia Energética] flag_all_zero_train=1 -> baseline_cero registrado.
[2025-09-21T22:31:46] [168|Licencias] flag_all_zero_train=1 -> baseline_cero registrado.
[2025-09-21T22:31:47] [171|Eficiencia Energética] flag_all_zero_train=1 -> baseline_cero registrado.
[2025-09-21T22:31:47] [171|Licencias] flag_all_zero_train=1 -> baseline_cero registrado.
[2025-09-21T22:31:49] [179|Servicios Extra] flag_all_zero_train=1 -> baseline_cero registrado.
[2025-09-21T22:31:50] [199|Licencias] flag_all_zero_train=1 -> baseline_cero registrado.
[2025-09-21T22:31:57] [215|Eficiencia Energética] flag_all_zero_train=1 -> baseline_cero registrado.
[2025-09-21T22:31:59] [224|Servicios Extra] flag_all_zero_train=1 -> baseline_cero registrado.
[2025-09-21T22:32:00] [229|Eficiencia Energética] flag_all_zero_train=1 -> baseline_cero registrado.
[2025-09-21T22:32:01] [231|Servicios Extra] flag_all_zero_train=1 -> baseline_cero registrado.
[2025-09-21T22:32:03] [239|Obras] flag_all_zero_train=1 -> baseline_cero registrado.
[2025-09-21T22:32:03] [239|Servicios Extra] flag_all_zero_train=1 -> baseline_cero registrado.
[2025-09-21T22:32:06] [260|Licencias] flag_all_zero_train=1 -> baseline_cero registrado.
[2025-09-21T22:32:07] [265|Obras] flag_all_zero_train=1 -> baseline_cero registrado.
[2025-09-21T22:32:12] [345|Servicios Extra] flag_all_zero_train=1 -> baseline_cero registrado.
[2025-09-21T22:32:14] [379|Obras] flag_all_zero_train=1 -> baseline_cero registrado.
[2025-09-21T22:32:14] [594|Obras] flag_all_zero_train=1 -> baseline_cero registrado.
[2025-09-21T22:32:14] [617|Licencias] flag_all_zero_train=1 -> baseline_cero registrado.
[2025-09-21T22:32:16] [713|Licencias] flag_all_zero_train=1 -> baseline_cero registrado.
[2025-09-21T22:32:18] [935|Eficiencia Energética] flag_all_zero_train=1 -> baseline_cero registrado.
[2025-09-21T22:32:19] [1007|Eficiencia Energética] flag_all_zero_train=1 -> baseline_cero registrado.
[2025-09-21T22:32:22] [1060|Servicios Extra] flag_all_zero_train=1 -> baseline_cero registrado.
[2025-09-21T22:32:22] [1065|Eficiencia Energética] flag_all_zero_train=1 -> baseline_cero registrado.
[2025-09-21T22:32:29] [1111|Servicios Extra] flag_all_zero_train=1 -> baseline_cero registrado.
[2025-09-21T22:32:33] [1306|Obras] flag_all_zero_train=1 -> baseline_cero registrado.
[2025-09-21T22:32:34] [1462|Licencias] flag_all_zero_train=1 -> baseline_cero registrado.
[2025-09-21T22:32:36] [1000026|Servicios Extra] flag_all_zero_train=1 -> baseline_cero registrado.
[2025-09-21T22:32:42] [1000270|Eficiencia Energética] flag_all_zero_train=1 -> baseline_cero registrado.
[2025-09-21T22:32:43] [1000345|Obras] flag_all_zero_train=1 -> baseline_cero registrado.
[2025-09-21T22:32:45] [1000409|Obras] flag_all_zero_train=1 -> baseline_cero registrado.
[2025-09-21T22:32:45] [1000413|Licencias] flag_all_zero_train=1 -> baseline_cero registrado.
[2025-09-21T22:32:45] [1000414|Licencias] flag_all_zero_train=1 -> baseline_cero registrado.
[2025-09-21T22:32:46] [1000431|Eficiencia Energética] flag_all_zero_train=1 -> baseline_cero registrado.
[2025-09-21T22:32:47] [1000437|Licencias] flag_all_zero_train=1 -> baseline_cero registrado.
[2025-09-21T22:32:47] [1000438|Obras] flag_all_zero_train=1 -> baseline_cero registrado.
[2025-09-21T22:32:48] [1000449|Licencias] flag_all_zero_train=1 -> baseline_cero registrado.
[2025-09-21T22:32:48] [1000450|Licencias] flag_all_zero_train=1 -> baseline_cero registrado.
[2025-09-21T22:32:48] [1000451|Obras] flag_all_zero_train=1 -> baseline_cero registrado.
[2025-09-21T22:32:49] [1000484|Obras] flag_all_zero_train=1 -> baseline_cero registrado.
[2025-09-21T22:32:50] [1000496|Obras] flag_all_zero_train=1 -> baseline_cero registrado.
[2025-09-21T22:32:50] [1000496|Servicios Extra] flag_all_zero_train=1 -> baseline_cero registrado.
[2025-09-21T22:32:52] [1000532|Obras] flag_all_zero_train=1 -> baseline_cero registrado.
[2025-09-21T22:32:52] [1000532|Servicios Extra] flag_all_zero_train=1 -> baseline_cero registrado.
[2025-09-21T22:32:56] [1000558|Eficiencia Energética] flag_all_zero_train=1 -> baseline_cero registrado.
[2025-09-21T22:32:58] [1000601|Licencias] flag_all_zero_train=1 -> baseline_cero registrado.
[2025-09-21T22:33:01] [1000666|Licencias] flag_all_zero_train=1 -> baseline_cero registrado.
[2025-09-21T22:33:01] [1000667|Eficiencia Energética] flag_all_zero_train=1 -> baseline_cero registrado.
[2025-09-21T22:33:01] [1000667|Licencias] flag_all_zero_train=1 -> baseline_cero registrado.
[2025-09-21T22:33:01] [1000667|Obras] flag_all_zero_train=1 -> baseline_cero registrado.
[2025-09-21T22:33:02] [1000733|Eficiencia Energética] flag_all_zero_train=1 -> baseline_cero registrado.
[2025-09-21T22:33:03] [1000758|Eficiencia Energética] flag_all_zero_train=1 -> baseline_cero registrado.
[2025-09-21T22:33:03] [1000761|Eficiencia Energética] flag_all_zero_train=1 -> baseline_cero registrado.
[2025-09-21T22:33:03] [1000762|Eficiencia Energética] flag_all_zero_train=1 -> baseline_cero registrado.
[2025-09-21T22:33:03] [1000763|Eficiencia Energética] flag_all_zero_train=1 -> baseline_cero registrado.
[2025-09-21T22:33:03] [1000764|Eficiencia Energética] flag_all_zero_train=1 -> baseline_cero registrado.
[2025-09-21T22:33:03] [1000765|Eficiencia Energética] flag_all_zero_train=1 -> baseline_cero registrado.
[2025-09-21T22:33:04] [1000766|Eficiencia Energética] flag_all_zero_train=1 -> baseline_cero registrado.
[2025-09-21T22:33:06] [1000817|Licencias] flag_all_zero_train=1 -> baseline_cero registrado.
[2025-09-21T22:33:10] [1000859|Obras] flag_all_zero_train=1 -> baseline_cero registrado.
[2025-09-21T22:33:12] [1000925|Obras] flag_all_zero_train=1 -> baseline_cero registrado.
[2025-09-21T22:33:13] [1001055|Mtto. Correctivo] flag_all_zero_train=1 -> baseline_cero registrado.
[2025-09-21T22:33:13] [1001077|Licencias] flag_all_zero_train=1 -> baseline_cero registrado.
[2025-09-21T22:33:15] [1001113|Licencias] flag_all_zero_train=1 -> baseline_cero registrado.
[2025-09-21T22:33:15] [1001113|Obras] flag_all_zero_train=1 -> baseline_cero registrado.
[2025-09-21T22:33:17] [1001119|Obras] flag_all_zero_train=1 -> baseline_cero registrado.
[2025-09-21T22:33:19] [1001122|Obras] flag_all_zero_train=1 -> baseline_cero registrado.
[2025-09-21T22:33:20] [1001126|Licencias] flag_all_zero_train=1 -> baseline_cero registrado.
[2025-09-21T22:33:20] [1001127|Eficiencia Energética] flag_all_zero_train=1 -> baseline_cero registrado.
[2025-09-21T22:33:25] [1001133|Obras] flag_all_zero_train=1 -> baseline_cero registrado.
[2025-09-21T22:33:26] [1001134|Obras] flag_all_zero_train=1 -> baseline_cero registrado.
[2025-09-21T22:33:26] [1001135|Obras] flag_all_zero_train=1 -> baseline_cero registrado.
[2025-09-21T22:33:27] [1001140|Obras] flag_all_zero_train=1 -> baseline_cero registrado.
[2025-09-21T22:33:28] [1001141|Obras] flag_all_zero_train=1 -> baseline_cero registrado.
[2025-09-21T22:33:28] [1001143|Obras] flag_all_zero_train=1 -> baseline_cero registrado.
[2025-09-21T22:33:30] [1001151|Obras] flag_all_zero_train=1 -> baseline_cero registrado.
[2025-09-21T22:33:31] [1001154|Licencias] flag_all_zero_train=1 -> baseline_cero registrado.
[2025-09-21T22:33:32] [1001156|Obras] flag_all_zero_train=1 -> baseline_cero registrado.
[2025-09-21T22:33:33] [1001159|Servicios Extra] flag_all_zero_train=1 -> baseline_cero registrado.
[2025-09-21T22:33:33] [1001160|Obras] flag_all_zero_train=1 -> baseline_cero registrado.
[2025-09-21T22:33:38] [1001173|Eficiencia Energética] flag_all_zero_train=1 -> baseline_cero registrado.
[2025-09-21T22:33:38] [1001173|Obras] flag_all_zero_train=1 -> baseline_cero registrado.
[2025-09-21T22:33:42] [1001205|Licencias] flag_all_zero_train=1 -> baseline_cero registrado.
[2025-09-21T22:33:42] [1001206|Licencias] flag_all_zero_train=1 -> baseline_cero registrado.
[2025-09-21T22:33:42] [1001207|Licencias] flag_all_zero_train=1 -> baseline_cero registrado.
[2025-09-21T22:33:42] [1001209|Licencias] flag_all_zero_train=1 -> baseline_cero registrado.
[2025-09-21T22:33:42] [1001213|Licencias] flag_all_zero_train=1 -> baseline_cero registrado.
[2025-09-21T22:33:43] [1001252|Eficiencia Energética] flag_all_zero_train=1 -> baseline_cero registrado.
[2025-09-21T22:33:44] [1001253|Eficiencia Energética] flag_all_zero_train=1 -> baseline_cero registrado.
[2025-09-21T22:33:45] [1001265|Licencias] flag_all_zero_train=1 -> baseline_cero registrado.
[2025-09-21T22:33:49] [1001365|Obras] flag_all_zero_train=1 -> baseline_cero registrado.
[2025-09-21T22:33:54] [1001388|Licencias] flag_all_zero_train=1 -> baseline_cero registrado.
[2025-09-21T22:33:56] [1001407|Licencias] flag_all_zero_train=1 -> baseline_cero registrado.
[2025-09-21T22:33:58] [1001471|Mtto. Correctivo] flag_all_zero_train=1 -> baseline_cero registrado.
[2025-09-21T22:33:58] [1001476|Licencias] flag_all_zero_train=1 -> baseline_cero registrado.
[2025-09-21T23:08:39] Guardado RAW: /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3/METRICAS/rolling_metrics_raw.csv con 45258 filas
[2025-09-21T23:08:39] Guardado AGG: /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3/METRICAS/rolling_metrics_agg.csv con 15086 filas
[2025-09-21T23:08:43] Guardado SELECCIÓN: /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3/METRICAS/seleccion_modelos_cv.csv con 2429 parejas
[2025-09-21T23:08:43] Punto D — % de series donde top1 supera a SNaive12_2024: 69.78%
[2025-09-21T23:08:43] Series por encima de umbrales (top 20): 20 casos

# ============================================================
# ESTRATEGIA 3 — PASO 5
# ENSEMBLES Y PREDICCIÓN 2024 (12M) POR SERIE
# ============================================================
# En este paso generamos ŷ_2024 por pareja (ID_BUILDING, FM_COST_TYPE)
# combinando top-2 modelos (50/50 o ponderado por 1/MASE12_mean),
# con fallbacks (SNaive12, baseline_cero) y trazabilidad en logs.
# ============================================================

# -----------------------------
# Bloque 1 — Parámetros y logging
# -----------------------------

# Fijamos semilla global para reproducibilidad de cualquier aleatoriedad interna.
np.random.seed(7)

# Asumimos que estas variables existen desde pasos previos; si no, definimos defaults seguros.
ruta_base_3 = globals().get("ruta_base_3", "/content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3")
CSV_SEP     = globals().get("CSV_SEP", ";")
PAIR_COLS   = globals().get("PAIR_COLS", ["ID_BUILDING","FM_COST_TYPE"])
DATECOL     = globals().get("DATECOL", "FECHA")
VALUE_TRAINTEST = globals().get("VALUE_TRAINTEST", "cost_float_mod")

# Rutas de trabajo
RUTA_METRICAS   = os.path.join(ruta_base_3, "METRICAS")
RUTA_LOGS       = os.path.join(ruta_base_3, "LOGS")
RUTA_RESULTADOS = os.path.join(ruta_base_3, "RESULTADOS")
RUTA_CONFIG     = os.path.join(ruta_base_3, "CONFIG")

os.makedirs(RUTA_METRICAS, exist_ok=True)
os.makedirs(RUTA_LOGS, exist_ok=True)
os.makedirs(RUTA_RESULTADOS, exist_ok=True)
os.makedirs(RUTA_CONFIG, exist_ok=True)

# Log del paso
LOG_PATH_STEP5 = os.path.join(RUTA_LOGS, "estrategia3_step5.log")
def log5(msg: str):
    # Dejamos traza en archivo y consola para auditoría
    stamp = datetime.now().isoformat(timespec="seconds")
    line = f"[{stamp}] {msg}"
    print(line)
    with open(LOG_PATH_STEP5, "a", encoding="utf-8") as f:
        f.write(line + "\n")

# Verificamos presencia de dataframes canónicos del Paso 0 (train_full/test_full)
assert "train_full" in globals(), "Falta train_full (generado en Paso 0)."
assert "test_full"  in globals(), "Falta test_full (generado en Paso 0)."

# Normalizamos fechas a mensual (MS)
def _ensure_ms(df: pd.DataFrame, datecol: str) -> pd.DataFrame:
    df = df.copy()
    df[datecol] = pd.to_datetime(df[datecol]).dt.to_period("M").dt.to_timestamp()
    return df

train_full = _ensure_ms(train_full, DATECOL)
test_full  = _ensure_ms(test_full,  DATECOL)

# Definimos rango futuro 2024
FREQ   = "MS"
H      = 12
FUTURE_2024 = pd.date_range("2024-01-01", periods=H, freq=FREQ)

log5("Iniciamos Paso 5 — Ensembles y predicción 2024.")

# -----------------------------------------
# Bloque 2 — Carga artefactos y panel de trabajo
# -----------------------------------------

# Artefactos esperados
PATH_SEG_ENR   = os.path.join(RUTA_METRICAS, "segmentation_intermitencia_step3.csv")
PATH_PAIRS_INT = os.path.join(RUTA_METRICAS, "pairs_intermittentes.csv")
PATH_PAIRS_GEN = os.path.join(RUTA_METRICAS, "pairs_generales.csv")
PATH_CFG_JSON  = os.path.join(RUTA_CONFIG,  "step3_menu_modelos_config.json")
PATH_ROLL_AGG  = os.path.join(RUTA_METRICAS, "rolling_metrics_agg.csv")
PATH_SEL_CV    = os.path.join(RUTA_METRICAS, "seleccion_modelos_cv.csv")
PATH_BASELINES = os.path.join(RUTA_METRICAS, "metrics_baselines_cv.csv")
PATH_SOLOTEST  = os.path.join(RUTA_METRICAS, "pairs_solo_test.csv")  # del Paso 0

# Cargamos artefactos mínimos
seg_enr   = pd.read_csv(PATH_SEG_ENR, sep=CSV_SEP)
pairs_int = pd.read_csv(PATH_PAIRS_INT, sep=CSV_SEP)
pairs_gen = pd.read_csv(PATH_PAIRS_GEN, sep=CSV_SEP)
sel_cv    = pd.read_csv(PATH_SEL_CV, sep=CSV_SEP)

# Cargamos rolling agg y baselines si existen (para ponderación por 1/MASE12 y comparación TSB vs SNaive12)
roll_agg  = pd.read_csv(PATH_ROLL_AGG, sep=CSV_SEP) if os.path.exists(PATH_ROLL_AGG) else pd.DataFrame()
df_base   = pd.read_csv(PATH_BASELINES, sep=CSV_SEP) if os.path.exists(PATH_BASELINES) else pd.DataFrame()
pairs_solo_test = pd.read_csv(PATH_SOLOTEST, sep=CSV_SEP) if os.path.exists(PATH_SOLOTEST) else pd.DataFrame(columns=PAIR_COLS)

# Panel de pares con route y bandera all_zero
flag_col = "flag_all_zero_train"
pairs_int["route"] = "INTERMITENTE"
pairs_gen["route"] = "GENERAL"
panel_pairs = pd.concat([pairs_int, pairs_gen], ignore_index=True)
panel_pairs = panel_pairs.merge(
    seg_enr[PAIR_COLS + ["segment", flag_col]].drop_duplicates(),
    on=PAIR_COLS, how="left"
)
panel_pairs["route"] = panel_pairs["route"].fillna(panel_pairs["segment"]).fillna("GENERAL")

# Añadimos top-1 y top-2 desde selección
panel_pairs = panel_pairs.merge(
    sel_cv[PAIR_COLS + ["modelo_top1", "modelo_top2"]],
    on=PAIR_COLS, how="left"
)

# Conteos de referencia
n_total = panel_pairs[PAIR_COLS].drop_duplicates().shape[0]
n_int   = (panel_pairs["route"] == "INTERMITENTE").sum()
n_gen   = (panel_pairs["route"] == "GENERAL").sum()
n_zero  = int(panel_pairs[flag_col].fillna(0).astype(int).sum())
log5(f"Población Paso 5 — total={n_total} | INTERMITENTE={n_int} | GENERAL={n_gen} | all_zero={n_zero}")

# ------------------------------------------------
# Bloque 3 — Wrappers de predicción (reutilizamos Paso 4)
# ------------------------------------------------

# Baselines y modelos
def _baseline_naive1(y_train, h):
    # Repetimos el último valor observado
    if len(y_train) == 0:
        return np.zeros(h, dtype=float)
    return np.repeat(float(y_train[-1]), h)

def _baseline_snaive12(y_train, dates_train, h):
    # Usamos el valor del mismo mes del año anterior si existe; si no, caemos al último valor
    if len(y_train) == 0:
        return np.zeros(h, dtype=float)
    df = pd.DataFrame({"FECHA": pd.to_datetime(dates_train), "y": y_train})
    df["MONTH_NUM"] = df["FECHA"].dt.month
    df["YEAR"]      = df["FECHA"].dt.year
    last_year = df["YEAR"].max()
    map_last_year = df[df["YEAR"] == last_year].set_index("MONTH_NUM")["y"].to_dict()
    last_date = pd.to_datetime(dates_train[-1]).to_period("M").to_timestamp()
    future = pd.date_range(last_date + pd.offsets.MonthBegin(1), periods=h, freq="MS")
    pred = []
    for d in future:
        pred.append(map_last_year.get(d.month, float(y_train[-1])))
    return np.asarray(pred, dtype=float)

def _baseline_median_seasonal(y_train, dates_train, h, m=12):
    # Mediana por mes sobre el train; fallback al último valor
    if len(y_train) == 0:
        return np.zeros(h, dtype=float)
    df = pd.DataFrame({"FECHA": pd.to_datetime(dates_train), "y": y_train})
    df["MONTH_NUM"] = df["FECHA"].dt.month
    med = df.groupby("MONTH_NUM")["y"].median().to_dict()
    last_date = pd.to_datetime(dates_train[-1]).to_period("M").to_timestamp()
    future = pd.date_range(last_date + pd.offsets.MonthBegin(1), periods=h, freq="MS")
    return np.array([med.get(d.month, float(y_train[-1])) for d in future], dtype=float)

def _theta_predict(y_train, dates_train, h):
    # Theta con period=12; si falla o no está disponible, caemos a mediana estacional
    if len(y_train) == 0:
        return np.zeros(h, dtype=float)
    if ThetaModel is None:
        return _baseline_median_seasonal(y_train, dates_train, h, m=12)
    try:
        tm = ThetaModel(np.asarray(y_train, dtype=float), period=12).fit()
        return np.asarray(tm.forecast(h), dtype=float)
    except Exception:
        return _baseline_median_seasonal(y_train, dates_train, h, m=12)

def _ets_auto_predict(y_train, h):
    # ETS aditivo amortiguado; si no, Naive1
    if len(y_train) < 3 or ExponentialSmoothing is None:
        return _baseline_naive1(y_train, h)
    try:
        model = ExponentialSmoothing(
            np.asarray(y_train, dtype=float),
            trend="add", seasonal="add", seasonal_periods=12,
            damped_trend=True, initialization_method="estimated"
        ).fit(optimized=True)
        return np.asarray(model.forecast(h), dtype=float)
    except Exception:
        return _baseline_naive1(y_train, h)

def _holt_damped_predict(y_train, h):
    # Holt amortiguado sin estacionalidad
    if len(y_train) < 3 or ExponentialSmoothing is None:
        return _baseline_naive1(y_train, h)
    try:
        model = ExponentialSmoothing(
            np.asarray(y_train, dtype=float),
            trend="add", seasonal=None,
            damped_trend=True, initialization_method="estimated"
        ).fit(optimized=True)
        return np.asarray(model.forecast(h), dtype=float)
    except Exception:
        return _baseline_naive1(y_train, h)

def _arima_auto_light(y_train, h):
    # Grid ligero ARIMA; si falla, Naive1
    if SARIMAX is None or len(y_train) < 8:
        return _baseline_naive1(y_train, h)
    y = np.asarray(y_train, dtype=float)
    best_aic = np.inf
    best_forecast = None
    for p in [0,1]:
        for d in [0,1]:
            for q in [0,1]:
                try:
                    mod = SARIMAX(y, order=(p,d,q), seasonal_order=(0,0,0,0),
                                  trend="c", enforce_stationarity=False, enforce_invertibility=False)
                    res = mod.fit(disp=False)
                    if res.aic < best_aic:
                        best_aic = res.aic
                        best_forecast = res.forecast(h)
                except Exception:
                    continue
    if best_forecast is None:
        return _baseline_naive1(y_train, h)
    return np.asarray(best_forecast, dtype=float)

# Modelos intermitentes
def _croston_classic(y, h, alpha=0.1):
    y = np.asarray(y, dtype=float)
    n = len(y)
    if n == 0:
        return np.zeros(h, dtype=float)
    z, p, k = 0.0, 0.0, 0
    first = True
    for v in y:
        k += 1
        if v > 0:
            if first:
                z, p, first = v, k, False
            else:
                z = z + alpha * (v - z)
                p = p + alpha * (k - p)
            k = 0
    f = (z / p) if p > 0 else 0.0
    return np.repeat(f, h)

def _croston_sba(y, h, alpha=0.1):
    base = _croston_classic(y, h, alpha=alpha)
    return base * (1.0 - alpha/2.0)

def _tsb(y, h, alpha=0.2, beta=0.1):
    y = np.asarray(y, dtype=float)
    if len(y) == 0:
        return np.zeros(h, dtype=float)
    demand = (y > 0).astype(int)
    sizes  = np.where(y > 0, y, 0.0)
    p = demand[0]
    z = sizes[0] if sizes[0] > 0 else (np.mean(sizes[sizes > 0]) if np.any(sizes > 0) else 0.0)
    for t in range(1, len(y)):
        p = p + alpha * (demand[t] - p)
        if y[t] > 0:
            z = z + beta * (y[t] - z)
    f = p * z
    return np.repeat(float(f), h)

def _adida_mapa(y_train, h):
    # ADIDA/MAPA simple: agregación trimestral + modelo LF + desagregación por perfil histórico
    y = np.asarray(y_train, dtype=float)
    if len(y) < 6:
        return _baseline_median_seasonal(y_train, np.arange(len(y)), h, m=12)
    pad = (3 - (len(y) % 3)) % 3
    y_pad = np.hstack([y, np.zeros(pad)])
    yQ = y_pad.reshape(-1, 3).sum(axis=1)
    # Modelo LF (preferimos ETS si hay suficientes puntos)
    if ExponentialSmoothing is not None and len(yQ) >= 3:
        try:
            mQ = ExponentialSmoothing(yQ, trend="add", seasonal=None, damped_trend=True).fit(optimized=True)
            fq = np.asarray(mQ.forecast(int(np.ceil(h/3))), dtype=float)
        except Exception:
            fq = np.repeat(float(yQ[-1]) if len(yQ) else 0.0, int(np.ceil(h/3)))
    else:
        try:
            tm = ThetaModel(yQ).fit()
            fq = np.asarray(tm.forecast(int(np.ceil(h/3))), dtype=float)
        except Exception:
            fq = np.repeat(float(yQ[-1]) if len(yQ) else 0.0, int(np.ceil(h/3)))
    # Pesos mensuales medios por trimestre
    blocks = y_pad.reshape(-1, 3)
    totals = blocks.sum(axis=1, keepdims=True)
    weights = np.divide(blocks, np.where(totals==0, 1.0, totals))
    w = np.nan_to_num(weights.mean(axis=0))
    if w.sum() == 0:
        w = np.array([1/3, 1/3, 1/3], dtype=float)
    # Desagregamos a meses
    f_months = []
    for qv in fq:
        f_months.extend(list(w * qv))
    return np.asarray(f_months[:h], dtype=float)

# Enrutador por nombre de modelo (ajustamos a los nombres usados en selección)
def get_model_spec_by_name(name: str):
    s = (name or "").strip().lower()
    # Mapeamos los nombres más probables de la selección / menús previos
    if s in ["snaive12","snaive_12","seasonal_naive","snaive"]:
        return ("baseline", "snaive12")
    if s in ["naive1","naive","last"]:
        return ("baseline", "naive1")
    if s in ["medianaestacional","median_seasonal","mediana_estacional"]:
        return ("baseline", "median_seasonal")
    if s in ["theta","thetamodel","theta_model"]:
        return ("theta", None)
    if s in ["ets","ets_auto","etsauto"]:
        return ("ets", None)
    if s in ["holt","holt_damped","holt-amortiguado","holt_damped_trend"]:
        return ("holt", None)
    if s in ["arima","arima_auto","sarima_light","auto_arima"]:
        return ("arima", None)
    if s in ["croston","croston_classic"]:
        return ("croston", "classic")
    if s in ["sba","croston_sba"]:
        return ("croston", "sba")
    if s in ["tsb"]:
        return ("tsb", None)
    if s in ["adida","mapa","adida_mapa","imapa"]:
        return ("adida_mapa", None)
    # Si no reconocemos, devolvemos baseline robusta
    return ("baseline", "median_seasonal")

def fit_predict_by_name(model_name: str, y_train, dates_train, h):
    fam, var = get_model_spec_by_name(model_name)
    try:
        if fam == "baseline":
            if var == "naive1":
                return _baseline_naive1(y_train, h)
            if var == "snaive12":
                return _baseline_snaive12(y_train, dates_train, h)
            if var == "median_seasonal":
                return _baseline_median_seasonal(y_train, dates_train, h, m=12)
            # Por seguridad
            return _baseline_naive1(y_train, h)
        if fam == "theta":
            return _theta_predict(y_train, dates_train, h)
        if fam == "ets":
            return _ets_auto_predict(y_train, h)
        if fam == "holt":
            return _holt_damped_predict(y_train, h)
        if fam == "arima":
            return _arima_auto_light(y_train, h)
        if fam == "croston":
            return _croston_classic(y_train, h) if var == "classic" else _croston_sba(y_train, h)
        if fam == "tsb":
            return _tsb(y_train, h)
        if fam == "adida_mapa":
            return _adida_mapa(y_train, h)
    except Exception as e:
        log5(f"AVISO: fallo en {model_name}: {e}. Aplicamos fallback SNaive12.")
        return _baseline_snaive12(y_train, dates_train, h)
    # Si algo no entra en ninguna rama, devolvemos Naive1
    return _baseline_naive1(y_train, h)

# Métricas auxiliares para pesos (eps para robustez)
def _safe_div(num, den, eps=1e-8):
    return num / (den + eps)

# ------------------------------------------------
# Bloque 4 — Lógica por pareja y generación de ŷ_2024
# ------------------------------------------------

# Construimos índice por pareja para recuperar serie completa
full_df = pd.concat([train_full, test_full], ignore_index=True)
full_df = full_df.sort_values(PAIR_COLS + [DATECOL]).reset_index(drop=True)

grouped = full_df.groupby(PAIR_COLS, sort=False)

# Para comparar TSB vs SNaive12 en intermitentes, preparamos una vista de roll_agg
roll_key_cols = ["ID_BUILDING","FM_COST_TYPE","model_name","MASE12_mean"]
if set(roll_key_cols).issubset(roll_agg.columns):
    roll_view = roll_agg[roll_key_cols].copy()
else:
    roll_view = pd.DataFrame(columns=roll_key_cols)

# Preparamos buffers de salida
rows_out = []
alertas_rows = []

# Preparamos set de SOLO TEST (si existiera)
solo_test_set = set(map(tuple, pairs_solo_test[PAIR_COLS].drop_duplicates().values.tolist())) if not pairs_solo_test.empty else set()

# Generamos un diccionario de MASE12_mean por (pair, model_name) para ponderación
def mase12_lookup(bid, ctype, model_name):
    if roll_view.empty:
        return np.nan
    sub = roll_view[(roll_view["ID_BUILDING"]==bid) & (roll_view["FM_COST_TYPE"]==ctype) & (roll_view["model_name"].str.lower()==str(model_name).lower())]
    if sub.empty:
        return np.nan
    return float(sub["MASE12_mean"].iloc[0])

# Bucle principal por pareja
for _, prow in panel_pairs.iterrows():
    bid, ctype = prow[PAIR_COLS[0]], prow[PAIR_COLS[1]]
    route      = str(prow.get("route", "GENERAL")).upper()
    is_zero    = int(prow.get(flag_col, 0)) == 1
    top1       = prow.get("modelo_top1", None)
    top2       = prow.get("modelo_top2", None)

    # Extraemos serie completa y particionamos train/test explícitamente
    try:
        sub = grouped.get_group((bid, ctype)).sort_values(DATECOL)
    except KeyError:
        log5(f"AVISO: no encontramos datos para pareja {(bid, ctype)}. Saltamos.")
        continue

    # Train 2021–2023
    tr = sub[(sub[DATECOL] >= pd.Timestamp("2021-01-01")) & (sub[DATECOL] <= pd.Timestamp("2023-12-01"))]
    y_tr = tr[VALUE_TRAINTEST].astype(float).values
    d_tr = tr[DATECOL].values

    # Test 2024 (verdad-terreno si existe)
    te = sub[(sub[DATECOL] >= pd.Timestamp("2024-01-01")) & (sub[DATECOL] <= pd.Timestamp("2024-12-01"))]
    y_true_2024 = te.set_index(DATECOL)[VALUE_TRAINTEST].reindex(FUTURE_2024).fillna(np.nan).values

    # Caso 1: cero estructural
    if is_zero:
        yhat = np.zeros(H, dtype=float)
        for i, d in enumerate(FUTURE_2024):
            rows_out.append({
                PAIR_COLS[0]: bid, PAIR_COLS[1]: ctype, "FECHA": d,
                "yhat_modelo1": 0.0, "yhat_modelo2": np.nan, "yhat_combo": 0.0,
                "fallback_flag": 1, "route": route,
                "modelo_top1": "baseline_cero", "modelo_top2": None,
                "combo_rule": "baseline_cero"
            })
            # Si observamos actividad real en 2024, registramos alerta
            if not np.isnan(y_true_2024[i]) and y_true_2024[i] > 0:
                alertas_rows.append({
                    PAIR_COLS[0]: bid, PAIR_COLS[1]: ctype, "FECHA": d,
                    "yhat_combo": 0.0, "y_true_2024": float(y_true_2024[i]),
                    "rule": "cero→actividad"
                })
        continue

    # Caso 2: SOLO TEST (sin histórico en 2021–2023)
    is_solo_test = (bid, ctype) in solo_test_set or (len(y_tr) == 0)
    if is_solo_test:
        yhat = _baseline_snaive12(y_tr, d_tr if len(d_tr)>0 else np.array([pd.Timestamp("2023-12-01")]), H)
        yhat = np.maximum(yhat, 0.0)
        for i, d in enumerate(FUTURE_2024):
            rows_out.append({
                PAIR_COLS[0]: bid, PAIR_COLS[1]: ctype, "FECHA": d,
                "yhat_modelo1": float(yhat[i]), "yhat_modelo2": np.nan, "yhat_combo": float(yhat[i]),
                "fallback_flag": 1, "route": route,
                "modelo_top1": "SNaive12", "modelo_top2": None,
                "combo_rule": "fallback_snaive12"
            })
        continue

    # Caso 3: intermitentes (prioridad TSB; si no mejora a SNaive12 en CV, usar SNaive12)
    if route == "INTERMITENTE":
        # Comparamos TSB vs SNaive12 con MASE12_mean si disponemos de roll_agg
        mase_tsb   = mase12_lookup(bid, ctype, "TSB")
        mase_snv   = mase12_lookup(bid, ctype, "SNaive12")
        use_snaive = False
        if not np.isnan(mase_tsb) and not np.isnan(mase_snv):
            use_snaive = (mase_tsb >= mase_snv)
        # Si no tenemos métricas, intentamos TSB por defecto
        if use_snaive:
            yhat1 = _baseline_snaive12(y_tr, d_tr, H)
            yhat2 = np.full(H, np.nan)
            combo = yhat1.copy()
            combo_rule = "fallback_snaive12_cv"
            fallback_flag = 1
            m1_name, m2_name = "SNaive12", None
        else:
            # TSB como top1; si hay tiempo/queremos, combinamos con ADIDA/MAPA
            yhat1 = _tsb(y_tr, H)
            # Probamos ADIDA/MAPA como modelo 2
            yhat2 = _adida_mapa(y_tr, H)
            # Intentamos ponderación por 1/MASE12_mean si existe en roll_agg; si no, 50/50
            m1_mase = mase12_lookup(bid, ctype, "TSB")
            m2_mase = mase12_lookup(bid, ctype, "ADIDA_MAPA")
            if not np.isnan(m1_mase) and not np.isnan(m2_mase) and m1_mase>0 and m2_mase>0:
                w1 = 1.0 / m1_mase
                w2 = 1.0 / m2_mase
                s  = w1 + w2
                combo = (w1/s) * yhat1 + (w2/s) * yhat2
                combo_rule = "weighted_by_inv_MASE12"
            else:
                combo = 0.5 * yhat1 + 0.5 * yhat2
                combo_rule = "mean50_50"
            fallback_flag = 0
            m1_name, m2_name = "TSB", "ADIDA/MAPA"
        # Aseguramos no negatividad
        yhat1 = np.maximum(yhat1, 0.0)
        if isinstance(yhat2, np.ndarray):
            yhat2 = np.maximum(yhat2, 0.0)
        combo = np.maximum(combo, 0.0)
        # Volcamos 12 filas
        for i, d in enumerate(FUTURE_2024):
            rows_out.append({
                PAIR_COLS[0]: bid, PAIR_COLS[1]: ctype, "FECHA": d,
                "yhat_modelo1": float(yhat1[i]),
                "yhat_modelo2": (float(yhat2[i]) if isinstance(yhat2, np.ndarray) else np.nan),
                "yhat_combo": float(combo[i]),
                "fallback_flag": int(fallback_flag),
                "route": route,
                "modelo_top1": m1_name,
                "modelo_top2": m2_name,
                "combo_rule": combo_rule
            })
        continue

    # Caso 4: generales (usamos top-1/top-2 de selección y combinamos)
    # Si faltara el nombre de top1, caemos a SNaive12
    if not isinstance(top1, str) or len(top1.strip()) == 0:
        top1 = "SNaive12"
    # Si top2 no existe, haremos combo=top1
    if not isinstance(top2, str) or len(top2.strip()) == 0:
        top2 = None

    # Ejecutamos modelos sobre todo el train 2021–2023
    yhat1 = fit_predict_by_name(top1, y_tr, d_tr, H)
    yhat2 = fit_predict_by_name(top2, y_tr, d_tr, H) if top2 else None

    # Aseguramos no negatividad
    yhat1 = np.maximum(yhat1, 0.0)
    if isinstance(yhat2, np.ndarray):
        yhat2 = np.maximum(yhat2, 0.0)

    # Combinación: por defecto 50/50; si tenemos MASE12_mean de ambos, ponderamos por 1/MASE12_mean
    if isinstance(yhat2, np.ndarray):
        m1_mase = mase12_lookup(bid, ctype, top1)
        m2_mase = mase12_lookup(bid, ctype, top2)
        if not np.isnan(m1_mase) and not np.isnan(m2_mase) and m1_mase>0 and m2_mase>0:
            w1 = 1.0 / m1_mase
            w2 = 1.0 / m2_mase
            s  = w1 + w2
            combo = (w1/s) * yhat1 + (w2/s) * yhat2
            combo_rule = "weighted_by_inv_MASE12"
        else:
            combo = 0.5 * yhat1 + 0.5 * yhat2
            combo_rule = "mean50_50"
    else:
        combo = yhat1.copy()
        combo_rule = "single_top1"

    combo = np.maximum(combo, 0.0)

    # Si algún modelo falló de forma silenciosa (NaNs, longitud), hacemos fallback a SNaive12
    if (len(yhat1) != H) or np.isnan(combo).any():
        log5(f"AVISO: predicciones inválidas en {(bid, ctype)}. Aplicamos fallback SNaive12.")
        combo = _baseline_snaive12(y_tr, d_tr, H)
        combo = np.maximum(combo, 0.0)
        top1 = "SNaive12"; top2 = None; yhat1 = combo.copy(); yhat2 = None
        combo_rule = "fallback_snaive12"

    # Volcamos 12 filas
    for i, d in enumerate(FUTURE_2024):
        rows_out.append({
            PAIR_COLS[0]: bid, PAIR_COLS[1]: ctype, "FECHA": d,
            "yhat_modelo1": float(yhat1[i]),
            "yhat_modelo2": (float(yhat2[i]) if isinstance(yhat2, np.ndarray) else np.nan),
            "yhat_combo": float(combo[i]),
            "fallback_flag": 0 if combo_rule in ["mean50_50","weighted_by_inv_MASE12","single_top1"] else 1,
            "route": route,
            "modelo_top1": top1,
            "modelo_top2": top2,
            "combo_rule": combo_rule
        })

# ----------------------------------------
# Bloque 5 — Guardado de resultados
# ----------------------------------------

# Construimos DataFrame de salida y aseguramos orden de columnas
cols_out = [
    PAIR_COLS[0], PAIR_COLS[1], "FECHA",
    "yhat_modelo1", "yhat_modelo2", "yhat_combo",
    "fallback_flag", "route", "modelo_top1", "modelo_top2", "combo_rule"
]

preds_df = pd.DataFrame(rows_out)
if not preds_df.empty:
    preds_df = preds_df[cols_out].sort_values(PAIR_COLS + ["FECHA"]).reset_index(drop=True)

OUT_PREDS = os.path.join(RUTA_RESULTADOS, "preds_por_serie_2024.csv")
preds_df.to_csv(OUT_PREDS, sep=CSV_SEP, index=False)
log5(f"Guardado: {OUT_PREDS} con {len(preds_df)} filas.")

# Guardamos alertas cero→actividad si existen
if len(alertas_rows) > 0:
    alertas_df = pd.DataFrame(alertas_rows)[PAIR_COLS + ["FECHA","yhat_combo","y_true_2024","rule"]]
    OUT_ALERT = os.path.join(RUTA_METRICAS, "alertas_cero_estructural_2024.csv")
    alertas_df.to_csv(OUT_ALERT, sep=CSV_SEP, index=False)
    log5(f"Guardado: {OUT_ALERT} con {len(alertas_df)} alertas.")
else:
    log5("No se registraron alertas cero→actividad en 2024.")

# (Opcional) Resumen por route y % fallback
if not preds_df.empty:
    resumen = (preds_df
               .groupby("route", as_index=False)
               .agg(n_rows=("FECHA","count"),
                    n_pairs=(PAIR_COLS[0], "nunique"))
               )
    # Calculamos %fallback por route
    fb = preds_df.groupby("route", as_index=False)["fallback_flag"].mean().rename(columns={"fallback_flag":"fallback_rate"})
    resumen = resumen.merge(fb, on="route", how="left")
    OUT_RES = os.path.join(RUTA_METRICAS, "preds_resumen_2024.csv")
    resumen.to_csv(OUT_RES, sep=CSV_SEP, index=False)
    log5(f"Guardado: {OUT_RES}")

# ----------------------------------------
# Bloque 6 — Punto de control E
# ----------------------------------------

# Validamos cobertura: nº parejas y 12 filas por pareja
pair_counts = preds_df.groupby(PAIR_COLS)["FECHA"].nunique().reset_index(name="n_meses")
missing_12  = pair_counts[pair_counts["n_meses"] != 12]

pred_pairs  = pair_counts.shape[0]
expected_pairs = panel_pairs[PAIR_COLS].drop_duplicates().shape[0]
missing_pairs  = expected_pairs - pred_pairs

fb_global = float(preds_df["fallback_flag"].mean()) if len(preds_df) else 0.0
fb_by_route = (preds_df.groupby("route")["fallback_flag"].mean().reset_index() if len(preds_df) else pd.DataFrame(columns=["route","fallback_flag"]))

log5(f"Punto E — parejas predichas={pred_pairs} / esperadas={expected_pairs} | missing_pairs={missing_pairs}")
log5(f"Punto E — %fallback global={fb_global:.2%}")
if not fb_by_route.empty:
    for _, r in fb_by_route.iterrows():
        log5(f"Punto E — %fallback route={r['route']}: {float(r['fallback_flag']):.2%}")

if not missing_12.empty:
    log5(f"Punto E — parejas sin 12 meses: {len(missing_12)}")
    # Mostramos hasta 10 para inspección
    log5("Ejemplos sin 12 meses (hasta 10): " + ", ".join(map(str, missing_12[PAIR_COLS].head(10).values.tolist())))
else:
    log5("Punto E — todas las parejas tienen 12 meses (enero–diciembre 2024).")

log5("Paso 5 finalizado.")

[2025-09-21T23:11:07] Iniciamos Paso 5 — Ensembles y predicción 2024.
[2025-09-21T23:11:07] Población Paso 5 — total=2429 | INTERMITENTE=1397 | GENERAL=1032 | all_zero=104
[2025-09-21T23:14:35] Guardado: /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3/RESULTADOS/preds_por_serie_2024.csv con 29148 filas.
[2025-09-21T23:14:35] Guardado: /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3/METRICAS/alertas_cero_estructural_2024.csv con 73 alertas.
[2025-09-21T23:14:35] Guardado: /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3/METRICAS/preds_resumen_2024.csv
[2025-09-21T23:14:35] Punto E — parejas predichas=2429 / esperadas=2429 | missing_pairs=0
[2025-09-21T23:14:35] Punto E — %fallback global=34.79%
[2025-09-21T23:14:35] Punto E — %fallback route=GENERAL: 0.00%
[2025-09-21T23:14:35] Punto E — %fallback route=INTERMITENTE: 60.49%
[2025-09-21T23:14:35] Punto E — todas las parejas tienen 12 meses (enero–diciembre 2024).
[2025-09-21T23:14:35] Paso 5 finalizado.

# ============================================================
# ESTRATEGIA 3 — PASO 6
# Comparación contra SNaive12 y consolidación de modelo_final por serie
# ============================================================
# En este paso:
#  - Recuperamos MASE12 de SNaive12 en 2024
#  - Comparamos con el top-1 seleccionado en Paso 4
#  - Calculamos delta y winner flag
#  - Aplicamos la regla de reemplazo por SNaive12 si ningún candidato mejora
#  - Respetamos casos especiales (cero estructural y solo test)
#  - Guardamos tablas finales y dejamos trazas en un log
# ============================================================

# Fijamos semilla global para reproducibilidad de cualquier aleatoriedad interna.
np.random.seed(7)


# =========================
# Bloque 1) Parámetros y rutas
# =========================
# Reutilizamos variables globales si existen; si no, definimos valores por defecto seguros.
ruta_base_3 = globals().get("ruta_base_3", "/content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3")
CSV_SEP     = globals().get("CSV_SEP", ";")
PAIR_COLS   = globals().get("PAIR_COLS", ["ID_BUILDING","FM_COST_TYPE"])
DATECOL     = globals().get("DATECOL", "FECHA")

RUTA_METRICAS = os.path.join(ruta_base_3, "METRICAS")
RUTA_RESULT   = os.path.join(ruta_base_3, "RESULTADOS")
RUTA_LOGS     = os.path.join(ruta_base_3, "LOGS")

os.makedirs(RUTA_METRICAS, exist_ok=True)
os.makedirs(RUTA_RESULT,   exist_ok=True)
os.makedirs(RUTA_LOGS,     exist_ok=True)

LOG_PATH_STEP6 = os.path.join(RUTA_LOGS, "estrategia3_step6.log")

# Ficheros de entrada esperados
PATH_SEL_CV   = os.path.join(RUTA_METRICAS, "seleccion_modelos_cv.csv")
PATH_ROLL_AGG = os.path.join(RUTA_METRICAS, "rolling_metrics_agg.csv")
PATH_BASE     = os.path.join(RUTA_METRICAS, "metrics_baselines_cv.csv")
PATH_SEG_ENR  = os.path.join(RUTA_METRICAS, "segmentation_intermitencia_step3.csv")
PATH_SOLOTEST = os.path.join(RUTA_METRICAS, "pairs_solo_test.csv")              # opcional
PATH_ALERTAS  = os.path.join(RUTA_METRICAS, "alertas_cero_estructural_2024.csv")# opcional del Paso 5

# Ficheros de salida de este paso
PATH_SNAIVE_VS_TOP1 = os.path.join(RUTA_METRICAS, "paso6_snaive_vs_top1.csv")
PATH_MODELO_FINAL   = os.path.join(RUTA_METRICAS, "modelo_final_paso6.csv")

# =========================
# Bloque 2) Funciones auxiliares
# =========================
def log6(msg: str):
    """Escribimos mensajes del Paso 6 en el log y también en consola para trazabilidad."""
    stamp = datetime.now().isoformat(timespec="seconds")
    line = f"[{stamp}] {msg}"
    print(line)
    with open(LOG_PATH_STEP6, "a", encoding="utf-8") as f:
        f.write(line + "\n")


def _read_csv(path: str, required: bool = True, sep: str = CSV_SEP) -> pd.DataFrame:
    """Leemos un CSV con el separador del proyecto. Si no existe y es opcional, devolvemos DF vacío."""
    if os.path.exists(path):
        return pd.read_csv(path, sep=sep)
    if required:
        raise FileNotFoundError(f"Falta el archivo requerido: {path}")
    log6(f"AVISO: no encontramos {os.path.basename(path)} (continuamos sin este artefacto).")
    return pd.DataFrame()


def _to_float_safe(s):
    """Convertimos una serie a float de forma robusta."""
    return pd.to_numeric(s, errors="coerce")


def _is_baseline_name(name: str) -> bool:
    """Determinamos si un nombre de modelo es baseline para excluirlo al elegir mejor_no_baseline."""
    if not isinstance(name, str):
        return False
    s = name.strip().lower()
    baselines = {
        "naive1", "snaive12", "medianaestacional", "medianaestac",
        "baseline_cero", "perfilescaso", "perfil_escaso", "perfilsparso",
        "perfil_escaso_uniforme"
    }
    return s in baselines


def _mejor_no_baseline(roll_agg_pair: pd.DataFrame) -> str | None:
    """Elegimos el mejor candidato no-baseline por MASE12_mean (desempate WAPE_mean)."""
    if roll_agg_pair.empty:
        return None
    df = roll_agg_pair.copy()
    df = df[~df["model_name"].apply(_is_baseline_name)]
    if df.empty:
        return None
    df = df.sort_values(["MASE12_mean", "WAPE_mean"], ascending=[True, True])
    return str(df.iloc[0]["model_name"])


def _min_mase12_no_baseline(roll_agg_pair: pd.DataFrame) -> float:
    """Calculamos el mínimo MASE12_mean entre no-baselines; devolvemos NaN si no hay candidatos."""
    df = roll_agg_pair.copy()
    df = df[~df["model_name"].apply(_is_baseline_name)]
    if df.empty:
        return np.nan
    return float(df["MASE12_mean"].min())


def _attach_route(sel_df: pd.DataFrame, seg_df: pd.DataFrame) -> pd.DataFrame:
    """Nos aseguramos de tener la columna route en la selección (si no está, la inferimos de segment)."""
    has_route = "route" in sel_df.columns
    if has_route:
        return sel_df
    # Si no estuviera, inferimos desde la segmentación (segment GENERAL/INTERMITENTE).
    aux = seg_df[PAIR_COLS + ["segment"]].drop_duplicates().rename(columns={"segment": "route"})
    out = sel_df.merge(aux, on=PAIR_COLS, how="left")
    return out

# =========================
# Bloque 3) Carga de artefactos
# =========================
log6("Iniciamos PASO 6 — Carga de artefactos.")

sel = _read_csv(PATH_SEL_CV, required=True)
roll = _read_csv(PATH_ROLL_AGG, required=True)
base = _read_csv(PATH_BASE, required=True)
seg  = _read_csv(PATH_SEG_ENR, required=True)

# Cargas opcionales
pairs_solo_test = _read_csv(PATH_SOLOTEST, required=False)
alertas_df      = _read_csv(PATH_ALERTAS, required=False)

# Tipamos métricas clave
for col in ["MAE_mean","WAPE_mean","SMAPE_mean","MASE1_mean","MASE12_mean"]:
    if col in roll.columns:
        roll[col] = _to_float_safe(roll[col])

for col in ["MAE","WAPE","SMAPE","MASE1","MASE12","MASE12_top1","MASE12_top2"]:
    if col in sel.columns:
        sel[col] = _to_float_safe(sel[col])

# Aseguramos que selección tenga route (si viniera sin ella)
sel = _attach_route(sel, seg)

# =========================
# Bloque 4) Recuperar SNaive12 (2024) y unir con selección
# =========================
# Extraemos SNaive12 desde metrics_baselines_cv.csv (Paso 1)
snaive = base.copy()
if "baseline" not in snaive.columns:
    raise ValueError("El archivo metrics_baselines_cv.csv no tiene columna 'baseline'.")
snaive = snaive[snaive["baseline"].str.strip().str.lower() == "snaive12"]

# Nos quedamos con la métrica MASE12 como referencia 2024 para la comparación
snaive = snaive[PAIR_COLS + ["MASE12"]].rename(columns={"MASE12": "MASE12_SNaive12_2024"})
snaive["MASE12_SNaive12_2024"] = _to_float_safe(snaive["MASE12_SNaive12_2024"])

# Unimos con la selección del Paso 4
merged = sel.merge(snaive, on=PAIR_COLS, how="left")

# Calculamos delta y winner flag (si no hay SNaive12_2024, el winner queda en 0 por cautela)
merged["delta_MASE12_vs_SNaive12"] = _to_float_safe(merged["MASE12_top1"]) - _to_float_safe(merged["MASE12_SNaive12_2024"])
merged["winner_beats_snaive12"] = ((merged["delta_MASE12_vs_SNaive12"] < 0).astype(int)).fillna(0).astype(int)
merged["regla_aplicada"] = "mantener_top1"

# Dejamos tabla base para la salida 1 (se actualizará con la regla de reemplazo si aplica)
cols_out_step6 = [
    *PAIR_COLS, "route", "modelo_top1", "modelo_top2",
    "MASE12_top1", "MASE12_SNaive12_2024", "delta_MASE12_vs_SNaive12",
    "winner_beats_snaive12", "regla_aplicada"
]
for c in ["MASE12_top1","MASE12_SNaive12_2024","delta_MASE12_vs_SNaive12"]:
    if c in merged.columns:
        merged[c] = _to_float_safe(merged[c])

# =========================
# Bloque 5) Aplicación de reglas por pareja
# =========================
log6("Aplicamos reglas de decisión por pareja (cero estructural, solo test y reemplazo por SNaive12).")

# Preparamos índices para acceso rápido
roll_idx = roll.set_index(PAIR_COLS, drop=False)
seg_idx  = seg.set_index(PAIR_COLS,  drop=False)

solo_test_set = set(map(tuple, pairs_solo_test[PAIR_COLS].drop_duplicates().values.tolist())) if not pairs_solo_test.empty else set()

# Vamos a construir la decisión final por pareja
final_rows = []

# También iremos actualizando 'merged' para reflejar si forzamos SNaive12 en modelo_top1 (regla_aplicada)
merged = merged.copy()

for i, r in merged.iterrows():
    bid, ctype = r[PAIR_COLS[0]], r[PAIR_COLS[1]]
    route      = r.get("route", "GENERAL")
    top1       = r.get("modelo_top1", None)
    top2       = r.get("modelo_top2", None)

    # Recuperamos bandera cero estructural
    flag_cero = 0
    try:
        flag_cero = int(seg_idx.loc[(bid, ctype)].get("flag_all_zero_train", 0))
    except Exception:
        flag_cero = 0

    # Caso 1: cero estructural (prioridad máxima)
    if flag_cero == 1:
        final_rows.append({
            PAIR_COLS[0]: bid, PAIR_COLS[1]: ctype, "route": route,
            "modelo_final": "baseline_cero",
            "motivo_final": "cero_estructural",
            "modelo_top2_final": None,
            "referencia_SNaive12": float(r.get("MASE12_SNaive12_2024")) if pd.notna(r.get("MASE12_SNaive12_2024")) else np.nan,
            "delta_MASE12_vs_SNaive12": float(r.get("delta_MASE12_vs_SNaive12")) if pd.notna(r.get("delta_MASE12_vs_SNaive12")) else np.nan
        })
        # No cambiamos merged['modelo_top1'] en este caso; el resultado final manda en modelo_final
        continue

    # Caso 2: SOLO TEST (sin histórico 2021–2023)
    is_solo_test = (bid, ctype) in solo_test_set
    if is_solo_test:
        final_rows.append({
            PAIR_COLS[0]: bid, PAIR_COLS[1]: ctype, "route": route,
            "modelo_final": "SNaive12",
            "motivo_final": "solo_test",
            "modelo_top2_final": None,
            "referencia_SNaive12": float(r.get("MASE12_SNaive12_2024")) if pd.notna(r.get("MASE12_SNaive12_2024")) else np.nan,
            "delta_MASE12_vs_SNaive12": float(r.get("delta_MASE12_vs_SNaive12")) if pd.notna(r.get("delta_MASE12_vs_SNaive12")) else np.nan
        })
        continue

    # Caso 3: comparación y posible reemplazo por SNaive12
    ref_snaive = r.get("MASE12_SNaive12_2024")
    if pd.isna(ref_snaive):
        # Si no tenemos SNaive12_2024, mantenemos top1 por prudencia y lo dejamos registrado
        final_rows.append({
            PAIR_COLS[0]: bid, PAIR_COLS[1]: ctype, "route": route,
            "modelo_final": top1,
            "motivo_final": "sin_snaive12_2024",
            "modelo_top2_final": top2,
            "referencia_SNaive12": np.nan,
            "delta_MASE12_vs_SNaive12": np.nan
        })
        merged.loc[i, "regla_aplicada"] = "sin_snaive12_2024"
        continue

    # Recuperamos performance agregada rolling para esta pareja
    try:
        roll_pair = roll_idx.loc[(bid, ctype)]
        roll_pair = roll_pair if isinstance(roll_pair, pd.DataFrame) else pd.DataFrame([roll_pair])
    except KeyError:
        roll_pair = pd.DataFrame(columns=roll.columns)

    # Calculamos si algún candidato NO-BASELINE mejora a SNaive12 (usamos MASE12_mean de CV)
    min_mase_nb = _min_mase12_no_baseline(roll_pair)
    mejora_existente = (pd.notna(min_mase_nb) and min_mase_nb < float(ref_snaive))

    if mejora_existente:
        # Conservamos el top-1 de la selección y registramos el motivo según delta
        motivo = "mejora_vs_SNaive12" if int(r.get("winner_beats_snaive12", 0)) == 1 else "mantener_top1"
        final_rows.append({
            PAIR_COLS[0]: bid, PAIR_COLS[1]: ctype, "route": route,
            "modelo_final": top1,
            "motivo_final": motivo,
            "modelo_top2_final": top2,
            "referencia_SNaive12": float(ref_snaive),
            "delta_MASE12_vs_SNaive12": float(r.get("delta_MASE12_vs_SNaive12")) if pd.notna(r.get("delta_MASE12_vs_SNaive12")) else np.nan
        })
    else:
        # Ningún candidato mejora: forzamos SNaive12 como top1 y elegimos mejor_no_baseline como top2_final
        mejor_nb = _mejor_no_baseline(roll_pair)
        merged.loc[i, "modelo_top1"] = "SNaive12"
        merged.loc[i, "regla_aplicada"] = "reemplazo_por_SNaive12"

        final_rows.append({
            PAIR_COLS[0]: bid, PAIR_COLS[1]: ctype, "route": route,
            "modelo_final": "SNaive12",
            "motivo_final": "reemplazo_por_SNaive12",
            "modelo_top2_final": mejor_nb,
            "referencia_SNaive12": float(ref_snaive),
            "delta_MASE12_vs_SNaive12": float(r.get("delta_MASE12_vs_SNaive12")) if pd.notna(r.get("delta_MASE12_vs_SNaive12")) else np.nan
        })

# =========================
# Bloque 6) Guardado de salidas
# =========================
# Salida 1: comparación top1 vs SNaive12 con regla aplicada
paso6_df = merged[cols_out_step6].copy()
paso6_df.to_csv(PATH_SNAIVE_VS_TOP1, sep=CSV_SEP, index=False)

# Salida 2: modelo final por pareja
modelo_final_df = pd.DataFrame(final_rows)[
    PAIR_COLS + ["route","modelo_final","motivo_final","modelo_top2_final","referencia_SNaive12","delta_MASE12_vs_SNaive12"]
].copy()
modelo_final_df = modelo_final_df.sort_values(PAIR_COLS).reset_index(drop=True)
modelo_final_df.to_csv(PATH_MODELO_FINAL, sep=CSV_SEP, index=False)

log6(f"Guardado: {PATH_SNAIVE_VS_TOP1} (comparación y reglas aplicadas)")
log6(f"Guardado: {PATH_MODELO_FINAL} (modelo_final por pareja)")

# =========================
# Bloque 7) Punto de control F — métricas y log
# =========================
n_pairs_total = modelo_final_df[PAIR_COLS].drop_duplicates().shape[0]
n_zero        = (modelo_final_df["motivo_final"] == "cero_estructural").sum()
n_solo_test   = (modelo_final_df["motivo_final"] == "solo_test").sum()
n_reemplazos  = (modelo_final_df["motivo_final"] == "reemplazo_por_SNaive12").sum()

# % donde top1 supera a SNaive12 (solo cuando tenemos referencia y no casos especiales)
work = paso6_df.merge(modelo_final_df[PAIR_COLS + ["motivo_final"]], on=PAIR_COLS, how="left")
mask_eval = (~work["MASE12_SNaive12_2024"].isna()) & (~work["motivo_final"].isin(["cero_estructural","solo_test"]))
pct_winners = float(work.loc[mask_eval, "winner_beats_snaive12"].mean()) if mask_eval.sum() > 0 else np.nan

# Nº de alertas cero→actividad (si existe el artefacto del Paso 5)
n_alertas = int(alertas_df.shape[0]) if not alertas_df.empty else 0

log6(f"Punto F — parejas totales en modelo_final: {n_pairs_total}")
log6(f"Punto F — % top1 que supera a SNaive12 (en pares evaluables): {('%.2f' % (pct_winners*100)) if not pd.isna(pct_winners) else 'NA'}%")
log6(f"Punto F — nº reemplazos por SNaive12: {n_reemplazos}")
log6(f"Punto F — nº cero estructural: {n_zero}")
log6(f"Punto F — nº solo test: {n_solo_test}")
log6(f"Punto F — nº alertas cero→actividad (2024): {n_alertas}")

# Ejemplos (hasta 10) por categoría para auditoría
def _ejemplos(df: pd.DataFrame, mask, cols_show, titulo: str):
    sub = df.loc[mask, cols_show].head(10)
    if not sub.empty:
        log6(f"Ejemplos — {titulo}:")
        for _, row in sub.iterrows():
            log6("  " + ", ".join([f"{c}={row[c]}" for c in cols_show]))

cols_show_pairs = PAIR_COLS + ["route"]

_ejemplos(modelo_final_df, modelo_final_df["motivo_final"]=="reemplazo_por_SNaive12", cols_show_pairs+["modelo_final","modelo_top2_final"], "reemplazos por SNaive12")
_ejemplos(modelo_final_df, modelo_final_df["motivo_final"]=="cero_estructural",           cols_show_pairs+["modelo_final"],              "cero estructural")
_ejemplos(modelo_final_df, modelo_final_df["motivo_final"]=="solo_test",                  cols_show_pairs+["modelo_final"],              "solo test")
_ejemplos(work,             mask_eval & (work["winner_beats_snaive12"]==1),               cols_show_pairs+["winner_beats_snaive12"],     "top1 supera a SNaive12")

print("\n== Paso 6 completado ==")
print(f"Comparación guardada en: {PATH_SNAIVE_VS_TOP1}")
print(f"Modelo final guardado en: {PATH_MODELO_FINAL}")
print(f"Log: {LOG_PATH_STEP6}")

[2025-09-21T23:33:35] Iniciamos PASO 6 — Carga de artefactos.
[2025-09-21T23:33:35] Aplicamos reglas de decisión por pareja (cero estructural, solo test y reemplazo por SNaive12).
[2025-09-21T23:33:39] Guardado: /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3/METRICAS/paso6_snaive_vs_top1.csv (comparación y reglas aplicadas)
[2025-09-21T23:33:39] Guardado: /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3/METRICAS/modelo_final_paso6.csv (modelo_final por pareja)
[2025-09-21T23:33:39] Punto F — parejas totales en modelo_final: 2429
[2025-09-21T23:33:39] Punto F — % top1 que supera a SNaive12 (en pares evaluables): 78.07%
[2025-09-21T23:33:39] Punto F — nº reemplazos por SNaive12: 660
[2025-09-21T23:33:39] Punto F — nº cero estructural: 104
[2025-09-21T23:33:39] Punto F — nº solo test: 0
[2025-09-21T23:33:39] Punto F — nº alertas cero→actividad (2024): 73
[2025-09-21T23:33:39] Ejemplos — reemplazos por SNaive12:
[2025-09-21T23:33:39]   ID_BUILDING=2, FM_COST_TYPE=Licencias, route=INTERMITENTE, modelo_final=SNaive12, modelo_top2_final=ADIDA_MAPA
[2025-09-21T23:33:39]   ID_BUILDING=2, FM_COST_TYPE=Obras, route=INTERMITENTE, modelo_final=SNaive12, modelo_top2_final=TSB
[2025-09-21T23:33:39]   ID_BUILDING=59, FM_COST_TYPE=Servicios Ctto., route=INTERMITENTE, modelo_final=SNaive12, modelo_top2_final=ADIDA_MAPA
[2025-09-21T23:33:39]   ID_BUILDING=74, FM_COST_TYPE=Servicios Ctto., route=INTERMITENTE, modelo_final=SNaive12, modelo_top2_final=TSB
[2025-09-21T23:33:39]   ID_BUILDING=74, FM_COST_TYPE=Servicios Extra, route=INTERMITENTE, modelo_final=SNaive12, modelo_top2_final=TSB
[2025-09-21T23:33:39]   ID_BUILDING=104, FM_COST_TYPE=Suministros, route=GENERAL, modelo_final=SNaive12, modelo_top2_final=Theta
[2025-09-21T23:33:39]   ID_BUILDING=105, FM_COST_TYPE=Servicios Extra, route=INTERMITENTE, modelo_final=SNaive12, modelo_top2_final=SBA
[2025-09-21T23:33:39]   ID_BUILDING=117, FM_COST_TYPE=Licencias, route=INTERMITENTE, modelo_final=SNaive12, modelo_top2_final=TSB
[2025-09-21T23:33:39]   ID_BUILDING=117, FM_COST_TYPE=Mtto. Correctivo, route=GENERAL, modelo_final=SNaive12, modelo_top2_final=Theta
[2025-09-21T23:33:39]   ID_BUILDING=121, FM_COST_TYPE=Mtto. Correctivo, route=GENERAL, modelo_final=SNaive12, modelo_top2_final=Holt_damped
[2025-09-21T23:33:39] Ejemplos — cero estructural:
[2025-09-21T23:33:39]   ID_BUILDING=2, FM_COST_TYPE=Eficiencia Energética, route=INTERMITENTE, modelo_final=baseline_cero
[2025-09-21T23:33:39]   ID_BUILDING=9, FM_COST_TYPE=Eficiencia Energética, route=INTERMITENTE, modelo_final=baseline_cero
[2025-09-21T23:33:39]   ID_BUILDING=18, FM_COST_TYPE=Eficiencia Energética, route=INTERMITENTE, modelo_final=baseline_cero
[2025-09-21T23:33:39]   ID_BUILDING=59, FM_COST_TYPE=Obras, route=INTERMITENTE, modelo_final=baseline_cero
[2025-09-21T23:33:39]   ID_BUILDING=62, FM_COST_TYPE=Licencias, route=INTERMITENTE, modelo_final=baseline_cero
[2025-09-21T23:33:39]   ID_BUILDING=74, FM_COST_TYPE=Licencias, route=INTERMITENTE, modelo_final=baseline_cero
[2025-09-21T23:33:39]   ID_BUILDING=127, FM_COST_TYPE=Servicios Extra, route=INTERMITENTE, modelo_final=baseline_cero
[2025-09-21T23:33:39]   ID_BUILDING=129, FM_COST_TYPE=Eficiencia Energética, route=INTERMITENTE, modelo_final=baseline_cero
[2025-09-21T23:33:39]   ID_BUILDING=137, FM_COST_TYPE=Obras, route=INTERMITENTE, modelo_final=baseline_cero
[2025-09-21T23:33:39]   ID_BUILDING=146, FM_COST_TYPE=Eficiencia Energética, route=INTERMITENTE, modelo_final=baseline_cero
[2025-09-21T23:33:39] Ejemplos — top1 supera a SNaive12:
[2025-09-21T23:33:39]   ID_BUILDING=2, FM_COST_TYPE=Licencias, route=INTERMITENTE, winner_beats_snaive12=1
[2025-09-21T23:33:39]   ID_BUILDING=2, FM_COST_TYPE=Mtto. Contratos, route=GENERAL, winner_beats_snaive12=1
[2025-09-21T23:33:39]   ID_BUILDING=2, FM_COST_TYPE=Mtto. Correctivo, route=GENERAL, winner_beats_snaive12=1
[2025-09-21T23:33:39]   ID_BUILDING=2, FM_COST_TYPE=Obras, route=INTERMITENTE, winner_beats_snaive12=1
[2025-09-21T23:33:39]   ID_BUILDING=2, FM_COST_TYPE=Servicios Ctto., route=GENERAL, winner_beats_snaive12=1
[2025-09-21T23:33:39]   ID_BUILDING=2, FM_COST_TYPE=Servicios Extra, route=INTERMITENTE, winner_beats_snaive12=1
[2025-09-21T23:33:39]   ID_BUILDING=2, FM_COST_TYPE=Suministros, route=GENERAL, winner_beats_snaive12=1
[2025-09-21T23:33:39]   ID_BUILDING=9, FM_COST_TYPE=Licencias, route=INTERMITENTE, winner_beats_snaive12=1
[2025-09-21T23:33:39]   ID_BUILDING=9, FM_COST_TYPE=Mtto. Contratos, route=GENERAL, winner_beats_snaive12=1
[2025-09-21T23:33:39]   ID_BUILDING=9, FM_COST_TYPE=Mtto. Correctivo, route=GENERAL, winner_beats_snaive12=1

== Paso 6 completado ==
Comparación guardada en: /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3/METRICAS/paso6_snaive_vs_top1.csv
Modelo final guardado en: /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3/METRICAS/modelo_final_paso6.csv
Log: /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3/LOGS/estrategia3_step6.log

# ===============================================================
# Paso 7 · Interfaces de modelos (mínimo viable)
# Implementamos un módulo autocontenido con wrappers de forecasting
# y una firma común: fit_predict(model_spec, y_train, dates_train, horizon, ...)
# En todo momento priorizamos robustez, no negatividad y reproducibilidad.
# ===============================================================

# Fijamos semilla global para reproducibilidad de cualquier aleatoriedad interna.
np.random.seed(7)


# ===============================================================
# Bloque 1) Constantes y utilidades generales
# ===============================================================

# Definimos un pequeño número epsilon para evitar divisiones por cero.
_EPS = 1e-12

def _to_numpy_1d(y):
    # Normalizamos a vector 1D float y saneamos NaNs/inf a 0.0
    if y is None:
        return np.zeros(0, dtype=float)
    arr = np.asarray(y, dtype=float).reshape(-1,)
    # Saneamos valores no finitos
    mask = np.isfinite(arr)
    if not mask.all():
        arr = np.where(mask, arr, 0.0)
    return arr



def _to_ms_dates(dates) -> np.ndarray:
    """Normalizamos las fechas a inicio de mes (MS) como pandas.Timestamp."""
    if dates is None:
        return np.array([], dtype="datetime64[ns]")
    if isinstance(dates, (list, tuple, np.ndarray, pd.Series, pd.Index)):
        ds = pd.to_datetime(pd.Series(dates)).dt.to_period("M").dt.to_timestamp()
        return ds.values
    # Si no podemos interpretar, devolvemos array vacío (el flujo downstream lo gestionará).
    return np.array([], dtype="datetime64[ns]")


def _future_months(last_date: pd.Timestamp, horizon: int, freq: str = "MS") -> pd.DatetimeIndex:
    """Generamos los meses futuros desde last_date con frecuencia mensual."""
    if horizon <= 0:
        return pd.DatetimeIndex([], dtype="datetime64[ns]")
    last_ms = pd.to_datetime(last_date).to_period("M").to_timestamp()
    start = last_ms + pd.offsets.MonthBegin(1)
    return pd.date_range(start, periods=horizon, freq=freq)


def _is_all_zero(y: np.ndarray) -> bool:
    """Determinamos si toda la serie es cero."""
    return y.size > 0 and np.allclose(y, 0.0)


def _max_zero_run_length(y: np.ndarray) -> int:
    """Calculamos la racha máxima de ceros consecutivos en y."""
    if y.size == 0:
        return 0
    is_zero = (y == 0.0)
    if not np.any(is_zero):
        return 0
    # Contamos rachas consecutivas
    max_run = run = 0
    for v in is_zero:
        if v:
            run += 1
            max_run = max(max_run, run)
        else:
            run = 0
    return max_run


def _prop_zeros(y: np.ndarray) -> float:
    """Calculamos la proporción de ceros en la serie."""
    if y.size == 0:
        return 0.0
    return float((y == 0.0).sum()) / float(y.size)


def _month_number_index(dates: np.ndarray) -> np.ndarray:
    """Extraemos el número de mes (1..12) para cada fecha."""
    if dates.size == 0:
        return np.array([], dtype=int)
    return pd.to_datetime(dates).month.values


def _last_complete_year_map(y: np.ndarray, dates: np.ndarray) -> Dict[int, float]:
    """
    Construimos un mapa mes->valor del último año completo disponible.
    Si no tenemos 12 meses en el último año, probamos hacia atrás.
    Si no encontramos ningún año completo, devolvemos dict vacío.
    """
    if y.size == 0 or dates.size == 0:
        return {}
    df = pd.DataFrame({"FECHA": pd.to_datetime(dates), "y": y})
    df["YEAR"] = df["FECHA"].dt.year
    df["MONTH"] = df["FECHA"].dt.month
    years = list(sorted(df["YEAR"].unique(), reverse=True))
    for year in years:
        sub = df[df["YEAR"] == year]
        if sub["MONTH"].nunique() == 12:
            return sub.set_index("MONTH")["y"].to_dict()
    return {}


def _median_by_month(y: np.ndarray, dates: np.ndarray) -> Dict[int, float]:
    """Calculamos la mediana por mes del histórico."""
    if y.size == 0 or dates.size == 0:
        return {}
    df = pd.DataFrame({"FECHA": pd.to_datetime(dates), "y": y})
    df["MONTH"] = df["FECHA"].dt.month
    med = df.groupby("MONTH")["y"].median().to_dict()
    return med

# ===============================================================
# Bloque 2) Baselines (independientes de statsmodels)
# ===============================================================

def _baseline_cero(h: int) -> np.ndarray:
    """Pronóstico estructural cero."""
    return np.zeros(h, dtype=float)


def _baseline_naive1(y: np.ndarray, h: int) -> np.ndarray:
    """Repetimos el último valor del histórico."""
    if y.size == 0:
        return np.zeros(h, dtype=float)
    return np.repeat(float(y[-1]), h)


def _baseline_median_seasonal(y: np.ndarray, dates: np.ndarray, h: int) -> np.ndarray:
    """Proyectamos la mediana por mes del histórico."""
    if y.size == 0 or dates.size == 0:
        return np.zeros(h, dtype=float)
    med = _median_by_month(y, dates)
    last_date = pd.to_datetime(dates[-1])
    future = _future_months(last_date, h, "MS")
    out = np.zeros(h, dtype=float)
    last = float(y[-1]) if y.size > 0 else 0.0
    for i, d in enumerate(future):
        out[i] = med.get(d.month, last)
    return out


def _baseline_snaive12(y: np.ndarray, dates: np.ndarray, h: int) -> np.ndarray:
    """
    SNaive12 robusto: usamos el mismo mes del último año completo disponible.
    Si no hay año completo, caemos a mediana estacional; si falta algo, último valor.
    """
    if h <= 0:
        return np.zeros(0, dtype=float)
    if y.size == 0 or dates.size == 0:
        return np.zeros(h, dtype=float)

    # Intentamos mapa del último año completo.
    last_year_map = _last_complete_year_map(y, dates)
    last_val = float(y[-1])
    last_date = pd.to_datetime(dates[-1])
    future = _future_months(last_date, h, "MS")

    if len(last_year_map) == 12:
        # Proyectamos con el mapa; si falta alguna clave rara, usamos last_val.
        out = np.array([last_year_map.get(d.month, last_val) for d in future], dtype=float)
        return out

    # Sin año completo -> mediana estacional
    med = _median_by_month(y, dates)
    if len(med) > 0:
        out = np.array([med.get(d.month, last_val) for d in future], dtype=float)
        return out

    # Último *fallback*: último valor
    return np.repeat(last_val, h)


def _baseline_perfil_escaso_uniforme(y: np.ndarray, h: int) -> np.ndarray:
    """Promedio global replicado; si no hay datos, ceros."""
    if y.size == 0:
        return np.zeros(h, dtype=float)
    mu = float(np.mean(y))
    return np.repeat(mu, h)

# ===============================================================
# Bloque 3) Modelos clásicos (statsmodels si está; fallbacks si no)
# ===============================================================

def _theta_predict(y: np.ndarray, dates: np.ndarray, h: int) -> np.ndarray:
    """Theta(period=12); si falla o no existe, caemos a mediana estacional."""
    if h <= 0:
        return np.zeros(0, dtype=float)
    if y.size == 0:
        return np.zeros(h, dtype=float)
    if ThetaModel is None:
        warnings.warn("ThetaModel no disponible; usamos fallback median_seasonal.")
        return _baseline_median_seasonal(y, dates, h)
    try:
        tm = ThetaModel(np.asarray(y, dtype=float), period=12)
        res = tm.fit()
        f = np.asarray(res.forecast(h), dtype=float)
        return f
    except Exception as e:
        warnings.warn(f"ThetaModel falló ({e}); usamos fallback median_seasonal.")
        return _baseline_median_seasonal(y, dates, h)


def _ets_auto_predict(y: np.ndarray, h: int) -> np.ndarray:
    """ETS aditivo estacional y tendencia amortiguada; fallback naive1."""
    if h <= 0:
        return np.zeros(0, dtype=float)
    if y.size < 3 or ExponentialSmoothing is None:
        if ExponentialSmoothing is None:
            warnings.warn("ExponentialSmoothing no disponible; usamos fallback naive1.")
        return _baseline_naive1(y, h)
    try:
        model = ExponentialSmoothing(
            np.asarray(y, dtype=float),
            trend="add",
            seasonal="add",
            seasonal_periods=12,
            damped_trend=True,
            initialization_method="estimated",
        ).fit(optimized=True)
        f = np.asarray(model.forecast(h), dtype=float)
        return f
    except Exception as e:
        warnings.warn(f"ETS falló ({e}); usamos fallback naive1.")
        return _baseline_naive1(y, h)


def _holt_damped_predict(y: np.ndarray, h: int) -> np.ndarray:
    """Holt amortiguado sin estacionalidad; fallback naive1."""
    if h <= 0:
        return np.zeros(0, dtype=float)
    if y.size < 3 or ExponentialSmoothing is None:
        if ExponentialSmoothing is None:
            warnings.warn("ExponentialSmoothing no disponible; usamos fallback naive1.")
        return _baseline_naive1(y, h)
    try:
        model = ExponentialSmoothing(
            np.asarray(y, dtype=float),
            trend="add",
            seasonal=None,
            damped_trend=True,
            initialization_method="estimated",
        ).fit(optimized=True)
        f = np.asarray(model.forecast(h), dtype=float)
        return f
    except Exception as e:
        warnings.warn(f"Holt falló ({e}); usamos fallback naive1.")
        return _baseline_naive1(y, h)


def _arima_auto_light(y: np.ndarray, h: int) -> np.ndarray:
    """Grid ligero ARIMA (p,d,q)∈{0,1} sin estacionalidad; fallback naive1."""
    if h <= 0:
        return np.zeros(0, dtype=float)
    if SARIMAX is None or y.size < 8:
        if SARIMAX is None:
            warnings.warn("SARIMAX no disponible; usamos fallback naive1.")
        return _baseline_naive1(y, h)
    y_ = np.asarray(y, dtype=float)
    best_aic = np.inf
    best_forecast = None
    for p in [0, 1]:
        for d in [0, 1]:
            for q in [0, 1]:
                try:
                    mod = SARIMAX(
                        y_, order=(p, d, q), seasonal_order=(0, 0, 0, 0),
                        trend="c", enforce_stationarity=False, enforce_invertibility=False
                    )
                    res = mod.fit(disp=False)
                    if res.aic < best_aic:
                        best_aic = res.aic
                        best_forecast = np.asarray(res.forecast(h), dtype=float)
                except Exception:
                    continue
    if best_forecast is None:
        warnings.warn("ARIMA grid no encontró solución; usamos fallback naive1.")
        return _baseline_naive1(y, h)
    return best_forecast

# ===============================================================
# Bloque 4) Intermitentes (implementación directa)
# ===============================================================

def _croston_classic(y: np.ndarray, h: int, alpha: float = 0.1) -> np.ndarray:
    """
    Implementamos Croston clásico: suavizamos tamaño y periodicidad, y proyectamos z/p.
    """
    if h <= 0:
        return np.zeros(0, dtype=float)
    y = np.asarray(y, dtype=float)
    n = y.size
    if n == 0:
        return np.zeros(h, dtype=float)
    z, p, k = 0.0, 0.0, 0
    first = True
    for v in y:
        k += 1
        if v > 0:
            if first:
                z, p, first = v, float(k), False
            else:
                z = z + alpha * (v - z)
                p = p + alpha * (k - p)
            k = 0
    f = (z / p) if p > 0 else 0.0
    return np.repeat(float(f), h)


def _croston_sba(y: np.ndarray, h: int, alpha: float = 0.1) -> np.ndarray:
    """
    Variación SBA: corrección de sesgo multiplicando por (1 - alpha/2).
    """
    base = _croston_classic(y, h, alpha=alpha)
    return base * (1.0 - alpha / 2.0)


def _tsb(y: np.ndarray, h: int, alpha: float = 0.2, beta: float = 0.1) -> np.ndarray:
    """
    TSB: modelamos por separado probabilidad de demanda y tamaño medio de demanda.
    """
    if h <= 0:
        return np.zeros(0, dtype=float)
    y = np.asarray(y, dtype=float)
    if y.size == 0:
        return np.zeros(h, dtype=float)
    demand = (y > 0).astype(int)
    sizes = np.where(y > 0, y, 0.0)

    # Inicializamos con primer valor (o media de positivos si el primero es cero)
    p = float(demand[0])
    z = float(sizes[0] if sizes[0] > 0 else (np.mean(sizes[sizes > 0]) if np.any(sizes > 0) else 0.0))

    for t in range(1, y.size):
        p = p + alpha * (float(demand[t]) - p)
        if y[t] > 0:
            z = z + beta * (float(y[t]) - z)

    f = p * z
    return np.repeat(float(f), h)

# ===============================================================
# Bloque 5) ADIDA/MAPA (mínimo viable)
# ===============================================================

def _adida_mapa(y: np.ndarray,
                h: int,
                low_freq_models: Optional[List[str]] = None) -> np.ndarray:
    """
    Implementamos un ADIDA/MAPA mínimo:
      1) Agregamos a trimestral (sumas de 3 meses; rellenamos con 0 para cerrar bloque).
      2) Pronosticamos a baja frecuencia con ETS o Theta (fallback: último valor).
      3) Desagregamos con perfil histórico medio por posición (mes1, mes2, mes3).
    """
    if h <= 0:
        return np.zeros(0, dtype=float)
    y = np.asarray(y, dtype=float)
    n = y.size
    if n < 1:
        return np.zeros(h, dtype=float)

    # Agregación a trimestral (M3)
    pad = (3 - (n % 3)) % 3
    y_pad = np.hstack([y, np.zeros(pad, dtype=float)])
    blocks = y_pad.reshape(-1, 3)
    yQ = blocks.sum(axis=1)

    # Modelo low frequency (priorizamos ETS; si no, Theta; si no, último valor)
    if low_freq_models is None:
        low_freq_models = ["ETS_auto", "Theta"]

    fq_h = int(np.ceil(h / 3))
    if fq_h <= 0:
        return np.zeros(0, dtype=float)

    def _lf_forecast(yQ_arr: np.ndarray, fq_h: int) -> np.ndarray:
        # Intentamos ETS si está disponible y hay datos suficientes.
        if "ETS_auto" in low_freq_models and ExponentialSmoothing is not None and yQ_arr.size >= 3:
            try:
                mQ = ExponentialSmoothing(yQ_arr, trend="add", seasonal=None, damped_trend=True).fit(optimized=True)
                return np.asarray(mQ.forecast(fq_h), dtype=float)
            except Exception:
                pass
        # Theta como alternativa si está disponible.
        if "Theta" in low_freq_models and ThetaModel is not None and yQ_arr.size >= 1:
            try:
                tm = ThetaModel(yQ_arr)  # periodo no crítico en trimestral sin estacionalidad fuerte
                res = tm.fit()
                return np.asarray(res.forecast(fq_h), dtype=float)
            except Exception:
                pass
        # Último fallback: repetir el último trimestre observado (o ceros si vacío).
        last_q = float(yQ_arr[-1]) if yQ_arr.size > 0 else 0.0
        return np.repeat(last_q, fq_h)

    fq = _lf_forecast(yQ, fq_h)

    # Desagregación por perfil mensual medio dentro del trimestre
    totals = blocks.sum(axis=1, keepdims=True)
    # Evitamos divisiones por cero
    weights = np.divide(blocks, np.where(totals == 0.0, 1.0, totals))
    w = np.nan_to_num(weights.mean(axis=0))
    if float(np.sum(w)) <= _EPS:
        w = np.array([1/3, 1/3, 1/3], dtype=float)

    # Expandimos cada trimestre pronosticado en 3 meses
    monthly = []
    for qv in fq:
        monthly.extend(list(w * qv))
    monthly = np.asarray(monthly, dtype=float)[:h]
    return monthly

# ===============================================================
# Bloque 6) Router: mapeo por family/name + reglas de ruta
# ===============================================================

def _normalize_spec(spec: Dict[str, Any]) -> Dict[str, Any]:
    """Normalizamos el model_spec para trabajar con claves en minúscula y defaults."""
    spec = dict(spec or {})
    name = str(spec.get("name", "")).strip()
    family = str(spec.get("family", "")).strip().lower()
    params = dict(spec.get("params", {}) or {})

    # Alias de nombre → family si no se indicó
    lname = name.strip().lower()
    aliases = {
        "snaive12": ("baseline", {"type": "snaive12"}),
        "naive1": ("baseline", {"type": "naive1"}),
        "median_seasonal": ("baseline", {"type": "median_seasonal"}),
        "medianaestac": ("baseline", {"type": "median_seasonal"}),
        "cero": ("baseline", {"type": "cero"}),
        "perfil_escaso_uniforme": ("baseline", {"type": "perfil_escaso_uniforme"}),
        "ets_auto": ("ets", {}),
        "ets": ("ets", {}),
        "theta": ("theta", {}),
        "arima_auto": ("arima", {"grid": "light"}),
        "holt_damped": ("holt", {}),
        "croston": ("croston", {"variant": "classic"}),
        "sba": ("croston", {"variant": "sba"}),
        "tsb": ("tsb", {}),
        "adida_mapa": ("adida_mapa", {}),
        "adida": ("adida_mapa", {}),
        "mapa": ("adida_mapa", {}),
    }
    if not family and lname in aliases:
        fam, default_p = aliases[lname]
        family = fam
        # Mezclamos defaults con los params existentes sin pisar lo que ya venga
        for k, v in default_p.items():
            params.setdefault(k, v)

    # Defaults específicos
    if family == "croston":
        params.setdefault("variant", "classic")  # classic | sba
        params.setdefault("alpha", 0.1)
    if family == "tsb":
        params.setdefault("alpha", 0.2)
        params.setdefault("beta", 0.1)
    if family == "adida_mapa":
        params.setdefault("agg_level", "Q")
        params.setdefault("model_low_freq", ["ETS_auto", "Theta"])
        params.setdefault("disagg", "perfil_historico")
    if family == "baseline":
        params.setdefault("type", "naive1")

    spec["name"] = name
    spec["family"] = family
    spec["params"] = params
    return spec


def _route_family_with_rules(family: str,
                             params: Dict[str, Any],
                             y: np.ndarray) -> Tuple[str, Dict[str, Any]]:
    """
    Aplicamos reglas internas según intermitencia y fuerza:
    - Si route="INTERMITENTE" y force=False:
        * Evitamos modelos pesados (ets/arima/holt/theta) y vamos a TSB o Croston.
    - ADIDA/MAPA restringido si prop_zeros ≥ 0.9 (preferimos TSB).
    """
    route = str(params.get("route", "")).strip().upper()
    force = bool(params.get("force", False))

    pz = _prop_zeros(y)
    mz = _max_zero_run_length(y)

    # Regla de cero estructural viene en fit_predict; aquí no la duplicamos.

    if route == "INTERMITENTE" and not force:
        # Si se pidió un modelo pesado, lo redirigimos a TSB por defecto.
        if family in {"ets", "arima", "holt", "theta"}:
            new_family = "tsb"
            new_params = dict(params)
            warnings.warn("route=INTERMITENTE y force=False -> redirigimos a TSB.")
            return new_family, new_params
        # Si se pidió ADIDA/MAPA pero la intermitencia es extrema, redirigimos a TSB.
        if family == "adida_mapa" and pz >= 0.90:
            new_family = "tsb"
            new_params = dict(params)
            warnings.warn("INTERMITENTE con prop_zeros>=0.90 -> preferimos TSB sobre ADIDA/MAPA.")
            return new_family, new_params

    return family, params

# ===============================================================
# Bloque 7) API pública
# ===============================================================

def fit_predict(model_spec: Dict[str, Any],
                y_train,
                dates_train,
                horizon: int,
                freq: str = "MS",
                exog: Optional[Dict[str, Any]] = None) -> np.ndarray:
    """
    Devolvemos un np.ndarray de longitud = horizon, con no negatividad garantizada.
    Aplicamos reglas internas:
      - Cero estructural automático (36 ceros seguidos o todo-cero) -> baseline cero.
      - route=INTERMITENTE y force=False -> preferimos TSB/Croston; limitamos ADIDA/MAPA en intermitentes extremas.
    """
    # Normalizamos entradas
    h = int(horizon)
    if h <= 0:
        return np.zeros(0, dtype=float)

    y = _to_numpy_1d(y_train)
    dates = _to_ms_dates(dates_train)

    # Si no hay fechas, generamos un proxy equiespaciado mensual empezando en 2000-01
    # para que las funciones que requieren fechas tengan algo razonable.
    if dates.size == 0 and y.size > 0:
        start = pd.Timestamp("2000-01-01")
        dates = pd.date_range(start, periods=y.size, freq="MS").values

    # Reglas de cero estructural
    if _is_all_zero(y) or _max_zero_run_length(y) >= 36:
        # Registramos la decisión con un warning suave para trazabilidad.
        warnings.warn("Detectamos cero estructural (todo ceros o racha>=36); devolvemos baseline cero.")
        return _baseline_cero(h)

    # Normalizamos el spec y aplicamos reglas de ruta
    spec = _normalize_spec(model_spec or {})
    family = spec.get("family", "")
    params = dict(spec.get("params", {}) or {})
    family, params = _route_family_with_rules(family, params, y)

    # Enrutamos por familia
    try:
        if family == "baseline":
            btype = str(params.get("type", "naive1")).strip().lower()
            if btype == "cero":
                pred = _baseline_cero(h)
            elif btype in {"naive1", "naive"}:
                pred = _baseline_naive1(y, h)
            elif btype in {"snaive12", "seasonal_naive", "snaive"}:
                pred = _baseline_snaive12(y, dates, h)
            elif btype in {"median_seasonal", "mediana_estacional", "medianaestac"}:
                pred = _baseline_median_seasonal(y, dates, h)
            elif btype in {"perfil_escaso_uniforme", "perfil_escaso"}:
                pred = _baseline_perfil_escaso_uniforme(y, h)
            else:
                warnings.warn(f"Baseline desconocido '{btype}'; usamos naive1.")
                pred = _baseline_naive1(y, h)

        elif family == "theta":
            pred = _theta_predict(y, dates, h)

        elif family == "ets":
            pred = _ets_auto_predict(y, h)

        elif family == "holt":
            pred = _holt_damped_predict(y, h)

        elif family == "arima":
            # Permitimos 'grid'='light' en params, aunque no lo usamos explícitamente aquí.
            pred = _arima_auto_light(y, h)

        elif family == "croston":
            variant = str(params.get("variant", "classic")).strip().lower()
            alpha = float(params.get("alpha", 0.1))
            if variant == "sba":
                pred = _croston_sba(y, h, alpha=alpha)
            else:
                pred = _croston_classic(y, h, alpha=alpha)

        elif family == "tsb":
            alpha = float(params.get("alpha", 0.2))
            beta = float(params.get("beta", 0.1))
            pred = _tsb(y, h, alpha=alpha, beta=beta)

        elif family == "adida_mapa":
            # Regla conservadora adicional aquí por seguridad
            if _prop_zeros(y) >= 0.95:
                warnings.warn("Intermitencia extrema (prop_zeros>=0.95); usamos TSB en lugar de ADIDA/MAPA.")
                pred = _tsb(y, h, alpha=float(params.get("alpha", 0.2)), beta=float(params.get("beta", 0.1)))
            else:
                lf = params.get("model_low_freq") or ["ETS_auto", "Theta"]
                pred = _adida_mapa(y, h, low_freq_models=list(lf))

        else:
            warnings.warn(f"Familia desconocida '{family}'; usamos baseline naive1.")
            pred = _baseline_naive1(y, h)

    except Exception as e:
        # Si algo explota, caemos a un fallback por familia o naive1 como red de seguridad.
        warnings.warn(f"Error en familia '{family}': {e}. Aplicamos fallback.")
        if family == "theta":
            pred = _baseline_median_seasonal(y, dates, h)
        elif family in {"ets", "holt", "arima"}:
            pred = _baseline_naive1(y, h)
        elif family in {"croston", "tsb", "adida_mapa"}:
            pred = _baseline_perfil_escaso_uniforme(y, h)
        else:
            pred = _baseline_naive1(y, h)

    # Clipping de no negatividad como último paso.
    pred = np.asarray(pred, dtype=float)
    if pred.size != h:
        # Si por alguna razón la longitud no cuadra, ajustamos de forma robusta.
        if pred.size > h:
            pred = pred[:h]
        else:
            pad = np.repeat(float(pred[-1]) if pred.size > 0 else 0.0, h - pred.size)
            pred = np.hstack([pred, pad])
    pred = np.maximum(pred, 0.0)
    pred = np.where(np.isfinite(pred), pred, 0.0)
    return pred

# ===============================================================
# Bloque 8) Tests ligeros (ejecución directa del módulo)
# ===============================================================
if __name__ == "__main__":
    # Generamos datos sintéticos y verificamos forma, finitud y no negatividad.
    # Comentamos en primera persona lo que comprobamos.

    def _check(name: str, arr: np.ndarray, h: int):
        ok_shape = (arr.shape == (h,))
        ok_finite = np.all(np.isfinite(arr))
        ok_nneg = np.all(arr >= 0.0)
        print(f"[{name}] shape={arr.shape} finite={ok_finite} nneg={ok_nneg}")
        if not (ok_shape and ok_finite and ok_nneg):
            warnings.warn(f"El test '{name}' no cumple alguno de los checks básicos.")

    # Parámetros comunes de test
    H = 12
    T = 60  # ~5 años de datos mensuales
    dates = pd.date_range("2019-01-01", periods=T, freq="MS")

    # 1) Estacional suave
    rng = np.random.default_rng(7)
    t = np.arange(T)
    y_seasonal = 10.0 + 2.0 * np.sin(2 * np.pi * t / 12.0) + rng.normal(0, 0.3, size=T)
    y_seasonal = np.maximum(y_seasonal, 0.0)

    specs_seasonal = [
        {"name": "Theta", "family": "theta", "params": {}},
        {"name": "ETS_auto", "family": "ets", "params": {}},
        {"name": "SNaive12", "family": "baseline", "params": {"type": "snaive12"}},
        {"name": "MedianaEstac", "family": "baseline", "params": {"type": "median_seasonal"}},
    ]

    print("\n== Test: Estacional suave ==")
    for sp in specs_seasonal:
        yhat = fit_predict(sp, y_seasonal, dates, H)
        _check(sp["name"], yhat, H)

    # 2) Intermitente (muchos ceros + picos)
    y_inter = np.zeros(T, dtype=float)
    # Simulamos picos aleatorios esporádicos
    peak_idx = rng.choice(np.arange(T), size=int(T * 0.15), replace=False)
    y_inter[peak_idx] = rng.uniform(1.0, 10.0, size=peak_idx.size)

    specs_interm = [
        {"name": "Croston", "family": "croston", "params": {"variant": "classic", "alpha": 0.1}},
        {"name": "SBA", "family": "croston", "params": {"variant": "sba", "alpha": 0.1}},
        {"name": "TSB", "family": "tsb", "params": {"alpha": 0.2, "beta": 0.1}},
        {"name": "PerfilEscaso", "family": "baseline", "params": {"type": "perfil_escaso_uniforme"}},
    ]

    print("\n== Test: Intermitente ==")
    for sp in specs_interm:
        yhat = fit_predict(sp, y_inter, dates, H)
        _check(sp["name"], yhat, H)

    # 2b) Enrutado sensible a ruta (INTERMITENTE redirigiendo ETS -> TSB)
    sp_heavy_inter = {"name": "ETS_auto", "family": "ets", "params": {"route": "INTERMITENTE", "force": False}}
    print("\n== Test: Enrutado route=INTERMITENTE (redirigimos ETS->TSB) ==")
    yhat = fit_predict(sp_heavy_inter, y_inter, dates, H)
    _check("ETS->TSB route", yhat, H)

    # 3) Todo-cero (debería activar baseline cero si detectamos racha>=36)
    y_zero = np.zeros(T, dtype=float)
    specs_zero = [
        {"name": "Cero", "family": "baseline", "params": {"type": "cero"}},
        {"name": "SNaive12", "family": "baseline", "params": {"type": "snaive12"}},
        {"name": "MedianaEstac", "family": "baseline", "params": {"type": "median_seasonal"}},
    ]

    print("\n== Test: Todo-cero (activamos cero estructural) ==")
    for sp in specs_zero:
        yhat = fit_predict(sp, y_zero, dates, H)
        _check(sp["name"], yhat, H)

    # 4) ARIMA fallback (si SARIMAX no está, deberíamos caer a naive1 sin romper)
    sp_arima = {"name": "ARIMA_auto", "family": "arima", "params": {"grid": "light"}}
    print("\n== Test: ARIMA fallback ==")
    yhat = fit_predict(sp_arima, y_seasonal, dates, H)
    _check("ARIMA_auto", yhat, H)

    # 5) ADIDA/MAPA en serie moderadamente intermitente
    sp_adida = {"name": "ADIDA_MAPA", "family": "adida_mapa",
                "params": {"agg_level": "Q", "model_low_freq": ["ETS_auto", "Theta"], "disagg": "perfil_historico"}}
    print("\n== Test: ADIDA/MAPA ==")
    yhat = fit_predict(sp_adida, y_inter, dates, H)
    _check("ADIDA_MAPA", yhat, H)

    # Interpretamos brevemente algunos outputs.
    # - En estacional suave, esperamos que theta/ets se comporten razonablemente y que las baselines
    #   den resultados no negativos con forma correcta.
    # - En intermitentes, Croston/TSB deben devolver niveles coherentes con la frecuencia de picos,
    #   y el enrutado de ETS a TSB cuando route=INTERMITENTE debe funcionar sin error.
    # - En todo-cero, verificamos que detectamos cero estructural y devolvemos ceros de forma estable.
    # - En ARIMA, confirmamos que si no hay dependencia, caemos a naive1 sin romper.
    print("\n== Tests completados ==")

== Test: Estacional suave ==
[Theta] shape=(12,) finite=True nneg=True
[ETS_auto] shape=(12,) finite=True nneg=True
[SNaive12] shape=(12,) finite=True nneg=True
[MedianaEstac] shape=(12,) finite=True nneg=True

== Test: Intermitente ==
[Croston] shape=(12,) finite=True nneg=True
[SBA] shape=(12,) finite=True nneg=True
[TSB] shape=(12,) finite=True nneg=True
[PerfilEscaso] shape=(12,) finite=True nneg=True

== Test: Enrutado route=INTERMITENTE (redirigimos ETS->TSB) ==
[ETS->TSB route] shape=(12,) finite=True nneg=True

== Test: Todo-cero (activamos cero estructural) ==
[Cero] shape=(12,) finite=True nneg=True
[SNaive12] shape=(12,) finite=True nneg=True
[MedianaEstac] shape=(12,) finite=True nneg=True

== Test: ARIMA fallback ==
[ARIMA_auto] shape=(12,) finite=True nneg=True

== Test: ADIDA/MAPA ==
[ADIDA_MAPA] shape=(12,) finite=True nneg=True

== Tests completados ==

# ============================================================
# ESTRATEGIA 3 — PASO 8: Métricas finales sobre 2024
# ============================================================
# === Paso 8 - Bloque preliminar: inspección de columnas ===

# Definimos rutas
PATH_TRAIN = "/content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3/RESULTADOS/train_full_2021_2023.csv"
PATH_TEST  = "/content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3/RESULTADOS/test_full_2024.csv"
PATH_PREDS = "/content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3/RESULTADOS/preds_por_serie_2024.csv"

# Leemos los ficheros sin forzar columnas
df_train_raw = pd.read_csv(PATH_TRAIN, sep=";")
df_test_raw  = pd.read_csv(PATH_TEST, sep=";")
df_preds_raw = pd.read_csv(PATH_PREDS, sep=";")

# Imprimimos columnas para verificar
print("Columnas en TRAIN:", df_train_raw.columns.tolist())
print("Columnas en TEST:", df_test_raw.columns.tolist())
print("Columnas en PREDS:", df_preds_raw.columns.tolist())

Columnas en TRAIN: ['ID_BUILDING', 'FM_COST_TYPE', 'FECHA', 'cost_float_mod']
Columnas en TEST: ['ID_BUILDING', 'FM_COST_TYPE', 'FECHA', 'cost_float_mod']
Columnas en PREDS: ['ID_BUILDING', 'FM_COST_TYPE', 'FECHA', 'yhat_modelo1', 'yhat_modelo2', 'yhat_combo', 'fallback_flag', 'route', 'modelo_top1', 'modelo_top2', 'combo_rule']

# === Paso 8 - Bloque preliminar: inspección de columnas ===

# Definimos rutas
PATH_BASELINES = "/content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3/METRICAS/metrics_baselines_cv.csv"


# Leemos los ficheros sin forzar columnas
df_baselines_raw = pd.read_csv(PATH_BASELINES, sep=";")


# Imprimimos columnas para verificar
print("Columnas en baselines:", df_baselines_raw.columns.tolist())

df_baselines_raw.head()

Columnas en baselines: ['ID_BUILDING', 'FM_COST_TYPE', 'baseline', 'n_obs', 'MAE', 'WAPE', 'SMAPE', 'MASE1', 'MASE12']

# ============================================================
# PASO 8 — Métricas finales 2024 (evaluación del modelo_final)
# Bloque 1) Constantes y parámetros
# ============================================================

# Fijamos semilla para reproducibilidad en cualquier componente que la use
np.random.seed(7)

# Definimos rutas base (reutilizamos si ya existen; si no, ponemos defaults seguros)
ruta_base_3 = globals().get("ruta_base_3", "/content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3")
CSV_SEP     = globals().get("CSV_SEP", ";")

# Rutas canónicas de salida
RUTA_RESULTADOS = os.path.join(ruta_base_3, "RESULTADOS")
RUTA_METRICAS   = os.path.join(ruta_base_3, "METRICAS")
RUTA_LOGS       = os.path.join(ruta_base_3, "LOGS")
RUTA_REPORTING  = os.path.join(ruta_base_3, "REPORTING")
RUTA_FIGS       = os.path.join(ruta_base_3, "FIGS")

# Aseguramos carpetas de salida
for _d in [RUTA_RESULTADOS, RUTA_METRICAS, RUTA_LOGS, RUTA_REPORTING, RUTA_FIGS]:
    os.makedirs(_d, exist_ok=True)

# Ficheros de entrada del paso
PATH_TRAIN = os.path.join(RUTA_RESULTADOS, "train_full_2021_2023.csv")
PATH_TEST  = os.path.join(RUTA_RESULTADOS, "test_full_2024.csv")
PATH_PREDS = os.path.join(RUTA_RESULTADOS, "preds_por_serie_2024.csv")
PATH_SEG   = os.path.join(RUTA_METRICAS,   "segmentation_intermitencia_step3.csv")
PATH_BASES = os.path.join(RUTA_METRICAS,   "metrics_baselines_cv.csv")   # para SNaive12_2024
PATH_FINAL  = os.path.join(RUTA_METRICAS, "modelo_final_paso6.csv")

# Ficheros de salida del paso
OUT_PAIR   = os.path.join(RUTA_METRICAS,  "paso8_metrics_2024_por_pareja.csv")
OUT_OVER   = os.path.join(RUTA_METRICAS,  "paso8_metrics_2024_overall.csv")
OUT_VS_SNV = os.path.join(RUTA_REPORTING, "paso8_resumen_vs_snaive12_por_ruta.csv")
OUT_ALERTS = os.path.join(RUTA_REPORTING, "paso8_intermitentes_alertas.csv")  # vistas para alertas

# Parámetros de claves y columnas
PAIR_COLS = ["ID_BUILDING","FM_COST_TYPE"]
DATECOL   = "FECHA"
VALUE_TRAINTEST = "cost_float_mod"    # columna de valor canónica
PRED_COL  = "yhat_combo"              # columna de predicción combinada del Paso 5

# Rango canónico de 2024 (malla MS)
FREQ = "MS"
H    = 12
MONTHS_2024 = pd.date_range("2024-01-01", periods=H, freq=FREQ)

# Epsilon para divisiones seguras y clips
EPS = 1e-8

# Log del paso 8
LOG_PATH_STEP8 = os.path.join(RUTA_LOGS, "estrategia3_step8.log")
def log8(msg: str):
    """Escribimos mensajes del Paso 8 en el log y en consola para trazabilidad."""
    stamp = datetime.now().isoformat(timespec="seconds")
    line = f"[{stamp}] {msg}"
    print(line)
    with open(LOG_PATH_STEP8, "a", encoding="utf-8") as f:
        f.write(line + "\n")

# ============================================================
# PASO 8 — Bloque 2) Funciones auxiliares
# ============================================================

# -- Utilidades de fecha e IO --

def _ensure_ms(df: pd.DataFrame, datecol: str) -> pd.DataFrame:
    """Normalizamos la columna de fecha a mensual (MS)."""
    df = df.copy()
    df[datecol] = pd.to_datetime(df[datecol]).dt.to_period("M").dt.to_timestamp()
    return df

def _read_csv(path: str, required: bool=True, sep: str=CSV_SEP) -> pd.DataFrame:
    """Leemos un CSV con el separador del proyecto; si es opcional y no existe, devolvemos vacío."""
    if os.path.exists(path):
        return pd.read_csv(path, sep=sep)
    if required:
        raise FileNotFoundError(f"Falta el archivo requerido: {path}")
    log8(f"AVISO: no encontramos {os.path.basename(path)} (continuamos sin este artefacto).")
    return pd.DataFrame()

def _agg_dup_and_reindex(df: pd.DataFrame, datecol: str, valuecol: str,
                         start: str, end: str, pair_cols=PAIR_COLS, agg="sum") -> pd.DataFrame:
    """Agregamos duplicados por (pair, mes) y reindexamos a malla mensual rellenando huecos a 0."""
    if df.empty:
        return pd.DataFrame(columns=[*pair_cols, datecol, valuecol])
    g = (df.groupby(pair_cols + [pd.Grouper(key=datecol, freq="MS")], as_index=False)[valuecol]
           .agg(agg))
    all_pairs = g[pair_cols].drop_duplicates()
    months = pd.date_range(start, end, freq="MS")
    out = []
    for a, b in all_pairs.values:
        sub = g[(g[pair_cols[0]]==a) & (g[pair_cols[1]]==b)].set_index(datecol)
        sub = sub.reindex(months).rename_axis(datecol).reset_index()
        sub[pair_cols[0]] = a; sub[pair_cols[1]] = b
        sub[valuecol] = sub[valuecol].fillna(0.0)
        out.append(sub)
    return pd.concat(out, ignore_index=True)[[*pair_cols, datecol, valuecol]]

# -- Métricas (reutilizamos criterio del proyecto) --

def _safe_div(num, den, eps=EPS):
    """Implementamos división segura para evitar divisiones por cero."""
    return num / (den + eps)

def _mae(y_true, y_pred):
    """Calculamos el MAE de forma directa."""
    y_true = np.asarray(y_true, dtype=float)
    y_pred = np.asarray(y_pred, dtype=float)
    return float(np.mean(np.abs(y_pred - y_true)))

def _wape_safe(y_true, y_pred, eps=EPS):
    """Calculamos el WAPE con denominador seguro sum(|y_true|)+eps."""
    y_true = np.asarray(y_true, dtype=float)
    y_pred = np.asarray(y_pred, dtype=float)
    num = np.sum(np.abs(y_pred - y_true))
    den = np.sum(np.abs(y_true))
    return float(_safe_div(num, den, eps))

def _smape_safe(y_true, y_pred, eps=EPS):
    """Calculamos el SMAPE con denominador simétrico y eps para estabilidad (en %)."""
    y_true = np.asarray(y_true, dtype=float)
    y_pred = np.asarray(y_pred, dtype=float)
    num = np.abs(y_pred - y_true)
    den = (np.abs(y_true) + np.abs(y_pred)) / 2.0
    return float(np.mean(_safe_div(num, den, eps)) * 100.0)

def _scale_diff(series, lag):
    """Calculamos la escala como media de |y_t - y_{t-lag}| sobre train."""
    arr = np.asarray(series, dtype=float)
    if len(arr) <= lag:
        return np.nan
    diffs = np.abs(arr[lag:] - arr[:-lag])
    if len(diffs) == 0:
        return np.nan
    return float(np.mean(diffs))

def mase_from_scale(errors, scale):
    """Calculamos el MASE a partir de un vector de errores y una escala precomputada."""
    if scale is None or np.isnan(scale) or scale == 0:
        return np.nan
    errors = np.asarray(errors, dtype=float)
    return float(np.mean(np.abs(errors)) / float(scale))

def _eval_series(y_true, y_pred, escalas_train: dict):
    """Calculamos las métricas comunes con las escalas calculadas en train (lag 1 y 12)."""
    y_true = np.asarray(y_true, dtype=float)
    y_pred = np.asarray(y_pred, dtype=float)
    err = y_pred - y_true
    mae   = _mae(y_true, y_pred)
    wape  = _wape_safe(y_true, y_pred)
    smape = _smape_safe(y_true, y_pred)
    mase1  = mase_from_scale(err,  escalas_train.get("scale1"))
    mase12 = mase_from_scale(err, escalas_train.get("scale12"))
    return {"MAE": mae, "WAPE": wape, "SMAPE": smape, "MASE1": mase1, "MASE12": mase12}

# -- Ayudas de empaquetado y vistas --

def _source_model_from_preds(df_pair_preds: pd.DataFrame) -> str:
    """Inferimos el tipo de modelo aplicado en 2024 para una pareja a partir de las columnas de preds."""
    # Priorizamos baseline_cero si aparece explícito
    tops = set(str(x).strip().lower() for x in df_pair_preds.get("modelo_top1", pd.Series(dtype=object)).dropna().unique())
    if "baseline_cero" in tops:
        return "baseline_cero"
    # Si todo el año es fallback de SNaive12, lo marcamos como fallback_snaive12
    if "fallback_flag" in df_pair_preds.columns and "modelo_top1" in df_pair_preds.columns:
        ff_all = int(df_pair_preds["fallback_flag"].fillna(0).astype(int).sum()) == len(df_pair_preds)
        top1_snv = all(str(x).strip().lower()=="snaive12" for x in df_pair_preds["modelo_top1"].dropna().unique())
        if ff_all and top1_snv:
            return "fallback_snaive12"
    # En el resto de casos consideramos que proviene del modelo_final (posiblemente combinado)
    return "modelo_final"

def _two_views(y_true: np.ndarray, y_pred: np.ndarray):
    """Construimos máscaras para vistas all_months y positive_only."""
    mask_all = np.ones_like(y_true, dtype=bool)
    mask_pos = (y_true > 0)
    return {"all_months": mask_all, "positive_only": mask_pos}

def _smape_clip(y_true, y_pred, eps=EPS):
    """SMAPE con denominador clippeado por eps a modo diagnóstico."""
    y_true = np.asarray(y_true, dtype=float)
    y_pred = np.asarray(y_pred, dtype=float)
    num = np.abs(y_pred - y_true)
    den = np.maximum((np.abs(y_true) + np.abs(y_pred)) / 2.0, eps)
    return float(np.mean(num / den) * 100.0)

def _percent(x):
    """Formateamos porcentaje de forma robusta."""
    try:
        return f"{100.0*float(x):.2f}%"
    except Exception:
        return "NA"

def _plot_hist(data: pd.Series, title: str, out_path: str):
    """Guardamos histograma simple si hay datos suficientes (opcional para diagnóstico)."""
    try:
        plt.figure()
        data.dropna().astype(float).plot(kind="hist", bins=20)
        plt.title(title)
        plt.xlabel(title)
        plt.ylabel("Frecuencia")
        plt.tight_layout()
        plt.savefig(out_path)
        plt.close()
        log8(f"Guardado histograma: {out_path}")
    except Exception as e:
        log8(f"AVISO: no fue posible guardar histograma {title}: {e}")

def _normalize_baselines_df(df: pd.DataFrame) -> pd.DataFrame:
    """
    Normalizamos nombres de columnas de metrics_baselines_cv.csv a canónicos, de forma case-insensitive.
    Intentamos quedarnos con: 'baseline', PAIR_COLS, 'SMAPE', 'MASE12'.
    Si no existen, devolvemos NaN en esas métricas tras el rename.
    """
    if df is None or df.empty:
        return pd.DataFrame(columns=[*PAIR_COLS, "baseline", "SMAPE", "MASE12"])

    cols = list(df.columns)
    lut  = {c.lower(): c for c in cols}  # mapeo case-insensitive

    # Mapeo a nombres canónicos
    rename_map = {}
    if "smape"    in lut: rename_map[lut["smape"]]    = "SMAPE"
    if "mase12"   in lut: rename_map[lut["mase12"]]   = "MASE12"
    if "baseline" in lut: rename_map[lut["baseline"]] = "baseline"

    out = df.rename(columns=rename_map).copy()

    # Garantizamos presencia de métricas y claves
    for c in ["SMAPE", "MASE12"]:
        if c not in out.columns:
            out[c] = np.nan
    if "baseline" not in out.columns:
        out["baseline"] = np.nan
    for k in PAIR_COLS:
        if k not in out.columns:
            out[k] = np.nan

    # Tipado robusto y limpieza de baseline
    out["SMAPE"]  = pd.to_numeric(out["SMAPE"],  errors="coerce")
    out["MASE12"] = pd.to_numeric(out["MASE12"], errors="coerce")
    out["baseline"] = out["baseline"].astype(str).str.strip()

    return out

# ============================================================
# PASO 8 — Bloque 3) Ejecución
# ============================================================

# 1) Cargamos entradas y normalizamos fechas
train_df = _ensure_ms(_read_csv(PATH_TRAIN, required=True), DATECOL)
test_df  = _ensure_ms(_read_csv(PATH_TEST,  required=True), DATECOL)
preds_df = _ensure_ms(_read_csv(PATH_PREDS, required=True), DATECOL)
seg_df   = _read_csv(PATH_SEG, required=True)
bases_df = _read_csv(PATH_BASES, required=False)   # opcional pero recomendado

# 2) Alineamos y aseguramos estructura en 2024
#    - test: agregamos duplicados por suma y reindexamos mensual a 2024
test_2024 = test_df.copy()
test_2024 = test_2024[[*PAIR_COLS, DATECOL, VALUE_TRAINTEST]]
test_2024 = _agg_dup_and_reindex(test_2024, DATECOL, VALUE_TRAINTEST,
                                 start="2024-01-01", end="2024-12-01",
                                 pair_cols=PAIR_COLS, agg="sum")

#    - preds: por diseño ya venía 12 filas por pareja, pero robustecemos agregando por suma si hubiera duplicados
if PRED_COL not in preds_df.columns:
    raise ValueError(f"Esperábamos la columna de predicción '{PRED_COL}' en {os.path.basename(PATH_PREDS)}.")
preds_2024 = preds_df.copy()
preds_2024 = preds_2024[[*PAIR_COLS, DATECOL, PRED_COL, "fallback_flag", "route", "modelo_top1", "modelo_top2", "combo_rule"] \
                        if "fallback_flag" in preds_df.columns else [*PAIR_COLS, DATECOL, PRED_COL, "route", "modelo_top1", "modelo_top2", "combo_rule"]]
preds_2024 = _ensure_ms(preds_2024, DATECOL)
preds_2024 = (preds_2024.groupby(PAIR_COLS + [DATECOL], as_index=False)
                        .agg({PRED_COL:"sum",
                              "fallback_flag":"max" if "fallback_flag" in preds_2024.columns else "size"}))

#    - construimos diccionarios de ruta/flag_all_zero_train
seg_cols = [c for c in ["segment","route","flag_all_zero_train"] if c in seg_df.columns]
seg_view = seg_df[PAIR_COLS + seg_cols].drop_duplicates().copy()
#       normalizamos a columna 'ruta'
if "route" in seg_view.columns:
    seg_view = seg_view.rename(columns={"route":"ruta"})
elif "segment" in seg_view.columns:
    seg_view = seg_view.rename(columns={"segment":"ruta"})
else:
    seg_view["ruta"] = "GENERAL"
if "flag_all_zero_train" not in seg_view.columns:
    seg_view["flag_all_zero_train"] = 0

# 3) Panel de parejas a evaluar (restringimos a modelo_final del Paso 6)
final_df = _read_csv(PATH_FINAL, required=True)  # contiene al menos PAIR_COLS
final_pairs = (final_df[PAIR_COLS]
               .drop_duplicates()
               .copy())

# Normalizamos claves por robustez (evitamos falsos duplicados por espacios)
def _norm_pair_keys(df):
    out = df.copy()
    if out.empty:
        return out
    # ID_BUILDING a int si es posible
    try:
        out[PAIR_COLS[0]] = pd.to_numeric(out[PAIR_COLS[0]], errors="coerce").astype("Int64")
    except Exception:
        pass
    # FM_COST_TYPE strip de espacios
    if out[PAIR_COLS[1]].dtype == object:
        out[PAIR_COLS[1]] = out[PAIR_COLS[1]].astype(str).str.strip()
    return out

final_pairs = _norm_pair_keys(final_pairs)
test_2024   = _norm_pair_keys(test_2024)
preds_2024  = _norm_pair_keys(preds_2024)
seg_view    = _norm_pair_keys(seg_view)

# Nos quedamos exclusivamente con las parejas del modelo_final
pairs_panel = (final_pairs
               .merge(seg_view, on=PAIR_COLS, how="left")
               .fillna({"flag_all_zero_train": 0, "ruta": "GENERAL"}))

# Diagnóstico: cuántas de esas parejas tienen test 2024 y pred 2024
pairs_test = test_2024[PAIR_COLS].drop_duplicates()
pairs_pred = preds_2024[PAIR_COLS].drop_duplicates()

missing_test = (final_pairs
                .merge(pairs_test, on=PAIR_COLS, how="left", indicator=True)
                .query("_merge == 'left_only'")[PAIR_COLS])
missing_pred = (final_pairs
                .merge(pairs_pred, on=PAIR_COLS, how="left", indicator=True)
                .query("_merge == 'left_only'")[PAIR_COLS])

n_pairs_expected = pairs_panel.shape[0]
log8(f"Punto G — total de parejas candidatas a evaluar (modelo_final): {n_pairs_expected}")

if len(missing_test) > 0:
    log8(f"AVISO: parejas del modelo_final sin test_2024: {len(missing_test)}. Ejemplos: {missing_test.head(10).values.tolist()}")
if len(missing_pred) > 0:
    log8(f"AVISO: parejas del modelo_final sin preds_2024: {len(missing_pred)}. Ejemplos: {missing_pred.head(10).values.tolist()}")

# Índices para acceso rápido (se mantienen igual)
train_idx = train_df.set_index(PAIR_COLS)
test_idx  = test_2024.set_index(PAIR_COLS)
pred_idx  = preds_2024.set_index(PAIR_COLS)


# 4) Preparamos índices para acceso rápido
train_idx = train_df.set_index(PAIR_COLS)
test_idx  = test_2024.set_index(PAIR_COLS)
pred_idx  = preds_2024.set_index(PAIR_COLS)

# 5) Métricas por pareja (dos vistas)
rows = []
n_scale_zero = 0
n_alertas = 0

for bid, ctype, ruta, flag_zero in pairs_panel[[*PAIR_COLS, "ruta", "flag_all_zero_train"]].itertuples(index=False):
    # Extraemos histórico 2021–2023
    try:
        tr_sub = train_idx.loc[(bid, ctype)].reset_index()
    except KeyError:
        tr_sub = pd.DataFrame(columns=[*PAIR_COLS, DATECOL, VALUE_TRAINTEST])
    tr_sub = tr_sub.sort_values(DATECOL)
    y_train = tr_sub[VALUE_TRAINTEST].astype(float).values if VALUE_TRAINTEST in tr_sub.columns else np.array([], dtype=float)

    # Escalas MASE con lag 1 y 12 sobre train
    scale1  = _scale_diff(y_train, 1)
    scale12 = _scale_diff(y_train, 12)
    scale_zero = int((pd.isna(scale1) or scale1==0) or (pd.isna(scale12) or scale12==0))
    if scale_zero == 1:
        n_scale_zero += 1
    escalas = {"scale1": (np.nan if scale_zero else scale1),
               "scale12": (np.nan if scale_zero else scale12)}

    # Extraemos verdad-terreno 2024 y predicciones 2024
    try:
        te_sub = test_idx.loc[(bid, ctype)].reset_index()
    except KeyError:
        te_sub = pd.DataFrame(columns=[*PAIR_COLS, DATECOL, VALUE_TRAINTEST])
    te_sub = te_sub.set_index(DATECOL).reindex(MONTHS_2024).reset_index().rename(columns={"index":DATECOL})
    y_true = te_sub[VALUE_TRAINTEST].astype(float).values if VALUE_TRAINTEST in te_sub.columns else np.zeros(H, dtype=float)

    try:
        pr_sub = pred_idx.loc[(bid, ctype)].reset_index()
    except KeyError:
        pr_sub = pd.DataFrame(columns=[*PAIR_COLS, DATECOL, PRED_COL, "fallback_flag"])
    pr_sub = _ensure_ms(pr_sub, DATECOL).set_index(DATECOL).reindex(MONTHS_2024).reset_index().rename(columns={"index":DATECOL})
    y_pred = pr_sub[PRED_COL].astype(float).fillna(0.0).values if PRED_COL in pr_sub.columns else np.zeros(H, dtype=float)

    # Determinamos metadata del modelo fuente a nivel pareja
    # (reconstruimos desde el conjunto de predicciones anual)
    try:
        full_preds_pair = preds_df[(preds_df[PAIR_COLS[0]]==bid) & (preds_df[PAIR_COLS[1]]==ctype)].copy()
    except Exception:
        full_preds_pair = pd.DataFrame(columns=preds_df.columns)
    source_model = _source_model_from_preds(full_preds_pair)

    # Señal de alerta: cero estructural en train y actividad en 2024
    alerta_actividad = int((int(flag_zero)==1) and np.any(y_true > 0))
    if alerta_actividad == 1:
        n_alertas += 1

    # Construimos vistas y calculamos métricas
    views = _two_views(y_true, y_pred)

    # Guardamos FECHA_MIN y FECHA_MAX para la cobertura efectiva del 2024 observado
    if np.all(np.isnan(y_true)):
        fecha_min, fecha_max = (pd.NaT, pd.NaT)
    else:
        idx_valid = np.where(~np.isnan(y_true))[0]
        fecha_min = MONTHS_2024[idx_valid[0]] if idx_valid.size>0 else pd.NaT
        fecha_max = MONTHS_2024[idx_valid[-1]] if idx_valid.size>0 else pd.NaT

    for vista, mask in views.items():
        # Filtramos por máscara y también por no-NaN en verdad-terreno
        mask_eff = mask & (~np.isnan(y_true))
        y_t = y_true[mask_eff]
        y_p = y_pred[mask_eff]
        n_obs = int(mask_eff.sum())
        n_obs_pos = int(np.sum(y_t > 0))

        if n_obs == 0:
            # Si no hay observaciones válidas, registramos NaNs en métricas
            rows.append({
                PAIR_COLS[0]: bid, PAIR_COLS[1]: ctype, "FECHA_MIN": fecha_min, "FECHA_MAX": fecha_max,
                "vista": vista, "ruta": ruta, "flag_all_zero_train": int(flag_zero),
                "source_model": source_model,
                "n_obs": 0, "n_obs_pos": 0,
                "scale1": scale1, "scale12": scale12, "scale_zero": scale_zero,
                "MAE": np.nan, "WAPE": np.nan, "SMAPE": np.nan, "MASE1": np.nan, "MASE12": np.nan,
                # Diagnóstico opcional
                "WAPE_micro": np.nan, "SMAPE_clip": np.nan,
                # Comparativa (rellenaremos después)
                "MASE12_SNaive12": np.nan, "SMAPE_SNaive12": np.nan,
                "delta_MASE12_vs_SNaive12": np.nan, "delta_SMAPE_vs_SNaive12": np.nan,
                # Info opcional del combo si existiera
                "combo_rule": full_preds_pair["combo_rule"].dropna().unique()[0] if "combo_rule" in full_preds_pair.columns and full_preds_pair["combo_rule"].notna().any() else None,
                "modelo_top1": full_preds_pair["modelo_top1"].dropna().unique()[0] if "modelo_top1" in full_preds_pair.columns and full_preds_pair["modelo_top1"].notna().any() else None,
                "modelo_top2": full_preds_pair["modelo_top2"].dropna().unique()[0] if "modelo_top2" in full_preds_pair.columns and full_preds_pair["modelo_top2"].notna().any() else None,
                "alerta_actividad_2024": alerta_actividad
            })
            continue

        # Calculamos métricas oficiales para la vista
        metrics = _eval_series(y_t, y_p, escalas)
        # Diagnóstico opcional (no sustituye oficiales)
        WAPE_micro = _wape_safe(y_t, y_p)
        SMAPE_clip = _smape_clip(y_t, y_p, eps=EPS)

        rows.append({
            PAIR_COLS[0]: bid, PAIR_COLS[1]: ctype, "FECHA_MIN": fecha_min, "FECHA_MAX": fecha_max,
            "vista": vista, "ruta": ruta, "flag_all_zero_train": int(flag_zero),
            "source_model": source_model,
            "n_obs": n_obs, "n_obs_pos": n_obs_pos,
            "scale1": scale1, "scale12": scale12, "scale_zero": scale_zero,
            **metrics,
            "WAPE_micro": WAPE_micro, "SMAPE_clip": SMAPE_clip,
            # Comparativa (rellenaremos después)
            "MASE12_SNaive12": np.nan, "SMAPE_SNaive12": np.nan,
            "delta_MASE12_vs_SNaive12": np.nan, "delta_SMAPE_vs_SNaive12": np.nan,
            # Info opcional de combinación
            "combo_rule": full_preds_pair["combo_rule"].dropna().unique()[0] if "combo_rule" in full_preds_pair.columns and full_preds_pair["combo_rule"].notna().any() else None,
            "modelo_top1": full_preds_pair["modelo_top1"].dropna().unique()[0] if "modelo_top1" in full_preds_pair.columns and full_preds_pair["modelo_top1"].notna().any() else None,
            "modelo_top2": full_preds_pair["modelo_top2"].dropna().unique()[0] if "modelo_top2" in full_preds_pair.columns and full_preds_pair["modelo_top2"].notna().any() else None,
            "alerta_actividad_2024": alerta_actividad
        })

# 6) Construimos DataFrame por pareja y persistimos (aún sin comparativa vs SNaive12)
pair_df = pd.DataFrame(rows)

# Trazabilidad básica
n_pairs_eval = pair_df[PAIR_COLS].drop_duplicates().shape[0]
log8(f"Punto H — parejas evaluadas: {n_pairs_eval} (esperado ~ 2.429)")
log8(f"Punto H — casos scale_zero=1: {n_scale_zero}")
log8(f"Punto H — nº alertas cero→actividad_2024: {n_alertas}")

# 7) Comparativa contra SNaive12 (si tenemos el artefacto)
if not bases_df.empty:
    # 7.1 Normalizamos el CSV de baselines (case-insensitive y columnas canónicas)
    bases_norm = _normalize_baselines_df(bases_df)

    # 7.2 Normalizamos claves para evitar falsos no-joins por tipos/espacios
    bases_norm = _norm_pair_keys(bases_norm)

    # 7.3 Filtramos SNaive12 por pareja y nos quedamos con SMAPE y MASE12 de 2024
    snaive = bases_norm[bases_norm["baseline"].str.lower() == "snaive12"][
        PAIR_COLS + ["SMAPE", "MASE12"]
    ].rename(columns={"SMAPE": "SMAPE_SNaive12", "MASE12": "MASE12_SNaive12"})

    # 7.4 Unimos con la vista all_months (comparativa oficial)
    mask_all = (pair_df["vista"] == "all_months")

    # Antes de merge: eliminamos en el lado izquierdo las columnas de comparativa si existen
    cols_cmp = ["MASE12_SNaive12", "SMAPE_SNaive12",
                "delta_MASE12_vs_SNaive12", "delta_SMAPE_vs_SNaive12"]
    left_all = pair_df[mask_all].drop(columns=cols_cmp, errors="ignore").copy()

    pair_df_all = left_all.merge(snaive, on=PAIR_COLS, how="left")

    # 7.5 Deltas (final - snaive12) con coerción segura
    pair_df_all["delta_MASE12_vs_SNaive12"] = (
        pd.to_numeric(pair_df_all["MASE12"], errors="coerce")
        - pd.to_numeric(pair_df_all["MASE12_SNaive12"], errors="coerce")
    )
    pair_df_all["delta_SMAPE_vs_SNaive12"] = (
        pd.to_numeric(pair_df_all["SMAPE"], errors="coerce")
        - pd.to_numeric(pair_df_all["SMAPE_SNaive12"], errors="coerce")
    )

    # 7.6 Volcamos comparativa de vuelta a pair_df (solo columnas necesarias)
    pair_df = pair_df.merge(
        pair_df_all[PAIR_COLS + ["MASE12_SNaive12", "SMAPE_SNaive12",
                                 "delta_MASE12_vs_SNaive12", "delta_SMAPE_vs_SNaive12"]],
        on=PAIR_COLS, how="left"
    )

    # 7.7 Resumen por ruta (solo all_months)
    cmp_df = pair_df_all.copy()
    cmp_df["mejora_MASE12"] = (
        pd.to_numeric(cmp_df["MASE12"], errors="coerce")
        < pd.to_numeric(cmp_df["MASE12_SNaive12"], errors="coerce")
    ).astype(int)
    cmp_df["mejora_SMAPE"] = (
        pd.to_numeric(cmp_df["SMAPE"], errors="coerce")
        < pd.to_numeric(cmp_df["SMAPE_SNaive12"], errors="coerce")
    ).astype(int)

    resumen_vs = (cmp_df.groupby("ruta")
                  .apply(lambda g: pd.Series({
                      "n_pairs": g[PAIR_COLS].drop_duplicates().shape[0],
                      "pct_mejora_MASE12": g["mejora_MASE12"].mean(),
                      "pct_mejora_SMAPE":  g["mejora_SMAPE"].mean(),
                      "delta_MASE12_medio": g["delta_MASE12_vs_SNaive12"].mean(),
                      "delta_SMAPE_medio":  g["delta_SMAPE_vs_SNaive12"].mean(),
                  }))
                  .reset_index())
    resumen_vs["pct_mejora_MASE12_fmt"] = resumen_vs["pct_mejora_MASE12"].map(_percent)
    resumen_vs["pct_mejora_SMAPE_fmt"]  = resumen_vs["pct_mejora_SMAPE"].map(_percent)
    resumen_vs.to_csv(OUT_VS_SNV, sep=CSV_SEP, index=False)
    log8(f"Guardado: {OUT_VS_SNV}")
else:
    log8("AVISO: no disponemos de metrics_baselines_cv.csv; omitimos comparativa vs SNaive12.")

# 7.8 Red de seguridad: garantizamos columnas de comparativa antes de ordenar/guardar
for c in ["MASE12_SNaive12", "SMAPE_SNaive12",
          "delta_MASE12_vs_SNaive12", "delta_SMAPE_vs_SNaive12"]:
    if c not in pair_df.columns:
        pair_df[c] = np.nan

# 8) Persistimos tabla por pareja (dos vistas)
#    Ordenamos columnas claves -> métricas -> comparativa
col_order = [
    *PAIR_COLS, "FECHA_MIN", "FECHA_MAX", "vista", "ruta", "flag_all_zero_train", "source_model",
    "n_obs", "n_obs_pos", "scale1", "scale12", "scale_zero",
    "MAE", "WAPE", "SMAPE", "MASE1", "MASE12",
    "WAPE_micro", "SMAPE_clip",
    "MASE12_SNaive12", "SMAPE_SNaive12", "delta_MASE12_vs_SNaive12", "delta_SMAPE_vs_SNaive12",
    "combo_rule", "modelo_top1", "modelo_top2", "alerta_actividad_2024"
]
# Añadimos cualquier columna faltante como NaN para guardar orden estable
for c in col_order:
    if c not in pair_df.columns:
        pair_df[c] = np.nan

pair_df = pair_df[col_order].sort_values(PAIR_COLS + ["vista"]).reset_index(drop=True)
pair_df.to_csv(OUT_PAIR, sep=CSV_SEP, index=False)
log8(f"Guardado: {OUT_PAIR} con {len(pair_df)} filas.")

# 9) Reporting agregado — overall (global, por ruta, por tipo_modelo)
#    Derivamos 'tipo_modelo' a partir de source_model
def _tipo_from_source(s: str) -> str:
    s = str(s or "").strip().lower()
    if s == "baseline_cero":
        return "baseline_cero"
    if s == "fallback_snaive12":
        return "fallback_snaive12"
    return "modelo_final"

pair_df["tipo_modelo"] = pair_df["source_model"].map(_tipo_from_source)

def _agg_metrics(df):
    return pd.Series({
        "n_pairs": df[PAIR_COLS].drop_duplicates().shape[0],
        "MAE_mean": df["MAE"].mean(),     "MAE_median": df["MAE"].median(),
        "WAPE_mean": df["WAPE"].mean(),   "WAPE_median": df["WAPE"].median(),
        "SMAPE_mean": df["SMAPE"].mean(), "SMAPE_median": df["SMAPE"].median(),
        "MASE1_mean": df["MASE1"].mean(), "MASE1_median": df["MASE1"].median(),
        "MASE12_mean": df["MASE12"].mean(),"MASE12_median": df["MASE12"].median(),
        "n_obs_mean": df["n_obs"].mean(), "n_obs_pos_mean": df["n_obs_pos"].mean()
    })

# Solo usamos vista all_months para el agregado oficial
pair_all = pair_df[pair_df["vista"]=="all_months"].copy()

overall_global = _agg_metrics(pair_all).to_frame().T.assign(n_rows=len(pair_all), scope="GLOBAL")
overall_by_route = (pair_all.groupby("ruta").apply(_agg_metrics).reset_index().assign(scope="POR_RUTA"))
overall_by_tipo  = (pair_all.groupby("tipo_modelo").apply(_agg_metrics).reset_index().assign(scope="POR_TIPO"))

overall_out = pd.concat([overall_global, overall_by_route, overall_by_tipo], ignore_index=True)
overall_out.to_csv(OUT_OVER, sep=CSV_SEP, index=False)
log8(f"Guardado: {OUT_OVER}")

# 10) Reporte de intermitentes con alertas (flag_all_zero_train=1 y alerta_actividad_2024=1)
alerts = pair_df[(pair_df["flag_all_zero_train"]==1) & (pair_df["alerta_actividad_2024"]==1)].copy()
if not alerts.empty:
    # Dejamos doble vista de métricas (all_months / positive_only)
    alerts = alerts.sort_values(PAIR_COLS + ["vista"]).reset_index(drop=True)
    alerts.to_csv(OUT_ALERTS, sep=CSV_SEP, index=False)
    log8(f"Guardado: {OUT_ALERTS} con {alerts[PAIR_COLS].drop_duplicates().shape[0]} parejas.")
else:
    log8("No hay parejas con alerta_actividad_2024=1 (cero estructural en train con actividad en 2024).")

# 11) Logs de trazabilidad y puntos de control
#     - Distribución por ruta y flag_all_zero_train
dist_ruta = (pair_all.groupby("ruta")[PAIR_COLS]
             .apply(lambda d: d.drop_duplicates().shape[0])
             .rename("n_pairs").reset_index())

dist_flag = (pair_all.groupby("flag_all_zero_train")[PAIR_COLS]
             .apply(lambda d: d.drop_duplicates().shape[0])
             .rename("n_pairs").reset_index())

log8(f"Distribución por ruta (n_pairs): {dist_ruta.to_dict(orient='records')}")
log8(f"Distribución por flag_all_zero_train (n_pairs): {dist_flag.to_dict(orient='records')}")

#     - Histogramas (opcionales) de n_obs y n_obs_pos
_plot_hist(pair_all["n_obs"], "n_obs", os.path.join(RUTA_FIGS, "paso8_hist_n_obs.png"))
_plot_hist(pair_all["n_obs_pos"], "n_obs_pos", os.path.join(RUTA_FIGS, "paso8_hist_n_obs_pos.png"))

#     - % de mejoras vs SNaive12 por ruta (si hay comparativa)
if "MASE12_SNaive12" in pair_all.columns and pair_all["MASE12_SNaive12"].notna().any():
    tmp = pair_all.dropna(subset=["MASE12","MASE12_SNaive12"]).copy()
    tmp["win_MASE12"] = (tmp["MASE12"] < tmp["MASE12_SNaive12"]).astype(int)
    pct_by_route = (tmp.groupby("ruta")["win_MASE12"].mean().rename("pct_win_MASE12").reset_index())
    log8("Porcentaje de mejoras vs SNaive12 (MASE12) por ruta: " + pct_by_route.to_dict(orient="records").__str__())

    if "SMAPE_SNaive12" in pair_all.columns and pair_all["SMAPE_SNaive12"].notna().any():
        tmp2 = pair_all.dropna(subset=["SMAPE","SMAPE_SNaive12"]).copy()
        tmp2["win_SMAPE"] = (tmp2["SMAPE"] < tmp2["SMAPE_SNaive12"]).astype(int)
        pct_by_route_smape = (tmp2.groupby("ruta")["win_SMAPE"].mean().rename("pct_win_SMAPE").reset_index())
        log8("Porcentaje de mejoras vs SNaive12 (SMAPE) por ruta: " + pct_by_route_smape.to_dict(orient="records").__str__())

#     - Pairs con n_obs != 12
bad_cov = pair_all[pair_all["n_obs"] != 12][PAIR_COLS].drop_duplicates()
log8(f"Parejas con n_obs != 12: {bad_cov.shape[0]}")
if bad_cov.shape[0] > 0:
    ejemplos = bad_cov.head(10).values.tolist()
    log8(f"Ejemplos (hasta 10): {ejemplos}")

#     - Top-10 peores por MASE12 (inspección rápida)
top_bad = pair_all.sort_values("MASE12", ascending=False)[PAIR_COLS + ["MASE12"]].head(10)
log8("Top-10 peores MASE12 (vista all_months): " + top_bad.to_dict(orient="records").__str__())

log8("Paso 8 finalizado.")

[2025-09-23T15:52:11] Punto G — total de parejas candidatas a evaluar (modelo_final): 2429
[2025-09-23T15:52:11] AVISO: parejas del modelo_final sin test_2024: 198. Ejemplos: [[2, 'Eficiencia Energética'], [57, 'Eficiencia Energética'], [57, 'Licencias'], [57, 'Mtto. Correctivo'], [57, 'Servicios Ctto.'], [57, 'Servicios Extra'], [57, 'Suministros'], [59, 'Licencias'], [59, 'Mtto. Correctivo'], [59, 'Obras']]
[2025-09-23T15:53:26] Punto H — parejas evaluadas: 2429 (esperado ~ 2.429)
[2025-09-23T15:53:26] Punto H — casos scale_zero=1: 106
[2025-09-23T15:53:27] Punto H — nº alertas cero→actividad_2024: 35
[2025-09-23T15:53:27] Guardado: /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3/REPORTING/paso8_resumen_vs_snaive12_por_ruta.csv
[2025-09-23T15:53:28] Guardado: /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3/METRICAS/paso8_metrics_2024_por_pareja.csv con 4858 filas.
[2025-09-23T15:53:28] Guardado: /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3/METRICAS/paso8_metrics_2024_overall.csv
[2025-09-23T15:53:28] Guardado: /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3/REPORTING/paso8_intermitentes_alertas.csv con 35 parejas.
[2025-09-23T15:53:28] Distribución por ruta (n_pairs): [{'ruta': 'GENERAL', 'n_pairs': 1032}, {'ruta': 'INTERMITENTE', 'n_pairs': 1397}]
[2025-09-23T15:53:28] Distribución por flag_all_zero_train (n_pairs): [{'flag_all_zero_train': 0, 'n_pairs': 2325}, {'flag_all_zero_train': 1, 'n_pairs': 104}]
[2025-09-23T15:53:29] Guardado histograma: /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3/FIGS/paso8_hist_n_obs.png
[2025-09-23T15:53:30] Guardado histograma: /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3/FIGS/paso8_hist_n_obs_pos.png
[2025-09-23T15:53:30] Parejas con n_obs != 12: 198
[2025-09-23T15:53:30] Ejemplos (hasta 10): [[2, 'Eficiencia Energética'], [57, 'Eficiencia Energética'], [57, 'Licencias'], [57, 'Mtto. Correctivo'], [57, 'Servicios Ctto.'], [57, 'Servicios Extra'], [57, 'Suministros'], [59, 'Licencias'], [59, 'Mtto. Correctivo'], [59, 'Obras']]
[2025-09-23T15:53:30] Top-10 peores MASE12 (vista all_months): [{'ID_BUILDING': 1001155, 'FM_COST_TYPE': 'Servicios Ctto.', 'MASE12': 797.9007421234954}, {'ID_BUILDING': 1001156, 'FM_COST_TYPE': 'Servicios Ctto.', 'MASE12': 712.7656069883597}, {'ID_BUILDING': 1001154, 'FM_COST_TYPE': 'Servicios Extra', 'MASE12': 703.7574615284893}, {'ID_BUILDING': 1001135, 'FM_COST_TYPE': 'Servicios Ctto.', 'MASE12': 583.1854774654578}, {'ID_BUILDING': 1001154, 'FM_COST_TYPE': 'Servicios Ctto.', 'MASE12': 519.3349754676485}, {'ID_BUILDING': 1000270, 'FM_COST_TYPE': 'Servicios Extra', 'MASE12': 313.793202614379}, {'ID_BUILDING': 1000607, 'FM_COST_TYPE': 'Servicios Ctto.', 'MASE12': 195.7336820334289}, {'ID_BUILDING': 617, 'FM_COST_TYPE': 'Servicios Ctto.', 'MASE12': 166.6442921512004}, {'ID_BUILDING': 1000574, 'FM_COST_TYPE': 'Servicios Ctto.', 'MASE12': 165.89353993751413}, {'ID_BUILDING': 257, 'FM_COST_TYPE': 'Servicios Extra', 'MASE12': 132.52504332427952}]
[2025-09-23T15:53:30] Paso 8 finalizado.

# ============================================================
# ANÁLISIS PASO 8: Valoración aleatoria del modelo masivo
# Bloque 1) Constantes y parámetros
# ============================================================


# Reutilizamos variables globales si existen; si no, ponemos defaults seguros
ruta_base_3 = globals().get("ruta_base_3", "/content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3")
CSV_SEP     = globals().get("CSV_SEP", ";")

RUTA_METRICAS   = os.path.join(ruta_base_3, "METRICAS")
RUTA_REPORTING  = os.path.join(ruta_base_3, "REPORTING")

PATH_PAIR   = os.path.join(RUTA_METRICAS,  "paso8_metrics_2024_por_pareja.csv")
PATH_OVER   = os.path.join(RUTA_METRICAS,  "paso8_metrics_2024_overall.csv")
PATH_VS_SNV = os.path.join(RUTA_REPORTING, "paso8_resumen_vs_snaive12_por_ruta.csv")

PAIR_COLS = ["ID_BUILDING","FM_COST_TYPE"]  # por coherencia
EPS = 1e-8

# Si ya tenemos log8 del Paso 8, lo reutilizamos; si no, definimos uno local
if "log8" not in globals():
    from datetime import datetime
    LOG_PATH_STEP8_ANALYSIS = os.path.join(ruta_base_3, "LOGS", "estrategia3_step8_analisis.log")
    os.makedirs(os.path.dirname(LOG_PATH_STEP8_ANALYSIS), exist_ok=True)
    def log8(msg: str):
        stamp = datetime.now().isoformat(timespec="seconds")
        line = f"[{stamp}] {msg}"
        print(line)
        with open(LOG_PATH_STEP8_ANALYSIS, "a", encoding="utf-8") as f:
            f.write(line + "\n")

# ============================================================
# Bloque 2) Funciones auxiliares
# ============================================================

def _read_csv_safe(path: str, sep: str = CSV_SEP) -> pd.DataFrame:
    """Leemos un CSV si existe; si no existe, devolvemos DataFrame vacío y avisamos."""
    if os.path.exists(path):
        return pd.read_csv(path, sep=sep)
    log8(f"AVISO: no encontramos {path}. Devolvemos DataFrame vacío para continuar el análisis.")
    return pd.DataFrame()

def _pct(x):
    """Formateamos porcentaje de forma robusta."""
    try:
        return f"{100.0*float(x):.2f}%"
    except Exception:
        return "NA"

def _num(x, nd=3):
    """Formateamos número con decimales controlados."""
    try:
        return f"{float(x):.{nd}f}"
    except Exception:
        return "NA"

def _summary_quantiles(s: pd.Series, qs=(0.1,0.25,0.5,0.75,0.9)):
    """Devolvemos un dict con cuantiles útiles para una serie numérica."""
    s = pd.to_numeric(s, errors="coerce").dropna()
    if s.empty:
        return {}
    out = {f"q{int(q*100)}": float(s.quantile(q)) for q in qs}
    out["mean"] = float(s.mean())
    return out

# ============================================================
# Bloque 3) Ejecución del análisis
# ============================================================

# 1) Cargamos artefactos del Paso 8
pair_df = _read_csv_safe(PATH_PAIR)
over_df = _read_csv_safe(PATH_OVER)
vs_df   = _read_csv_safe(PATH_VS_SNV)

log8("Iniciamos análisis post–Paso 8.")

# 2) Análisis de comparativa vs SNaive12 (por ruta)
if not vs_df.empty:
    # Nos aseguramos de que existan columnas esperadas
    expected_cols = {"ruta","n_pairs","pct_mejora_MASE12","pct_mejora_SMAPE","delta_MASE12_medio","delta_SMAPE_medio"}
    missing = expected_cols - set(vs_df.columns)
    if missing:
        log8(f"AVISO: faltan columnas en resumen_vs_snaive12: {missing}. Continuamos con las disponibles.")

    # Imprimimos un resumen formateado por ruta
    log8("Resumen vs SNaive12 por ruta (siendo negativo mejor que baseline en deltas):")
    cols_show = [c for c in ["ruta","n_pairs","pct_mejora_MASE12","pct_mejora_SMAPE","delta_MASE12_medio","delta_SMAPE_medio"] if c in vs_df.columns]
    for _, r in vs_df[cols_show].iterrows():
        ruta   = r.get("ruta","NA")
        nps    = r.get("n_pairs", np.nan)
        pm12   = r.get("pct_mejora_MASE12", np.nan)
        psm    = r.get("pct_mejora_SMAPE", np.nan)
        d12    = r.get("delta_MASE12_medio", np.nan)
        dsm    = r.get("delta_SMAPE_medio", np.nan)
        log8(f"Ruta={ruta} | n_pairs={int(nps) if pd.notna(nps) else 'NA'} | "
             f"% mejora MASE12={_pct(pm12)} | % mejora SMAPE={_pct(psm)} | "
             f"ΔMASE12 medio={_num(d12)} | ΔSMAPE medio={_num(dsm)}")

    # Interpretación operacional
    log8("Interpretación: si el % de mejora MASE12 supera el 50% en GENERAL y el ΔMASE12 medio es negativo, concluimos que el modelo supera a SNaive12 en esa ruta.")
else:
    log8("No disponemos de paso8_resumen_vs_snaive12_por_ruta.csv; omitimos la comparativa por ruta.")

# 3) Agregados globales y segmentados (overall)
if not over_df.empty:
    # Mostramos filas GLOBAL, POR_RUTA y POR_TIPO si existen
    if "scope" in over_df.columns:
        # Global
        gl = over_df[over_df["scope"]=="GLOBAL"]
        if not gl.empty:
            r = gl.iloc[0].to_dict()
            log8("Overall GLOBAL (vista all_months): "
                 f"MAE_mediana="+_num(r.get("MAE_median"))+", WAPE_mediana="+_num(r.get("WAPE_median"))+
                 ", SMAPE_mediana="+_num(r.get("SMAPE_median"))+", MASE12_mediana="+_num(r.get("MASE12_median")))
        # Por ruta
        pr = over_df[over_df["scope"]=="POR_RUTA"]
        if not pr.empty:
            log8("Overall POR_RUTA (medianas):")
            for _, row in pr.iterrows():
                log8(f"  Ruta={row.get('ruta','NA')} | SMAPE_mediana={_num(row.get('SMAPE_median'))} "
                     f"| MASE12_mediana={_num(row.get('MASE12_median'))}")
        # Por tipo de modelo
        pt = over_df[over_df["scope"]=="POR_TIPO"]
        if not pt.empty:
            log8("Overall POR_TIPO (medianas):")
            for _, row in pt.iterrows():
                log8(f"  Tipo={row.get('tipo_modelo','NA')} | SMAPE_mediana={_num(row.get('SMAPE_median'))} "
                     f"| MASE12_mediana={_num(row.get('MASE12_median'))}")
    else:
        log8("AVISO: el overall no trae columna 'scope'; mostramos primeras filas para inspección rápida.")
        log8(over_df.head(5).to_string(index=False))
else:
    log8("No disponemos de paso8_metrics_2024_overall.csv; omitimos agregados globales y segmentados.")

# 4) Profundizamos en la tabla por pareja (vista all_months)
if not pair_df.empty:
    # Nos quedamos con la vista oficial del año completo
    all_df = pair_df[pair_df.get("vista","all_months")=="all_months"].copy()

    # Cuantiles de SMAPE y MASE12 (global)
    smape_q = _summary_quantiles(all_df["SMAPE"])
    mase12_q = _summary_quantiles(all_df["MASE12"])
    log8(f"Distribución SMAPE (all_months) — { {k:_num(v) for k,v in smape_q.items()} }")
    log8(f"Distribución MASE12 (all_months) — { {k:_num(v) for k,v in mase12_q.items()} }")

    # Por ruta
    if "ruta" in all_df.columns:
        by_ruta = []
        for ruta, g in all_df.groupby("ruta"):
            sm_q = _summary_quantiles(g["SMAPE"])
            ms_q = _summary_quantiles(g["MASE12"])
            by_ruta.append((ruta, sm_q, ms_q))
        log8("Distribución por ruta (medianas clave):")
        for ruta, sm_q, ms_q in by_ruta:
            log8(f"  Ruta={ruta} | SMAPE_mediana={_num(sm_q.get('q50'))} | MASE12_mediana={_num(ms_q.get('q50'))}")

    # Si vienen columnas de comparativa, calculamos % de victorias otra vez como sanity check
    cols_needed = {"MASE12","MASE12_SNaive12"}
    if cols_needed.issubset(all_df.columns) and all_df["MASE12_SNaive12"].notna().any():
        tmp = all_df.dropna(subset=list(cols_needed)).copy()
        tmp["win_MASE12"] = (pd.to_numeric(tmp["MASE12"], errors="coerce") <
                             pd.to_numeric(tmp["MASE12_SNaive12"], errors="coerce")).astype(int)
        win_global = tmp["win_MASE12"].mean()
        log8(f"% de parejas que mejoran a SNaive12 en MASE12 (recalculo rápido): {_pct(win_global)}")
        if "ruta" in tmp.columns:
            win_by_ruta = tmp.groupby("ruta")["win_MASE12"].mean()
            for ruta, val in win_by_ruta.items():
                log8(f"  Ruta={ruta} | % win MASE12={_pct(val)}")

    # Interpretación operativa breve
    log8("Interpretación: si la mediana de SMAPE y MASE12 del modelo_final es sensiblemente menor que la de los fallback, "
         "y además vemos >50% de victorias vs SNaive12, concluimos que el modelo masivo generaliza por encima del baseline.")
else:
    log8("No disponemos de paso8_metrics_2024_por_pareja.csv; no podemos profundizar por serie.")

log8("Análisis post–Paso 8 finalizado.")

[2025-09-23T15:53:30] Iniciamos análisis post–Paso 8.
[2025-09-23T15:53:30] Resumen vs SNaive12 por ruta (siendo negativo mejor que baseline en deltas):
[2025-09-23T15:53:30] Ruta=GENERAL | n_pairs=1032 | % mejora MASE12=64.53% | % mejora SMAPE=62.31% | ΔMASE12 medio=-0.072 | ΔSMAPE medio=-2.262
[2025-09-23T15:53:30] Ruta=INTERMITENTE | n_pairs=1397 | % mejora MASE12=23.84% | % mejora SMAPE=11.38% | ΔMASE12 medio=-0.044 | ΔSMAPE medio=32.836
[2025-09-23T15:53:30] Interpretación: si el % de mejora MASE12 supera el 50% en GENERAL y el ΔMASE12 medio es negativo, concluimos que el modelo supera a SNaive12 en esa ruta.
[2025-09-23T15:53:30] Overall GLOBAL (vista all_months): MAE_mediana=159.472, WAPE_mediana=0.858, SMAPE_mediana=63.589, MASE12_mediana=1.046
[2025-09-23T15:53:30] Overall POR_RUTA (medianas):
[2025-09-23T15:53:30]   Ruta=GENERAL | SMAPE_mediana=42.361 | MASE12_mediana=0.672
[2025-09-23T15:53:30]   Ruta=INTERMITENTE | SMAPE_mediana=97.282 | MASE12_mediana=1.676
[2025-09-23T15:53:30] Overall POR_TIPO (medianas):
[2025-09-23T15:53:30]   Tipo=baseline_cero | SMAPE_mediana=16.667 | MASE12_mediana=nan
[2025-09-23T15:53:30]   Tipo=fallback_snaive12 | SMAPE_mediana=50.000 | MASE12_mediana=1.378
[2025-09-23T15:53:30]   Tipo=modelo_final | SMAPE_mediana=75.220 | MASE12_mediana=0.943
[2025-09-23T15:53:30] Distribución SMAPE (all_months) — {'q10': '10.722', 'q25': '32.529', 'q50': '63.589', 'q75': '125.493', 'q90': '178.561', 'mean': '78.767'}
[2025-09-23T15:53:30] Distribución MASE12 (all_months) — {'q10': '0.215', 'q25': '0.445', 'q50': '1.046', 'q75': '2.325', 'q90': '4.622', 'mean': '4.479'}
[2025-09-23T15:53:30] Distribución por ruta (medianas clave):
[2025-09-23T15:53:30]   Ruta=GENERAL | SMAPE_mediana=42.361 | MASE12_mediana=0.672
[2025-09-23T15:53:30]   Ruta=INTERMITENTE | SMAPE_mediana=97.282 | MASE12_mediana=1.676
[2025-09-23T15:53:30] Interpretación: si la mediana de SMAPE y MASE12 del modelo_final es sensiblemente menor que la de los fallback, y además vemos >50% de victorias vs SNaive12, concluimos que el modelo masivo generaliza por encima del baseline.
[2025-09-23T15:53:30] Análisis post–Paso 8 finalizado.

# ============================================================
# ANÁLISIS GRÁFICO — 10 parejas al azar (real 2024 vs modelo_final)
# Bloque 1) Constantes y parámetros
# ============================================================


# Reutilizamos globals si existen; si no, definimos por defecto
ruta_base_3 = globals().get("ruta_base_3", "/content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3")
CSV_SEP     = globals().get("CSV_SEP", ";")

RUTA_RESULTADOS = os.path.join(ruta_base_3, "RESULTADOS")
RUTA_FIGS       = os.path.join(ruta_base_3, "FIGS")
os.makedirs(RUTA_FIGS, exist_ok=True)

# Claves y columnas estándar del proyecto
PAIR_COLS = globals().get("PAIR_COLS", ["ID_BUILDING","FM_COST_TYPE"])
DATECOL   = globals().get("DATECOL", "FECHA")
VALUE_TRAINTEST = globals().get("VALUE_TRAINTEST", "cost_float_mod")  # valor real
PRED_COL  = globals().get("PRED_COL", "yhat_combo")                   # predicción combinada

# Rango 2024 canónico
FREQ = "MS"
H    = 12
MONTHS_2024 = pd.date_range("2024-01-01", periods=H, freq=FREQ)

np.random.seed(42)

# ============================================================
# Bloque 2) Funciones auxiliares
# ============================================================

def _ensure_ms(df: pd.DataFrame, datecol: str) -> pd.DataFrame:
    """Normalizamos la columna de fecha a frecuencia mensual (MS)."""
    df = df.copy()
    df[datecol] = pd.to_datetime(df[datecol]).dt.to_period("M").dt.to_timestamp()
    return df

def _agg_dup_and_reindex(df: pd.DataFrame, datecol: str, valuecol: str,
                         start: str, end: str, pair_cols=PAIR_COLS, agg="sum") -> pd.DataFrame:
    """Agregamos duplicados por (pair, mes) y reindexamos a 2024 MS; rellenamos huecos a 0."""
    if df.empty:
        return pd.DataFrame(columns=[*pair_cols, datecol, valuecol])
    g = (df.groupby(pair_cols + [pd.Grouper(key=datecol, freq="MS")], as_index=False)[valuecol]
           .agg(agg))
    months = pd.date_range(start, end, freq="MS")
    out = []
    for keys, sub in g.groupby(pair_cols):
        sub = sub.set_index(datecol).reindex(months).rename_axis(datecol).reset_index()
        for i, k in enumerate(pair_cols):
            sub[k] = keys[i]
        sub[valuecol] = sub[valuecol].fillna(0.0)
        out.append(sub)
    return pd.concat(out, ignore_index=True)[[*pair_cols, datecol, valuecol]]

def _plot_pair(sub_df: pd.DataFrame, bid, ctype, out_dir: str):
    """Generamos la figura por pareja y la guardamos a FIGS además de mostrarla."""
    plt.figure(figsize=(10, 4))
    plt.plot(sub_df[DATECOL], sub_df[f"{VALUE_TRAINTEST}"], marker="o", label="Real 2024")
    plt.plot(sub_df[DATECOL], sub_df[PRED_COL], marker="x", label="Predicción")
    plt.title(f"ID_BUILDING={bid} | FM_COST_TYPE={ctype}")
    plt.xlabel("Fecha")
    plt.ylabel("Coste")
    plt.grid(True)
    plt.legend()
    fname = f"cmp_2024_{bid}_{str(ctype).replace(' ', '_')}.png"
    plt.tight_layout()
    plt.savefig(os.path.join(out_dir, fname))
    plt.show()

# ============================================================
# Bloque 3) Ejecución
# ============================================================

# 1) Cargamos verdad-terreno y predicciones, y normalizamos FECHA
test_df  = pd.read_csv(os.path.join(RUTA_RESULTADOS, "test_full_2024.csv"), sep=CSV_SEP)
preds_df = pd.read_csv(os.path.join(RUTA_RESULTADOS, "preds_por_serie_2024.csv"), sep=CSV_SEP)

test_df  = _ensure_ms(test_df, DATECOL)
preds_df = _ensure_ms(preds_df, DATECOL)

# 2) Alineamos a 2024 y resolvemos duplicados por suma (criterio Paso 0)
test_2024  = test_df[[*PAIR_COLS, DATECOL, VALUE_TRAINTEST]].copy()
test_2024  = _agg_dup_and_reindex(test_2024, DATECOL, VALUE_TRAINTEST,
                                  start="2024-01-01", end="2024-12-01", pair_cols=PAIR_COLS, agg="sum")

preds_2024 = preds_df[[*PAIR_COLS, DATECOL, PRED_COL]].copy()
preds_2024 = (preds_2024
              .groupby(PAIR_COLS + [DATECOL], as_index=False)[PRED_COL].sum())

# 3) Unimos real y pred por pareja+mes y reindexamos a la malla canónica por pareja
merged = (test_2024
          .merge(preds_2024, on=PAIR_COLS + [DATECOL], how="inner"))

# 4) Elegimos 10 parejas al azar de la intersección real∩pred
pairs_all = merged[PAIR_COLS].drop_duplicates()
n_sample = min(10, len(pairs_all))
sample_pairs = pairs_all.sample(n_sample, random_state=42).values.tolist()

# 5) Para cada pareja, reindexamos explícitamente a MONTHS_2024 y graficamos
for bid, ctype in sample_pairs:
    sub = merged[(merged[PAIR_COLS[0]] == bid) & (merged[PAIR_COLS[1]] == ctype)].copy()
    if sub.empty:
        continue
    sub = (sub.set_index(DATECOL)
              .reindex(MONTHS_2024)
              .rename_axis(DATECOL)
              .reset_index())
    # Rellenamos NaN a 0 por consistencia visual si hay huecos
    for col in [VALUE_TRAINTEST, PRED_COL]:
        if col in sub.columns:
            sub[col] = sub[col].fillna(0.0)
    sub[PAIR_COLS[0]] = bid
    sub[PAIR_COLS[1]] = ctype

    _plot_pair(sub, bid, ctype, RUTA_FIGS)

print(f"Graficadas {n_sample} parejas y guardadas en {RUTA_FIGS}")

Graficadas 10 parejas y guardadas en /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3/FIGS

# ============================================================
# ANÁLISIS POST–PASO 8: Visualización de los top-10 mejores pares
# ============================================================


# 1) Cargamos métricas y seleccionamos la vista all_months
metrics_df = pd.read_csv(OUT_PAIR, sep=CSV_SEP)
metrics_all = metrics_df[metrics_df["vista"] == "all_months"].copy()

# 2) Seleccionamos las 10 mejores parejas (menor MASE12)
top10_best = metrics_all.sort_values("MASE12", ascending=True).head(10)[PAIR_COLS]
log8(f"Seleccionadas top-10 parejas con mejor MASE12: {top10_best.values.tolist()}")

# 3) Para cada pareja, trazamos la serie real vs predicha
fig, axes = plt.subplots(5, 2, figsize=(14, 18), sharex=True)
axes = axes.flatten()

for i, (bid, ctype) in enumerate(top10_best.values):
    ax = axes[i]

    # Extraemos serie real
    real = test_2024[(test_2024["ID_BUILDING"]==bid) & (test_2024["FM_COST_TYPE"]==ctype)]
    pred = preds_2024[(preds_2024["ID_BUILDING"]==bid) & (preds_2024["FM_COST_TYPE"]==ctype)]

    if real.empty or pred.empty:
        ax.set_title(f"({bid}, {ctype}) — sin datos")
        continue

    ax.plot(real[DATECOL], real[VALUE_TRAINTEST], marker="o", label="Real 2024")
    ax.plot(pred[DATECOL], pred[PRED_COL], marker="x", linestyle="--", label="Predicción")

    ax.set_title(f"ID={bid} | Tipo={ctype}")
    ax.legend()
    ax.grid(True)

plt.tight_layout()
plt.show()

[2025-09-23T13:36:04] Seleccionadas top-10 parejas con mejor MASE12: [[715, 'Licencias'], [1000547, 'Licencias'], [379, 'Licencias'], [1000434, 'Mtto. Contratos'], [1001159, 'Licencias'], [1059, 'Licencias'], [356, 'Licencias'], [1001161, 'Licencias'], [1114, 'Licencias'], [1000596, 'Licencias']]

# ============================================================
# TOP-10 MEJORES (evitando series triviales a cero) Y GRÁFICOS
# ============================================================


# ---------------------------
# Constantes y parámetros
# ---------------------------
ruta_base_3 = "/content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3"
CSV_SEP     = ";"

# Artefactos del Paso 8 y datos 2024
OUT_PAIR    = os.path.join(ruta_base_3, "METRICAS",  "paso8_metrics_2024_por_pareja.csv")
PATH_TEST   = os.path.join(ruta_base_3, "RESULTADOS","test_full_2024.csv")
PATH_PREDS  = os.path.join(ruta_base_3, "RESULTADOS","preds_por_serie_2024.csv")

# Convenciones de clave y columnas
PAIR_COLS   = ["ID_BUILDING","FM_COST_TYPE"]
DATECOL     = "FECHA"
VALUE_COL   = "cost_float_mod"   # verdad-terreno
PRED_COL    = "yhat_combo"       # predicción del modelo

# Malla 2024
FREQ = "MS"
H    = 12
MONTHS_2024 = pd.date_range("2024-01-01", periods=H, freq=FREQ)

# ---------------------------
# Funciones auxiliares
# ---------------------------

def _ensure_ms(df: pd.DataFrame, datecol: str) -> pd.DataFrame:
    """Normalizamos la columna de fecha a mensual (MS)."""
    df = df.copy()
    df[datecol] = pd.to_datetime(df[datecol]).dt.to_period("M").dt.to_timestamp()
    return df

def _agg_dup_and_reindex(df: pd.DataFrame, start: str, end: str,
                         datecol: str, valuecol: str,
                         pair_cols=PAIR_COLS, agg="sum") -> pd.DataFrame:
    """Agregamos duplicados por (pair, mes) y reindexamos a malla mensual."""
    if df.empty:
        return pd.DataFrame(columns=[*pair_cols, datecol, valuecol])
    g = (df.groupby(pair_cols + [pd.Grouper(key=datecol, freq="MS")], as_index=False)[valuecol]
           .agg(agg))
    all_pairs = g[pair_cols].drop_duplicates()
    months = pd.date_range(start, end, freq="MS")
    out = []
    for a, b in all_pairs.values:
        sub = g[(g[pair_cols[0]]==a) & (g[pair_cols[1]]==b)].set_index(datecol)
        sub = sub.reindex(months).rename_axis(datecol).reset_index()
        sub[pair_cols[0]] = a; sub[pair_cols[1]] = b
        sub[valuecol] = sub[valuecol].fillna(0.0)
        out.append(sub)
    return pd.concat(out, ignore_index=True)[[*pair_cols, datecol, valuecol]]

# ---------------------------
# Carga de datos
# ---------------------------

metrics_df = pd.read_csv(OUT_PAIR, sep=CSV_SEP)
metrics_all = metrics_df[metrics_df["vista"] == "all_months"].copy()

# Leemos verdad-terreno y predicciones para construir las curvas
test_df  = _ensure_ms(pd.read_csv(PATH_TEST,  sep=CSV_SEP), DATECOL)
preds_df = _ensure_ms(pd.read_csv(PATH_PREDS, sep=CSV_SEP), DATECOL)

# Normalizamos y reindexamos a 2024
test_2024 = test_df[[*PAIR_COLS, DATECOL, VALUE_COL]].copy()
test_2024 = _agg_dup_and_reindex(test_2024, "2024-01-01", "2024-12-01", DATECOL, VALUE_COL, PAIR_COLS, agg="sum")

# En preds garantizamos columna PRED_COL
if PRED_COL not in preds_df.columns:
    raise ValueError(f"Esperábamos la columna '{PRED_COL}' en preds_por_serie_2024.csv")

preds_2024 = preds_df[[*PAIR_COLS, DATECOL, PRED_COL]].copy()
preds_2024 = (preds_2024
              .groupby(PAIR_COLS + [DATECOL], as_index=False)
              .agg({PRED_COL: "sum"}))

# ---------------------------
# Filtro para evitar series triviales
# ---------------------------
# Filtramos:
# - tener actividad en 2024 (n_obs_pos > 0)
# - excluir baseline_cero
# - MASE12 válido y positivo
flt = (
    (metrics_all["n_obs_pos"] > 0) &
    (metrics_all["source_model"].str.lower() != "baseline_cero") &
    (metrics_all["MASE12"].notna()) &
    (metrics_all["MASE12"] > 0)
)

candidatas = metrics_all.loc[flt].copy()

# Si también queremos evitar SMAPE = 0 por predicción exacta con todo >0, podríamos añadir:
# candidatas = candidatas[candidatas["SMAPE"] > 0]

# Ordenamos por mejor MASE12 y tomamos top-10
top10 = (candidatas
         .sort_values(["MASE12","SMAPE","WAPE"], ascending=[True, True, True])
         .head(10)
         [[*PAIR_COLS, "MASE12", "SMAPE", "WAPE", "source_model"]]
         .reset_index(drop=True))

print("Top-10 (ya sin series triviales):")
print(top10)

# ---------------------------
# Gráficos comparación y_true vs y_pred
# ---------------------------

n = len(top10)
cols = 2
rows = int(np.ceil(n / cols))

plt.figure(figsize=(12, 3.5*rows))
for i, row in top10.iterrows():
    bid, ctype = row[PAIR_COLS[0]], row[PAIR_COLS[1]]

    # Extraemos y_true e y_pred para 2024
    te_sub = (test_2024[(test_2024[PAIR_COLS[0]]==bid) & (test_2024[PAIR_COLS[1]]==ctype)]
              .set_index(DATECOL)
              .reindex(MONTHS_2024))
    pr_sub = (preds_2024[(preds_2024[PAIR_COLS[0]]==bid) & (preds_2024[PAIR_COLS[1]]==ctype)]
              .set_index(DATECOL)
              .reindex(MONTHS_2024))

    # Rellenamos a 0 donde falte
    y_t = te_sub[VALUE_COL].fillna(0.0).values if VALUE_COL in te_sub.columns else np.zeros(H)
    y_p = pr_sub[PRED_COL].fillna(0.0).values if PRED_COL in pr_sub.columns else np.zeros(H)

    ax = plt.subplot(rows, cols, i+1)
    ax.plot(MONTHS_2024, y_t, label="Real 2024")
    ax.plot(MONTHS_2024, y_p, label="Pred 2024", linestyle="--")
    ax.set_title(f"{bid} | {ctype}\nMASE12={row['MASE12']:.3f} · SMAPE={row['SMAPE']:.1f} · {row['source_model']}")
    ax.set_xlabel("Mes")
    ax.set_ylabel("Coste")
    ax.grid(True)
    ax.legend()

plt.tight_layout()
plt.show()

Top-10 (ya sin series triviales):
   ID_BUILDING     FM_COST_TYPE    MASE12      SMAPE      WAPE  \
0            2            Obras  0.009560  16.666667  1.000000
1         1075  Servicios Ctto.  0.017328   0.249082  0.002490
2      1000009  Mtto. Contratos  0.019284   0.682261  0.006554
3         1084  Servicios Ctto.  0.027392   0.610810  0.006162
4      1000872  Servicios Ctto.  0.031868   1.588242  0.016501
5         1076  Mtto. Contratos  0.036939   0.682261  0.006554
6         1087  Mtto. Contratos  0.036939   0.682261  0.006554
7         1092  Mtto. Contratos  0.036939   0.682261  0.006554
8         1096  Mtto. Contratos  0.036939   0.682261  0.006554
9      1000632  Mtto. Contratos  0.036939   0.682261  0.006554

        source_model
0  fallback_snaive12
1       modelo_final
2       modelo_final
3       modelo_final
4  fallback_snaive12
5       modelo_final
6       modelo_final
7       modelo_final
8       modelo_final
9       modelo_final

##### NO EJECUTAR  ####

# ============================================================
# === Paso 9 - Bloque preliminar: Cargamos dataframe para enriquecer e inspeccionamos las columnas ===
# ============================================================
# ============================================================
# Carga df_fd1_v5_ITE1_train / df_fd1_v5_ITE1_test y verificación de columnas
# ============================================================


# Reutiliza globals si existen
CSV_SEP = globals().get("CSV_SEP", ";")

PATH_E2 = "/content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_2"

def _read_any(path_base: str) -> pd.DataFrame:
    """
    Intenta leer primero .csv (con CSV_SEP) y si no existe, .xlsx.
    Lanza FileNotFoundError si no encuentra ninguno.
    """
    path_csv  = f"{path_base}.csv"
    path_xlsx = f"{path_base}.xlsx"
    if os.path.exists(path_csv):
        return pd.read_csv(path_csv, sep=CSV_SEP)
    if os.path.exists(path_xlsx):
        return pd.read_excel(path_xlsx)
    raise FileNotFoundError(f"No se encontró ni {path_csv} ni {path_xlsx}")

# 1) Cargamos en variables con el mismo nombre del archivo (sin extensión)
df_fd1_v5_ITE1_train = _read_any(os.path.join(PATH_E2, "df_fd1_v5_ITE1_train"))
df_fd1_v5_ITE1_test  = _read_any(os.path.join(PATH_E2, "df_fd1_v5_ITE1_test"))

# 2) Verificamos columnas esperadas
expected_cols = [
    "ID_BUILDING","FM_COST_TYPE","MONTH","YEAR","cost_float_mod",
    "SUPPLIER_TYPE_MOD_2","FM_RESPONSIBLE_MOD","TIPO_USO",
    "ID_REGION_GRUPO","COUNTRY_DEF"
]

missing_train = [c for c in expected_cols if c not in df_fd1_v5_ITE1_train.columns]
missing_test  = [c for c in expected_cols if c not in df_fd1_v5_ITE1_test.columns]

print("[verificación] columnas train faltantes:", missing_train)
print("[verificación] columnas test  faltantes:", missing_test)
print("[verificación] shape train:", df_fd1_v5_ITE1_train.shape,
      "| shape test:", df_fd1_v5_ITE1_test.shape)

[verificación] columnas train faltantes: []
[verificación] columnas test  faltantes: []
[verificación] shape train: (65466, 11) | shape test: (23750, 11)

##### NO EJECUTAR  ####

# df_fd1_v5_ITE1_test.columns

Index(['ID_BUILDING', 'FM_COST_TYPE', 'FECHA', 'YEAR', 'MONTH',
       'cost_float_mod', 'COUNTRY_DEF', 'ID_REGION_GRUPO', 'TIPO_USO',
       'SUPPLIER_TYPE_MOD_2', 'FM_RESPONSIBLE_MOD'],
      dtype='object')

##### NO EJECUTAR  ####

# ============================================================
# Construcción de dim_buildings.csv (solo con TEST y TRAIN con dominio de TEST sobre TRAIN)
# ============================================================

# Reutiliza globals si existen
ruta_base_3 = globals().get("ruta_base_3", "/content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3")
CSV_SEP     = globals().get("CSV_SEP", ";")

RUTA_METRICAS = os.path.join(ruta_base_3, "METRICAS")
os.makedirs(RUTA_METRICAS, exist_ok=True)
PATH_DIM_OUT  = os.path.join(RUTA_METRICAS, "dim_buildings.csv")

# --- Utilidades ---
def _norm_keys(df: pd.DataFrame) -> pd.DataFrame:
    if df is None or df.empty:
        return df
    out = df.copy()
    if "ID_BUILDING" in out.columns:
        try:
            out["ID_BUILDING"] = pd.to_numeric(out["ID_BUILDING"], errors="coerce").astype("Int64")
        except Exception:
            pass
    for c in out.select_dtypes(include="object").columns:
        out[c] = out[c].astype(str).str.strip()
    return out

def _pick_first_nonnull(s):
    return next((v for v in s if pd.notna(v) and str(v).strip() != ""), np.nan)

# --- Normalizamos fuentes (ya cargadas previamente) ---
train_enr = _norm_keys(df_fd1_v5_ITE1_train.copy())
test_enr  = _norm_keys(df_fd1_v5_ITE1_test.copy())

# Universo de edificios (incluye los que solo aparecen en 2024)
ids = (pd.concat([train_enr[["ID_BUILDING"]], test_enr[["ID_BUILDING"]]], ignore_index=True)
         .drop_duplicates()
         .dropna(subset=["ID_BUILDING"]))

# Columnas disponibles en ambos datasets (según tu lista)
cols_meta = [
    "COUNTRY_DEF",          # país
    "ID_REGION_GRUPO",      # región/código
    "TIPO_USO",             # uso edificio
    "FM_RESPONSIBLE_MOD",   # responsable FM
    "SUPPLIER_TYPE_MOD_2"   # tipo proveedor
]

def _collapse_meta(df: pd.DataFrame, cols: list[str]) -> pd.DataFrame:
    keep = ["ID_BUILDING"] + [c for c in cols if c in df.columns]
    agg = {c: _pick_first_nonnull for c in keep if c != "ID_BUILDING"}
    return (df[keep].groupby("ID_BUILDING", as_index=False).agg(agg))

# Agregamos por edificio en cada fuente
meta_train = _collapse_meta(train_enr, cols_meta)
meta_test  = _collapse_meta(test_enr,  cols_meta)

# Coalesce por edificio: TEST ≻ TRAIN
meta_test_idx  = meta_test.set_index("ID_BUILDING")
meta_train_idx = meta_train.set_index("ID_BUILDING")
meta = meta_test_idx.combine_first(meta_train_idx).reset_index()

# Construimos la dimensión con nombres canónicos esperados por Paso 9
dim = meta.rename(columns={
    "COUNTRY_DEF":          "PAIS",
    "ID_REGION_GRUPO":      "REGION",
    "TIPO_USO":             "TIPO_USO",
    "FM_RESPONSIBLE_MOD":   "FM_RESP",
    "SUPPLIER_TYPE_MOD_2":  "SUPPLIER"
})[["ID_BUILDING", "PAIS", "REGION", "TIPO_USO", "FM_RESP", "SUPPLIER"]]

# Normalizaciones ligeras
def _norm_upper(x):  return str(x).upper() if pd.notna(x) else x
def _norm_title(x):
    try:
        s = str(x).strip()
        return s.title() if s else np.nan
    except Exception:
        return x

if "PAIS" in dim.columns:
    dim["PAIS"] = dim["PAIS"].apply(_norm_upper)
if "REGION" in dim.columns:
    dim["REGION"] = dim["REGION"].astype(str).str.strip()  # suele ser código; sin title-case
if "TIPO_USO" in dim.columns:
    dim["TIPO_USO"] = dim["TIPO_USO"].apply(_norm_title)

# Guardado
dim_out = dim.drop_duplicates().reset_index(drop=True)
dim_out.to_csv(PATH_DIM_OUT, sep=CSV_SEP, index=False)
print(f"[dim_buildings] Guardado: {PATH_DIM_OUT} con {len(dim_out)} edificios.")
print("[dim_buildings] Columnas:", list(dim_out.columns))

[dim_buildings] Guardado: /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3/METRICAS/dim_buildings.csv con 561 edificios.
[dim_buildings] Columnas: ['ID_BUILDING', 'PAIS', 'REGION', 'TIPO_USO', 'FM_RESP', 'SUPPLIER']

##### NO EJECUTAR  ####

# ============================================================
# Construcción de dim_buildings.csv completa - con catálogo FM que domina; fallback con dominio de TEST sobre TRAIN)
# ============================================================


CSV_SEP = globals().get("CSV_SEP", ";")
ruta_base_3 = globals().get("ruta_base_3", "/content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3")
RUTA_METRICAS = os.path.join(ruta_base_3, "METRICAS")
os.makedirs(RUTA_METRICAS, exist_ok=True)

# Carpeta del catálogo FM (tu ruta)
CAT_DIR = "/content/drive/MyDrive/PROYECTO_NEITH/FUENTE_DATOS/CATALOGOS"
# Nombre típico del archivo (ajústalo si cambia en el futuro)
CAT_FILENAME = "Catalogo_inmuebles_Actualizado_a_092025.xlsx"
PATH_FM_CAT = os.path.join(CAT_DIR, CAT_FILENAME)

# Si el archivo exacto no existe, tomamos el último .xlsx del directorio como fallback
if not os.path.exists(PATH_FM_CAT):
    cand = sorted(glob.glob(os.path.join(CAT_DIR, "*.xlsx")))
    PATH_FM_CAT = cand[-1] if cand else PATH_FM_CAT

# Dim previa (opcional, como fallback)
PATH_DIM_PREV = os.path.join(RUTA_METRICAS, "dim_buildings.csv")

# Salida final (mantenemos el mismo nombre para compatibilidad con Paso 10)
PATH_DIM_FINAL = os.path.join(RUTA_METRICAS, "dim_buildings.csv")

def _norm_id_building(s):
    s = pd.to_numeric(s, errors="coerce")
    return s.astype("Int64")

def _norm_upper(x):
    try:
        return str(x).strip().upper() if pd.notna(x) else np.nan
    except Exception:
        return np.nan

def _norm_str(x):
    try:
        s = str(x).strip()
        return s if s else np.nan
    except Exception:
        return np.nan

# 1) Leemos catálogo FM (si existe)
if os.path.exists(PATH_FM_CAT):
    fm = pd.read_excel(PATH_FM_CAT)
    fm.columns = [c.strip() for c in fm.columns]
    keep_map = {
        "ID_BUILDING": "ID_BUILDING",
        "ID_REGION":   "REGION",   # renombramos a REGION
        "COUNTRY":     "PAIS",     # renombramos a PAIS
        "TIPO_USO":    "TIPO_USO"
    }
    fm = fm[[c for c in keep_map if c in fm.columns]].rename(columns=keep_map)
    if "ID_BUILDING" in fm.columns:
        fm["ID_BUILDING"] = _norm_id_building(fm["ID_BUILDING"])
    if "PAIS" in fm.columns:
        fm["PAIS"] = fm["PAIS"].apply(_norm_upper)
    if "REGION" in fm.columns:
        fm["REGION"] = fm["REGION"].apply(_norm_str)
    if "TIPO_USO" in fm.columns:
        fm["TIPO_USO"] = fm["TIPO_USO"].apply(_norm_str)
    fm = fm.drop_duplicates(subset=["ID_BUILDING"]).reset_index(drop=True)
else:
    fm = pd.DataFrame(columns=["ID_BUILDING","PAIS","REGION","TIPO_USO"])

# 2) Leemos dim previa (si existe) para buildings que no estén en el catálogo
if os.path.exists(PATH_DIM_PREV):
    prev = pd.read_csv(PATH_DIM_PREV, sep=CSV_SEP)
    prev.columns = [c.strip() for c in prev.columns]
    rename_prev = {}
    for c in prev.columns:
        cu = c.strip().upper()
        if cu == "ID_BUILDING": rename_prev[c] = "ID_BUILDING"
        if cu == "PAIS":        rename_prev[c] = "PAIS"
        if cu == "REGION":      rename_prev[c] = "REGION"
        if cu == "TIPO_USO":    rename_prev[c] = "TIPO_USO"
    prev = prev.rename(columns=rename_prev)
    if "ID_BUILDING" in prev.columns:
        prev["ID_BUILDING"] = _norm_id_building(prev["ID_BUILDING"])
    for c in ["PAIS","REGION","TIPO_USO"]:
        if c in prev.columns:
            prev[c] = prev[c].apply(_norm_str)
else:
    prev = pd.DataFrame(columns=["ID_BUILDING","PAIS","REGION","TIPO_USO"])

# 3) Combinamos: catálogo FM domina; añadimos solo los ID_BUILDING que falten
if not fm.empty and not prev.empty:
    ids_fm = set(fm["ID_BUILDING"].dropna().tolist())
    prev_only = prev[~prev["ID_BUILDING"].isin(ids_fm)]
    dim_final = pd.concat([fm, prev_only], ignore_index=True)
elif not fm.empty:
    dim_final = fm.copy()
else:
    dim_final = prev.copy()

# 4) Limpieza final y guardado
dim_final = dim_final[["ID_BUILDING","PAIS","REGION","TIPO_USO"]].drop_duplicates()
dim_final = dim_final.sort_values("ID_BUILDING").reset_index(drop=True)
dim_final.to_csv(PATH_DIM_FINAL, sep=CSV_SEP, index=False)

# 5) Informe rápido
total = len(dim_final)
n_pais_null   = dim_final["PAIS"].isna().sum() if "PAIS" in dim_final.columns else total
n_region_null = dim_final["REGION"].isna().sum() if "REGION" in dim_final.columns else total
print("[dim híbrida] Fuente catálogo:", os.path.basename(PATH_FM_CAT) if os.path.exists(PATH_FM_CAT) else "NO ENCONTRADA")
print("[dim híbrida] Guardado:", PATH_DIM_FINAL, "filas:", total)
print(f"[dim híbrida] %PAIS nulo:   {100*n_pais_null/max(total,1):.2f}%")
print(f"[dim híbrida] %REGION nulo: {100*n_region_null/max(total,1):.2f}%")

[dim híbrida] Fuente catálogo: Catalogo_inmuebles_Actualizado_a_092025.xlsx
[dim híbrida] Guardado: /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3/METRICAS/dim_buildings.csv filas: 1650
[dim híbrida] %PAIS nulo:   0.00%
[dim híbrida] %REGION nulo: 0.00%

#### dim_final.head()

#### NO EJECUTAR ####
# ============================================================
# Construcción de dim_buildings.csv (catálogo ≻ fallback TEST≻TRAIN)
# ============================================================

# Reutiliza globals si existen
ruta_base_3 = globals().get("ruta_base_3", "/content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3")
CSV_SEP     = globals().get("CSV_SEP", ";")

RUTA_METRICAS = os.path.join(ruta_base_3, "METRICAS")
os.makedirs(RUTA_METRICAS, exist_ok=True)
PATH_DIM_OUT  = os.path.join(RUTA_METRICAS, "dim_buildings.csv")

# Ruta del catálogo FM
PATH_CATALOGO = "/content/drive/MyDrive/PROYECTO_NEITH/FUENTE_DATOS/CATALOGOS/Catalogo_inmuebles_Actualizado_a_092025.xlsx"

# Logger local del bloque (usa log9 si ya existe)
def _dim_log(msg: str):
    tag = "[dim_buildings]"
    if "log9" in globals() and callable(globals()["log9"]):
        globals()["log9"](f"{tag} {msg}")
    else:
        print(f"{tag} {msg}")

# ---------- Normalizadores vectorizados seguros ----------
import numpy as np
import pandas as pd

def _to_upper_series(s: pd.Series) -> pd.Series:
    s = s.astype("string").str.strip()
    s = s.mask(s.eq("") | s.str.lower().isin(["nan", "none"]))
    return s.str.upper()

def _to_title_series(s: pd.Series) -> pd.Series:
    s = s.astype("string").str.strip()
    s = s.mask(s.eq("") | s.str.lower().isin(["nan", "none"]))
    return s.apply(lambda x: x.title() if isinstance(x, str) else np.nan)

def _norm_id_series(s: pd.Series) -> pd.Series:
    return pd.to_numeric(s, errors="coerce").astype("Int64")

# ---------- Intento 1: construir desde CATALOGO ----------
dim_out = pd.DataFrame()
_catalog_ok = False

try:
    if os.path.exists(PATH_CATALOGO):
        cat = pd.read_excel(PATH_CATALOGO)
        # Limpieza de columnas y eliminación de completamente vacías
        cat.columns = [str(c).strip() for c in cat.columns]
        empty_cols = [c for c in cat.columns if cat[c].isna().all()]
        if empty_cols:
            cat = cat.drop(columns=empty_cols)

        _dim_log(f"[{pd.Timestamp.now().isoformat(timespec='seconds')}] Catálogo cargado: {len(cat)} filas. Columnas: {list(cat.columns)}")

        # Consolidamos ID_BUILDING si viene duplicado como 'ID_BUILDING.1'
        if "ID_BUILDING.1" in cat.columns:
            idc  = pd.to_numeric(cat.get("ID_BUILDING"), errors="coerce")
            idc2 = pd.to_numeric(cat.get("ID_BUILDING.1"), errors="coerce")
            cat["ID_BUILDING"] = idc.fillna(idc2)

        # Chequeo de columnas mínimas del catálogo
        needed = {"ID_BUILDING","ID_REGION","COUNTRY","STATUS","TIPO_USO"}
        if not needed.issubset(set(cat.columns)):
            faltan = sorted(list(needed - set(cat.columns)))
            _dim_log(f"AVISO: el catálogo no tiene columnas mínimas {faltan}. Intentaremos fallback TEST≻TRAIN.")
        else:
            # Tipamos y normalizamos
            cat["ID_BUILDING"] = _norm_id_series(cat["ID_BUILDING"])
            dim_raw = pd.DataFrame({
                "ID_BUILDING": cat["ID_BUILDING"],
                "PAIS":        _to_upper_series(cat["COUNTRY"]),
                "REGION":      cat["ID_REGION"].astype("string").str.strip().mask(lambda x: x.eq("") | x.str.lower().isin(["nan","none"])),
                "TIPO_USO":    _to_title_series(cat["TIPO_USO"]),
                # STATUS binario para priorizar activos
                "STATUS_BIN":  cat["STATUS"].astype("string").str.strip().isin(["1","1.0","true","True","ACTIVO","Activo"]).astype(int)
            })

            # Deduplicamos por ID_BUILDING priorizando activos (STATUS=1)
            dim_out = (dim_raw
                       .sort_values(["ID_BUILDING","STATUS_BIN"], ascending=[True, False])
                       .drop_duplicates(subset=["ID_BUILDING"], keep="first")
                       .drop(columns=["STATUS_BIN"])
                       .dropna(subset=["ID_BUILDING"])
                       .reset_index(drop=True))

            _catalog_ok = len(dim_out) > 0

            _dim_log(f"Construido desde catálogo: {len(dim_out)} edificios únicos.")
            _dim_log(f"Audit nulos — PAIS={100*dim_out['PAIS'].isna().mean():.1f}% | REGION={100*dim_out['REGION'].isna().mean():.1f}% | TIPO_USO={100*dim_out['TIPO_USO'].isna().mean():.1f}%")
    else:
        _dim_log(f"AVISO: no existe el fichero de catálogo en {PATH_CATALOGO}. Intentaremos fallback TEST≻TRAIN.")
except Exception as e:
    _dim_log(f"AVISO: error construyendo desde catálogo ({e}). Intentaremos fallback TEST≻TRAIN.")

# ---------- Intento 2 (fallback): TEST ≻ TRAIN como tenías ----------
if not _catalog_ok:
    _dim_log("Activando fallback TEST≻TRAIN para dim_buildings.csv.")

    def _norm_keys(df: pd.DataFrame) -> pd.DataFrame:
        if df is None or df.empty:
            return df
        out = df.copy()
        if "ID_BUILDING" in out.columns:
            try:
                out["ID_BUILDING"] = pd.to_numeric(out["ID_BUILDING"], errors="coerce").astype("Int64")
            except Exception:
                pass
        for c in out.select_dtypes(include="object").columns:
            out[c] = out[c].astype(str).str.strip()
        return out

    def _pick_first_nonnull(s):
        return next((v for v in s if pd.notna(v) and str(v).strip() != ""), np.nan)

    # Necesitamos df_fd1_v5_ITE1_train / test ya cargados en memoria (como en tu Paso 9)
    if not all(name in globals() for name in ["df_fd1_v5_ITE1_train","df_fd1_v5_ITE1_test"]):
        raise RuntimeError("Fallback TEST≻TRAIN requiere df_fd1_v5_ITE1_train y df_fd1_v5_ITE1_test ya cargados en memoria.")

    train_enr = _norm_keys(globals()["df_fd1_v5_ITE1_train"].copy())
    test_enr  = _norm_keys(globals()["df_fd1_v5_ITE1_test"].copy())

    cols_meta = [
        "COUNTRY_DEF",          # país
        "ID_REGION_GRUPO",      # región/código
        "TIPO_USO",             # uso edificio
        "FM_RESPONSIBLE_MOD",   # responsable FM
        "SUPPLIER_TYPE_MOD_2"   # tipo proveedor
    ]

    def _collapse_meta(df: pd.DataFrame, cols: list[str]) -> pd.DataFrame:
        keep = ["ID_BUILDING"] + [c for c in cols if c in df.columns]
        agg = {c: _pick_first_nonnull for c in keep if c != "ID_BUILDING"}
        return (df[keep].groupby("ID_BUILDING", as_index=False).agg(agg))

    meta_train = _collapse_meta(train_enr, cols_meta)
    meta_test  = _collapse_meta(test_enr,  cols_meta)

    meta = meta_test.set_index("ID_BUILDING").combine_first(meta_train.set_index("ID_BUILDING")).reset_index()

    dim_out = meta.rename(columns={
        "COUNTRY_DEF":          "PAIS",
        "ID_REGION_GRUPO":      "REGION",
        "TIPO_USO":             "TIPO_USO",
        "FM_RESPONSIBLE_MOD":   "FM_RESP",
        "SUPPLIER_TYPE_MOD_2":  "SUPPLIER"
    })[["ID_BUILDING", "PAIS", "REGION", "TIPO_USO", "FM_RESP", "SUPPLIER"]]

    # Normalizaciones suaves
    dim_out["PAIS"]    = _to_upper_series(dim_out["PAIS"]) if "PAIS" in dim_out.columns else dim_out.get("PAIS")
    if "REGION" in dim_out.columns:
        dim_out["REGION"] = dim_out["REGION"].astype("string").str.strip()
    if "TIPO_USO" in dim_out.columns:
        dim_out["TIPO_USO"] = _to_title_series(dim_out["TIPO_USO"])

    dim_out = dim_out.drop_duplicates().reset_index(drop=True)
    _dim_log(f"Fallback TEST≻TRAIN construido: {len(dim_out)} edificios.")

# ---------- Guardado ----------
dim_out.to_csv(PATH_DIM_OUT, sep=CSV_SEP, index=False)
_dim_log(f"Guardado: {PATH_DIM_OUT} con {len(dim_out)} edificios.")
_dim_log(f"Columnas finales: {list(dim_out.columns)}")

[dim_buildings] [2025-09-24T12:17:53] [dim_buildings] [2025-09-24T12:17:53] Catálogo cargado: 1650 filas. Columnas: ['ID_BUILDING', 'ID_REGION', 'COUNTRY', 'STATUS', 'FM_PERIMETER', 'SERVICE_LEVEL', 'TIPO_USO', 'M2_AVAIL_PERIMETER', 'M2_AVAIL', 'M2_PROM_USO', 'ID_BUILDING.1']
[dim_buildings] [2025-09-24T12:17:53] [dim_buildings] Construido desde catálogo: 1650 edificios únicos.
[dim_buildings] [2025-09-24T12:17:53] [dim_buildings] Audit nulos — PAIS=0.0% | REGION=0.0% | TIPO_USO=0.0%
[dim_buildings] [2025-09-24T12:17:53] [dim_buildings] Guardado: /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3/METRICAS/dim_buildings.csv con 1650 edificios.
[dim_buildings] [2025-09-24T12:17:53] [dim_buildings] Columnas finales: ['ID_BUILDING', 'PAIS', 'REGION', 'TIPO_USO']

# ============================================================
# Bloque Preliminar: Construcción de dim_buildings.csv DESDE CATÁLOGO
# ============================================================

# Rutas (mantenemos nombres de la Estrategia 3)
ruta_base_3 = "/content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3"
CSV_SEP     = ";"

RUTA_METRICAS = os.path.join(ruta_base_3, "METRICAS")
os.makedirs(RUTA_METRICAS, exist_ok=True)
PATH_DIM_OUT  = os.path.join(RUTA_METRICAS, "dim_buildings.csv")

# Catálogo fuente (status actual de los inmuebles)
CATALOGO_DIR  = "/content/drive/MyDrive/PROYECTO_NEITH/FUENTE_DATOS/CATALOGOS"
CATALOGO_XLSX = os.path.join(CATALOGO_DIR, "Catalogo_inmuebles_Actualizado_a_092025.xlsx")

# -----------------------
# Helpers de normalización
# -----------------------
def _norm_id_building(series: pd.Series) -> pd.Series:
    # Tipamos a entero nullable para conservar posibles NaNs
    return pd.to_numeric(series, errors="coerce").astype("Int64")

def _norm_upper_series(series: pd.Series) -> pd.Series:
    s = series.astype(str).str.strip()
    s = s.replace({"": np.nan, "nan": np.nan, "None": np.nan})
    return s.str.upper()

def _norm_title_series(series: pd.Series) -> pd.Series:
    # Title-case seguro; si viene NaN lo dejamos como tal
    return series.apply(lambda x: str(x).strip().title() if pd.notna(x) and str(x).strip() != "" else np.nan)

def _norm_status(series: pd.Series) -> pd.Series:
    # Mapeamos a 0/1 de forma robusta
    s = pd.to_numeric(series, errors="coerce")
    s = s.fillna(0).astype(int)
    s = s.clip(lower=0, upper=1)
    return s

# ---------------
# Carga Catálogo
# ---------------
cat = pd.read_excel(CATALOGO_XLSX)

# Eliminamos columnas completamente vacías (p.ej. 'Unnamed: 10', etc.)
empty_cols = [c for c in cat.columns if cat[c].isna().all()]
cat = cat.drop(columns=empty_cols)

print(f"[dim_buildings] [{datetime.now().isoformat(timespec='seconds')}] "
      f"Catálogo cargado: {len(cat)} filas. Columnas: {list(cat.columns)}")

# Renombrado y selección de columnas canónicas
# (el catálogo trae: ID_BUILDING, ID_REGION, COUNTRY, STATUS, TIPO_USO, ...)
cols_needed = ["ID_BUILDING", "COUNTRY", "ID_REGION", "TIPO_USO", "STATUS"]
missing = [c for c in cols_needed if c not in cat.columns]
if missing:
    raise KeyError(f"Faltan columnas en el catálogo: {missing}. Revisa {CATALOGO_XLSX}")

# Normalización de valores
dim_raw = pd.DataFrame({
    "ID_BUILDING": _norm_id_building(cat["ID_BUILDING"]),
    "PAIS":        _norm_upper_series(cat["COUNTRY"]),
    "REGION":      cat["ID_REGION"].astype(str).str.strip().replace({"": np.nan}),
    "TIPO_USO":    _norm_title_series(cat["TIPO_USO"]),
    "STATUS":      _norm_status(cat["STATUS"]),
})

# Quitamos filas sin ID_BUILDING y deduplicamos por seguridad (no se espera histórico aquí)
dim_out = (dim_raw
           .dropna(subset=["ID_BUILDING"])
           .drop_duplicates(subset=["ID_BUILDING"])
           .reset_index(drop=True))

# Guardado
dim_out.to_csv(PATH_DIM_OUT, sep=CSV_SEP, index=False)

# Resumen final
n_total   = len(dim_out)
n_activos = int(dim_out["STATUS"].eq(1).sum())
n_baja    = int(dim_out["STATUS"].eq(0).sum())
print(f"[dim_buildings] Guardado: {PATH_DIM_OUT} con {n_total} edificios. "
      f"Activos={n_activos} | Baja={n_baja}")
print("[dim_buildings] Columnas:", list(dim_out.columns))

[dim_buildings] [2025-09-24T12:35:32] Catálogo cargado: 1650 filas. Columnas: ['ID_BUILDING', 'ID_REGION', 'COUNTRY', 'STATUS', 'FM_PERIMETER', 'SERVICE_LEVEL', 'TIPO_USO', 'M2_AVAIL_PERIMETER', 'M2_AVAIL', 'M2_PROM_USO', 'ID_BUILDING.1']
[dim_buildings] Guardado: /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3/METRICAS/dim_buildings.csv con 1650 edificios. Activos=826 | Baja=824
[dim_buildings] Columnas: ['ID_BUILDING', 'PAIS', 'REGION', 'TIPO_USO', 'STATUS']

# ============================================================
# PASO 9 · Evaluación final y tableros de métricas (negocio/dirección)
# Bloque 0) Parámetros, rutas y logging
# ============================================================


# Fijamos semilla para reproducibilidad
np.random.seed(7)

# Reutilizamos variables globales del proyecto si ya existen; si no, ponemos defaults seguros
ruta_base_3 = globals().get("ruta_base_3", "/content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3")
CSV_SEP     = globals().get("CSV_SEP", ";")
PAIR_COLS   = globals().get("PAIR_COLS", ["ID_BUILDING", "FM_COST_TYPE"])
DATECOL     = globals().get("DATECOL", "FECHA")
VALUE_TRAINTEST = globals().get("VALUE_TRAINTEST", "cost_float_mod")

# Frecuencia y malla canónica
FREQ = "MS"
H    = 12
MONTHS_2024 = pd.date_range("2024-01-01", periods=H, freq=FREQ)

# Carpeta de trabajo
RUTA_RESULTADOS = os.path.join(ruta_base_3, "RESULTADOS")
RUTA_METRICAS   = os.path.join(ruta_base_3, "METRICAS")
RUTA_LOGS       = os.path.join(ruta_base_3, "LOGS")
RUTA_REPORTING  = os.path.join(ruta_base_3, "REPORTING")

for _d in [RUTA_RESULTADOS, RUTA_METRICAS, RUTA_LOGS, RUTA_REPORTING]:
    os.makedirs(_d, exist_ok=True)

# Entradas
PATH_TRAIN   = os.path.join(RUTA_RESULTADOS, "train_full_2021_2023.csv")
PATH_TEST    = os.path.join(RUTA_RESULTADOS, "test_full_2024.csv")
PATH_PREDS   = os.path.join(RUTA_RESULTADOS, "preds_por_serie_2024.csv")
PATH_FINAL   = os.path.join(RUTA_METRICAS,   "modelo_final_paso6.csv")
PATH_PASO8   = os.path.join(RUTA_METRICAS,   "paso8_metrics_2024_por_pareja.csv")  # insumo opcional
PATH_BASES   = os.path.join(RUTA_METRICAS,   "metrics_baselines_cv.csv")           # opcional
PATH_DIM     = os.path.join(RUTA_METRICAS,   "dim_buildings.csv")                  # opcional

# Salidas
OUT_SERIES    = os.path.join(RUTA_METRICAS,  "metrics_series_mensual.csv")
OUT_SEGMENTOS = os.path.join(RUTA_METRICAS,  "metrics_segmento_mensual.csv")
OUT_PORTFOLIO = os.path.join(RUTA_METRICAS,  "metrics_portfolio_mensual.csv")
OUT_TOP10     = os.path.join(RUTA_REPORTING, "top10_series_contrib_error.csv")
OUT_TOP5      = os.path.join(RUTA_REPORTING, "top5_segmentos_rojos_WAPE.csv")
OUT_SUBSET    = os.path.join(RUTA_METRICAS,  "metrics_subset_estable.csv")
OUT_COMPARE_MOD3 = os.path.join(RUTA_REPORTING, "compare_portfolio_mod3.csv")

# Logging
LOG_PATH_STEP9 = os.path.join(RUTA_LOGS, "estrategia3_step9.log")

def log9(msg: str):
    """Escribimos mensajes del Paso 9 en el log y en consola para trazabilidad."""
    stamp = datetime.now().isoformat(timespec="seconds")
    line = f"[{stamp}] {msg}"
    print(line)
    with open(LOG_PATH_STEP9, "a", encoding="utf-8") as f:
        f.write(line + "\n")

# ============================================================
# Bloque 1) Utilidades de carga, fechas y normalización
# ============================================================

EPS = 1e-8

def _ensure_ms(df: pd.DataFrame, datecol: str) -> pd.DataFrame:
    # Normalizamos la fecha a comienzo de mes para alinear todo a la malla MS
    df = df.copy()
    df[datecol] = pd.to_datetime(df[datecol]).dt.to_period("M").dt.to_timestamp()
    return df

def _read_csv(path: str, required: bool=True, sep: str=CSV_SEP) -> pd.DataFrame:
    if os.path.exists(path):
        df = pd.read_csv(path, sep=sep)
        # Normalización fuerte de nombres de columnas (espacios/BOM)
        df.columns = [str(c).strip().lstrip("\ufeff").rstrip() for c in df.columns]
        return df
    if required:
        raise FileNotFoundError(f"Falta el archivo requerido: {path}")
    log9(f"AVISO: no encontramos {os.path.basename(path)} (continuamos sin este artefacto).")
    return pd.DataFrame()

def _norm_pair_keys(df: pd.DataFrame) -> pd.DataFrame:
    """Normaliza claves y asegura que existan exactamente 'ID_BUILDING' y 'FM_COST_TYPE'."""
    if df is None or df.empty:
        return df

    out = df.copy()

    # 2.1 Limpia nombres de columnas
    out.columns = [str(c).strip().lstrip("\ufeff").rstrip() for c in out.columns]

    # 2.2 Detecta y renombra alias (por si aparece con otro case o con espacios)
    cols_lower = {c.lower(): c for c in out.columns}
    if "id_building" in cols_lower and "ID_BUILDING" not in out.columns:
        out = out.rename(columns={cols_lower["id_building"]: "ID_BUILDING"})
    if "fm_cost_type" in cols_lower and "FM_COST_TYPE" not in out.columns:
        out = out.rename(columns={cols_lower["fm_cost_type"]: "FM_COST_TYPE"})

    # 2.3 Tipea y limpia valores
    if "ID_BUILDING" in out.columns:
        try:
            out["ID_BUILDING"] = pd.to_numeric(out["ID_BUILDING"], errors="coerce").astype("Int64")
        except Exception:
            pass
    if "FM_COST_TYPE" in out.columns and out["FM_COST_TYPE"].dtype == object:
        out["FM_COST_TYPE"] = out["FM_COST_TYPE"].astype(str).str.strip()

    return out


def _agg_dup_and_reindex(df: pd.DataFrame, datecol: str, valuecol: str,
                         months: pd.DatetimeIndex, pair_cols=PAIR_COLS, agg="sum") -> pd.DataFrame:
    # Agregamos duplicados por (pair, mes) y reindexamos a la malla fija de meses
    if df.empty:
        return pd.DataFrame(columns=[*pair_cols, datecol, valuecol])
    g = (df.groupby(pair_cols + [pd.Grouper(key=datecol, freq="MS")], as_index=False)[valuecol]
           .agg(agg))
    out = []
    for a, b in g[pair_cols].drop_duplicates().values:
        sub = g[(g[pair_cols[0]]==a) & (g[pair_cols[1]]==b)].set_index(datecol)
        sub = sub.reindex(months).rename_axis(datecol).reset_index()
        sub[pair_cols[0]] = a; sub[pair_cols[1]] = b
        sub[valuecol] = sub[valuecol].fillna(np.nan)  # dejamos NaN para cobertura
        out.append(sub)
    return pd.concat(out, ignore_index=True)[[*pair_cols, datecol, valuecol]]

# ============================================================
# Bloque 2) Métricas y escalas MASE
# ============================================================

def _safe_div(num, den, eps=EPS):
    return num / (den + eps)

def mae_vec(y_true, y_pred):
    # MAE como media de errores absolutos en el conjunto evaluado
    y_true = np.asarray(y_true, dtype=float)
    y_pred = np.asarray(y_pred, dtype=float)
    return float(np.mean(np.abs(y_pred - y_true)))

def wape_micro(y_true, y_pred, eps=EPS):
    # WAPE micro: sumatorio de |e| sobre sumatorio de |y|
    y_true = np.asarray(y_true, dtype=float)
    y_pred = np.asarray(y_pred, dtype=float)
    return float(_safe_div(np.sum(np.abs(y_pred - y_true)), np.sum(np.abs(y_true)), eps))

def smape_vec(y_true, y_pred, eps=EPS):
    # SMAPE medio con denominador simétrico y eps para evitar divisiones por 0
    y_true = np.asarray(y_true, dtype=float)
    y_pred = np.asarray(y_pred, dtype=float)
    num = np.abs(y_pred - y_true)
    den = (np.abs(y_true) + np.abs(y_pred)) / 2.0
    return float(np.mean(_safe_div(num, den, eps)) * 100.0)

def _scale_diff(series: np.ndarray, lag: int):
    # Escala para MASE: media de |y_t - y_{t-lag}| sobre el histórico
    arr = np.asarray(series, dtype=float)
    if len(arr) <= lag:
        return np.nan
    diffs = np.abs(arr[lag:] - arr[:-lag])
    if diffs.size == 0:
        return np.nan
    return float(np.mean(diffs))

def mase_from_scale(errors: np.ndarray, scale: float):
    # MASE con escala precomputada; si la escala no es válida, devolvemos NaN
    if scale is None or np.isnan(scale) or scale == 0:
        return np.nan
    errors = np.asarray(errors, dtype=float)
    return float(np.mean(np.abs(errors)) / float(scale))

def mase_point_abs_err(abs_err: float, scale: float):
    # MASE puntual por fila (lo usamos para MASE1/MASE12 en serie–mes)
    if scale is None or np.isnan(scale) or scale == 0:
        return np.nan
    return float(abs_err / float(scale))

# ============================================================
# Bloque 3) Baseline SNaive12 para Control F (último año completo -> mediana estacional -> último valor)
# ============================================================

def _last_complete_year_map(y: np.ndarray, dates: np.ndarray):
    # Buscamos el último año con los 12 meses completos y devolvemos mapa mes->valor
    if y.size == 0 or dates.size == 0:
        return {}
    df = pd.DataFrame({"FECHA": pd.to_datetime(dates), "y": y})
    df["YEAR"] = df["FECHA"].dt.year
    df["MONTH"] = df["FECHA"].dt.month
    years = sorted(df["YEAR"].unique())[::-1]
    for year in years:
        sub = df[df["YEAR"] == year]
        if sub["MONTH"].nunique() == 12:
            return sub.set_index("MONTH")["y"].to_dict()
    return {}

def _median_by_month(y: np.ndarray, dates: np.ndarray):
    if y.size == 0 or dates.size == 0:
        return {}
    df = pd.DataFrame({"FECHA": pd.to_datetime(dates), "y": y})
    df["MONTH"] = df["FECHA"].dt.month
    return df.groupby("MONTH")["y"].median().to_dict()

def snaive12_forecast(y_train: np.ndarray, dates_train: np.ndarray, horizon: int):
    # Implementamos SNaive12 con la misma lógica del proyecto
    if horizon <= 0:
        return np.zeros(0, dtype=float)
    if y_train.size == 0:
        return np.zeros(horizon, dtype=float)
    last_date = pd.to_datetime(dates_train[-1])
    future = pd.date_range(last_date.to_period("M").to_timestamp() + pd.offsets.MonthBegin(1),
                           periods=horizon, freq="MS")
    last_y = float(y_train[-1])

    m = _last_complete_year_map(y_train, dates_train)
    if len(m) == 12:
        return np.array([m.get(d.month, last_y) for d in future], dtype=float)

    med = _median_by_month(y_train, dates_train)
    if len(med) > 0:
        return np.array([med.get(d.month, last_y) for d in future], dtype=float)

    return np.repeat(last_y, horizon).astype(float)

# ============================================================
# Bloque 4) Carga y preparación de datos  (con validación)
# ============================================================

# --- Utilidad de validación ---
def _ensure_pair_cols(df: pd.DataFrame, name: str, require: list[str] | None = None) -> pd.DataFrame:
    """
    Verifica que el DataFrame contenga las columnas definidas en PAIR_COLS
    y, opcionalmente, otras columnas requeridas (require).
    Si faltan, lanza KeyError. Devuelve el mismo DataFrame si está correcto.
    """
    need = set(PAIR_COLS)
    if require:
        need |= set(require)
    have = set(df.columns)
    missing = list(need - have)
    if missing:
        raise KeyError(f"[{name}] faltan columnas {missing}. Columnas presentes: {sorted(have)}")
    return df

# --- Cargamos entradas canónicas ---
train_df = _ensure_ms(_read_csv(PATH_TRAIN, required=True), DATECOL)
test_df  = _ensure_ms(_read_csv(PATH_TEST,  required=True), DATECOL)
preds_df = _ensure_ms(_read_csv(PATH_PREDS, required=True), DATECOL)
final_df = _read_csv(PATH_FINAL, required=True)
paso8_df = _read_csv(PATH_PASO8, required=False)  # opcional
bases_df = _read_csv(PATH_BASES, required=False)  # opcional

# --- Validaciones tempranas de esquema ---
_ensure_pair_cols(train_df, "train_df", require=[DATECOL, VALUE_TRAINTEST])
_ensure_pair_cols(test_df,  "test_df",  require=[DATECOL, VALUE_TRAINTEST])
_ensure_pair_cols(preds_df, "preds_df", require=[DATECOL, "yhat_combo"])
_ensure_pair_cols(final_df, "final_df")  # aquí solo exigimos PAIR_COLS

# Dimensiones opcionales (PAIS, REGION, TIPO_USO)
dim_df = _read_csv(PATH_DIM, required=False)
if not dim_df.empty:
    keep = [c for c in ["ID_BUILDING","PAIS","REGION","TIPO_USO"] if c in dim_df.columns]
    dim_df = dim_df[keep].drop_duplicates()
else:
    dim_df = pd.DataFrame(columns=["ID_BUILDING","PAIS","REGION","TIPO_USO"])

# --- Normalizamos claves ---
train_df = _norm_pair_keys(train_df)
test_df  = _norm_pair_keys(test_df)
preds_df = _norm_pair_keys(preds_df)
final_df = _norm_pair_keys(final_df)
dim_df   = _norm_pair_keys(dim_df)

# Revalidamos tras normalización (por si algún cast dejara NaNs/strings raros)
_ensure_pair_cols(train_df, "train_df(norm)", require=[DATECOL, VALUE_TRAINTEST])
_ensure_pair_cols(test_df,  "test_df(norm)",  require=[DATECOL, VALUE_TRAINTEST])
_ensure_pair_cols(preds_df, "preds_df(norm)", require=[DATECOL, "yhat_combo"])
_ensure_pair_cols(final_df, "final_df(norm)")

# --- Construimos malla 2024 para verdad-terreno agregada y reindexada ---
test_2024 = test_df[[*PAIR_COLS, DATECOL, VALUE_TRAINTEST]].copy()
test_2024 = _agg_dup_and_reindex(
    test_2024, DATECOL, VALUE_TRAINTEST,
    months=MONTHS_2024, pair_cols=PAIR_COLS, agg="sum"
)

# --- Predicciones 2024 (aseguramos 12 filas por pareja) ---
preds_2024 = (preds_df
              .groupby(PAIR_COLS + [DATECOL], as_index=False)
              .agg({"yhat_combo":"sum"}))
preds_2024 = _agg_dup_and_reindex(
    preds_2024, DATECOL, "yhat_combo",
    months=MONTHS_2024, pair_cols=PAIR_COLS, agg="sum"
)

# --- Panel de evaluación (universo del modelo_final del Paso 6) ---
panel_pairs = final_df[PAIR_COLS].drop_duplicates().copy()
total_series_portfolio = panel_pairs.shape[0]
log9(f"Total de series en modelo_final (panel): {total_series_portfolio}")

# --- Unimos verdad y predicción a nivel serie–mes ---
series_month = (panel_pairs
                .merge(test_2024, on=PAIR_COLS, how="left")
                .merge(preds_2024, on=PAIR_COLS + [DATECOL], how="left"))

# --- Renombrados y campos base ---
series_month = series_month.rename(columns={VALUE_TRAINTEST: "y_true", "yhat_combo": "y_pred"})
series_month["abs_err"] = np.abs(series_month["y_pred"] - series_month["y_true"])
series_month["cobertura_mes"] = (~series_month["y_true"].isna()).astype(int)
series_month["peso_gasto_mes"] = np.abs(series_month["y_true"]).fillna(0.0)

# --- Adjuntamos dimensiones si existen ---
series_month = series_month.merge(dim_df, on="ID_BUILDING", how="left")
for c in ["PAIS","REGION","TIPO_USO"]:
    if c not in series_month.columns:
        series_month[c] = None

[2025-09-24T12:36:58] Total de series en modelo_final (panel): 2429

# ============================================================
# Bloque 5) Escalas MASE por serie y métricas por fila
# ============================================================

# Preparamos escalas MASE1/MASE12 por serie usando train 2021–2023
train_idx = (_ensure_ms(train_df, DATECOL)
             .sort_values(PAIR_COLS + [DATECOL])
             .set_index(PAIR_COLS))

scales = {}
for key, g in train_idx.groupby(level=[0,1]):
    y_hist = g[VALUE_TRAINTEST].astype(float).values
    s1  = _scale_diff(y_hist, 1)
    s12 = _scale_diff(y_hist, 12)
    scales[key] = {"scale1": s1, "scale12": s12}

def _lookup_scale(row):
    key = (row[PAIR_COLS[0]], row[PAIR_COLS[1]])
    sc = scales.get(key, {"scale1": np.nan, "scale12": np.nan})
    return pd.Series({"scale1": sc["scale1"], "scale12": sc["scale12"]})

series_month = series_month.merge(
    series_month.apply(_lookup_scale, axis=1),
    left_index=True, right_index=True
)

# Métricas por fila: MAE fila = abs_err; WAPE fila = abs_err / |y_true|; SMAPE puntual
def _smape_point(y, yhat, eps=EPS):
    y = float(y) if pd.notna(y) else np.nan
    yhat = float(yhat) if pd.notna(yhat) else np.nan
    if np.isnan(y) or np.isnan(yhat):
        return np.nan
    den = max((abs(y) + abs(yhat))/2.0, eps)
    return 100.0 * abs(yhat - y) / den

series_month["MAE"]    = series_month["abs_err"]
series_month["WAPE"]   = _safe_div(series_month["abs_err"], series_month["peso_gasto_mes"].replace(0.0, np.nan)).replace([np.inf, -np.inf], np.nan)
series_month["SMAPE"]  = series_month.apply(lambda r: _smape_point(r["y_true"], r["y_pred"], eps=EPS), axis=1)
series_month["MASE1"]  = series_month.apply(lambda r: mase_point_abs_err(r["abs_err"], r["scale1"]), axis=1)
series_month["MASE12"] = series_month.apply(lambda r: mase_point_abs_err(r["abs_err"], r["scale12"]), axis=1)

# Orden estable de columnas y guardado
cols_series = [*PAIR_COLS, DATECOL, "y_true", "y_pred", "abs_err", "MAE", "SMAPE", "WAPE", "MASE1", "MASE12",
               "cobertura_mes", "peso_gasto_mes", "PAIS", "REGION", "TIPO_USO"]
series_month = series_month[cols_series].sort_values(PAIR_COLS + [DATECOL]).reset_index(drop=True)
series_month.to_csv(OUT_SERIES, sep=CSV_SEP, index=False)
log9(f"Guardado: {OUT_SERIES} con {len(series_month)} filas.")

[2025-09-24T12:37:08] Guardado: /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3/METRICAS/metrics_series_mensual.csv con 26970 filas.

# ----- Diagnóstico de cobertura y duplicados -----
print("[check] filas esperadas si todas las series tienen 12 meses:", panel_pairs.shape[0] * 12)
print("[check] filas reales en series_month:", len(series_month))

# pares únicos y meses por par
cov = (series_month.groupby(PAIR_COLS, as_index=False)[DATECOL]
       .nunique().rename(columns={DATECOL: "n_meses"}))
print("[check] distribución n_meses por serie:")
print(cov["n_meses"].value_counts().sort_index())

# pares con n_meses != 12
bad = cov[cov["n_meses"] != 12]
print(f"[check] series con n_meses != 12: {len(bad)} (muestra 10)")
print(bad.head(10))

# ¿hay duplicados exactos de (pair, FECHA)?
dups = series_month.duplicated(subset=PAIR_COLS + [DATECOL]).sum()
print("[check] duplicados exactos (pair, FECHA):", dups)

# ¿cuántos pares del panel no aparecen en test_2024?
pairs_test = test_2024[PAIR_COLS].drop_duplicates()
missing_in_test = (panel_pairs.merge(pairs_test, on=PAIR_COLS, how="left", indicator=True)
                   .query("_merge=='left_only'")[PAIR_COLS])
print("[check] pares del panel sin filas en test_2024:", len(missing_in_test))
print(missing_in_test.head(10))

[check] filas esperadas si todas las series tienen 12 meses: 29148
[check] filas reales en series_month: 26970
[check] distribución n_meses por serie:
n_meses
0      198
12    2231
Name: count, dtype: int64
[check] series con n_meses != 12: 198 (muestra 10)
    ID_BUILDING           FM_COST_TYPE  n_meses
0             2  Eficiencia Energética        0
24           57  Eficiencia Energética        0
25           57              Licencias        0
27           57       Mtto. Correctivo        0
28           57        Servicios Ctto.        0
29           57        Servicios Extra        0
30           57            Suministros        0
31           59              Licencias        0
33           59       Mtto. Correctivo        0
34           59                  Obras        0
[check] duplicados exactos (pair, FECHA): 0
[check] pares del panel sin filas en test_2024: 198
    ID_BUILDING           FM_COST_TYPE
0             2  Eficiencia Energética
24           57  Eficiencia Energética
25           57              Licencias
27           57       Mtto. Correctivo
28           57        Servicios Ctto.
29           57        Servicios Extra
30           57            Suministros
31           59              Licencias
33           59       Mtto. Correctivo
34           59                  Obras

# ============================================================
# Bloque 6) Agregación por segmento–mes
# ============================================================

# Preparamos función de agregado por dimensión
def _agg_segment_month(df: pd.DataFrame, dim_col: str, dim_name: str, total_series_segmento: pd.DataFrame):
    # Calculamos métricas agregadas por mes, además de cobertura relativa y pesos
    g = df.groupby([dim_col, DATECOL], as_index=False)
    out = g.apply(lambda d: pd.Series({
        "MAE":    d["MAE"].mean(),
        "SMAPE":  d["SMAPE"].mean(),
        "WAPE":   wape_micro(d["y_true"].values, d["y_pred"].values, eps=EPS),
        "MASE1":  d["MASE1"].mean(),
        "MASE12": d["MASE12"].mean(),
        "n_series_mes": int(d.loc[d["cobertura_mes"]==1, [PAIR_COLS[0],PAIR_COLS[1]]].drop_duplicates().shape[0]),
        "peso_gasto_mes": float(np.nansum(np.abs(d["y_true"]))),
    })).reset_index(drop=True)

    # Cobertura relativa: n_series_mes / total_series_segmento
    out = out.merge(total_series_segmento, left_on=dim_col, right_on=dim_col, how="left")
    out["cobertura_relativa"] = out["n_series_mes"] / out["total_series_segmento"].replace(0, np.nan)

    # Añadimos identificadores de nivel y clave
    out["nivel_segmento"] = dim_name
    out = out.rename(columns={dim_col: "clave_segmento"})
    # Orden de columnas
    out = out[["nivel_segmento","clave_segmento", DATECOL,
               "MAE","SMAPE","WAPE","MASE1","MASE12",
               "n_series_mes","peso_gasto_mes","cobertura_relativa"]]
    return out

# Preparamos conteos de universo por dimensión (series posibles por segmento)
def _universe_by_dim(panel_pairs: pd.DataFrame, series_month: pd.DataFrame, dim_col: str):
    # Determinamos el universo de pares por segmento (tomamos únicos del panel final)
    if dim_col == "FM_COST_TYPE":
        base = panel_pairs.copy()
        base = base.rename(columns={PAIR_COLS[1]: "FM_COST_TYPE"})
        u = base.groupby("FM_COST_TYPE")[PAIR_COLS].apply(lambda x: x.drop_duplicates().shape[0]).rename("total_series_segmento").reset_index()
        return u.rename(columns={"FM_COST_TYPE": dim_col})
    else:
        if dim_col not in series_month.columns:
            return pd.DataFrame(columns=[dim_col,"total_series_segmento"])
        base = (series_month[[*PAIR_COLS, dim_col]]
                .dropna(subset=[dim_col])
                .drop_duplicates())
        u = base.groupby(dim_col)[PAIR_COLS].apply(lambda x: x.drop_duplicates().shape[0]).rename("total_series_segmento").reset_index()
        return u

# Preparamos las dimensiones disponibles
dim_specs = [("FM_COST_TYPE","FM_COST_TYPE"), ("PAIS","PAIS"), ("REGION","REGION"), ("TIPO_USO","TIPO_USO")]
segment_frames = []

for dim_col, dim_name in dim_specs:
    if (dim_col == "FM_COST_TYPE") or (dim_col in series_month.columns):
        uni = _universe_by_dim(panel_pairs, series_month, dim_col)
        seg_df = _agg_segment_month(series_month, dim_col, dim_name, uni)
        segment_frames.append(seg_df)
    else:
        log9(f"AVISO: omitimos dimensión {dim_col} por no disponer de metadatos.")

metrics_segmento = pd.concat(segment_frames, ignore_index=True) if len(segment_frames)>0 else pd.DataFrame(
    columns=["nivel_segmento","clave_segmento",DATECOL,"MAE","SMAPE","WAPE","MASE1","MASE12","n_series_mes","peso_gasto_mes","cobertura_relativa"]
)

metrics_segmento = metrics_segmento.sort_values(["nivel_segmento","clave_segmento",DATECOL]).reset_index(drop=True)
metrics_segmento.to_csv(OUT_SEGMENTOS, sep=CSV_SEP, index=False)
log9(f"Guardado: {OUT_SEGMENTOS} con {len(metrics_segmento)} filas.")

[2025-09-24T12:37:11] Guardado: /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3/METRICAS/metrics_segmento_mensual.csv con 1560 filas.

# ============================================================
# Bloque 7) Agregación de portfolio mensual y anual
# ============================================================

# Mensual
g = series_month.groupby(DATECOL, as_index=False)
portfolio_month = g.apply(lambda d: pd.Series({
    "MAE":    d["MAE"].mean(),
    "SMAPE":  d["SMAPE"].mean(),
    "WAPE":   wape_micro(d["y_true"].values, d["y_pred"].values, eps=EPS),
    "MASE1":  d["MASE1"].mean(),
    "MASE12": d["MASE12"].mean(),
    "n_series_mes": int(d.loc[d["cobertura_mes"]==1, [PAIR_COLS[0],PAIR_COLS[1]]].drop_duplicates().shape[0]),
    "peso_gasto_mes": float(np.nansum(np.abs(d["y_true"]))),
    "cobertura_relativa_mes": float(d.loc[d["cobertura_mes"]==1, [PAIR_COLS[0],PAIR_COLS[1]]].drop_duplicates().shape[0]) / max(total_series_portfolio, 1),
})).reset_index(drop=True)

# Anual (fila agregada 2024)
mask_valid = series_month["cobertura_mes"]==1
d_all = series_month.loc[mask_valid].copy()
annual_row = pd.DataFrame([{
    "FECHA": pd.Timestamp("2024-12-01"),
    "MAE_2024":    d_all["MAE"].mean(),
    "SMAPE_2024":  d_all["SMAPE"].mean(),
    "WAPE_2024":   wape_micro(d_all["y_true"].values, d_all["y_pred"].values, eps=EPS),
    "MASE1_2024":  d_all["MASE1"].mean(),
    "MASE12_2024": d_all["MASE12"].mean(),
    "n_series_activas_2024": int(d_all[[*PAIR_COLS]].drop_duplicates().shape[0]),
    "peso_gasto_2024": float(np.nansum(np.abs(d_all["y_true"]))),
    "cobertura_relativa_media": float(portfolio_month["cobertura_relativa_mes"].mean()) if len(portfolio_month)>0 else np.nan
}])

# Unimos y guardamos
portfolio_out = portfolio_month.copy()
for col in annual_row.columns:
    if col not in portfolio_out.columns:
        portfolio_out[col] = np.nan
portfolio_out = pd.concat([portfolio_out, annual_row], ignore_index=True)
portfolio_out.to_csv(OUT_PORTFOLIO, sep=CSV_SEP, index=False)
log9(f"Guardado: {OUT_PORTFOLIO} con {len(portfolio_out)} filas.")

[2025-09-24T12:37:12] Guardado: /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3/METRICAS/metrics_portfolio_mensual.csv con 13 filas.

# ============================================================
# Bloque 8) Listados de apoyo a la acción
# ============================================================

# Top-10 series por contribución al error absoluto anual
contrib = (series_month
           .groupby(PAIR_COLS, as_index=False)
           .agg(contrib_abs_error_2024=("abs_err","sum"),
                peso_gasto_2024=("peso_gasto_mes","sum")))

# Intentamos añadir 'route' si venía en preds (opcional)
if "route" in preds_df.columns:
    route_pairs = preds_df[PAIR_COLS + ["route"]].drop_duplicates()
    contrib = contrib.merge(route_pairs, on=PAIR_COLS, how="left")

top10_series = contrib.sort_values("contrib_abs_error_2024", ascending=False).head(10)
top10_series.to_csv(OUT_TOP10, sep=CSV_SEP, index=False)
log9(f"Guardado: {OUT_TOP10}")

# Top-5 segmentos “rojos” por WAPE 2024 (al menos por FM_COST_TYPE)
top5_frames = []
for dim_col, dim_name in [("FM_COST_TYPE","FM_COST_TYPE"), ("PAIS","PAIS"), ("REGION","REGION"), ("TIPO_USO","TIPO_USO")]:
    if (dim_col == "FM_COST_TYPE") or (dim_col in series_month.columns):
        df_dim = series_month.copy()
        if dim_col != "FM_COST_TYPE":
            df_dim = df_dim.dropna(subset=[dim_col])
        grp = df_dim.groupby(dim_col, as_index=False).apply(
            lambda d: pd.Series({
                "WAPE_2024": wape_micro(d["y_true"].values, d["y_pred"].values, eps=EPS),
                "peso_gasto_2024": float(np.nansum(np.abs(d["y_true"]))),
                "n_series": int(d[[*PAIR_COLS]].drop_duplicates().shape[0])
            })
        ).reset_index(drop=True)
        if grp.empty:
            continue
        grp = grp.rename(columns={dim_col: "clave_segmento"})
        grp["nivel_segmento"] = dim_name
        grp = grp.sort_values("WAPE_2024", ascending=False).head(5)
        grp["ranking"] = np.arange(1, len(grp)+1)
        top5_frames.append(grp)

top5_out = pd.concat(top5_frames, ignore_index=True) if len(top5_frames)>0 else pd.DataFrame(
    columns=["nivel_segmento","clave_segmento","WAPE_2024","peso_gasto_2024","n_series","ranking"]
)
top5_out.to_csv(OUT_TOP5, sep=CSV_SEP, index=False)
log9(f"Guardado: {OUT_TOP5}")

[2025-09-24T12:37:12] Guardado: /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3/REPORTING/top10_series_contrib_error.csv
[2025-09-24T12:37:12] Guardado: /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3/REPORTING/top5_segmentos_rojos_WAPE.csv

# ============================================================
# Bloque 9) Subset estable y agregados específicos
# ============================================================

# Identificamos series con 12/12 en 2024
cov_2024 = (series_month[series_month["cobertura_mes"]==1]
            .groupby(PAIR_COLS, as_index=False)[DATECOL].nunique()
            .rename(columns={DATECOL:"n_meses_2024"}))
stable_12m = cov_2024[cov_2024["n_meses_2024"]==12][PAIR_COLS]

# Comprobamos histórico con >=36 meses y sin huecos (2021-01..2023-12 => 36 MS)
train_ms = _ensure_ms(train_df, DATECOL)
train_panel = (train_ms[(train_ms[DATECOL] >= "2021-01-01") & (train_ms[DATECOL] <= "2023-12-01")]
               .groupby(PAIR_COLS, as_index=False)[DATECOL].nunique()
               .rename(columns={DATECOL:"n_train_ms"}))
stable_hist = train_panel[train_panel["n_train_ms"]==36][PAIR_COLS]

# Subset estable = intersección
stable_pairs = stable_12m.merge(stable_hist, on=PAIR_COLS, how="inner")
log9(f"Subset estable: {stable_pairs.shape[0]} series.")

subset_df = series_month.merge(stable_pairs, on=PAIR_COLS, how="inner")

# Métricas de portfolio del subset
g = subset_df.groupby(DATECOL, as_index=False)
subset_portfolio_month = g.apply(lambda d: pd.Series({
    "MAE": d["MAE"].mean(),
    "SMAPE": d["SMAPE"].mean(),
    "WAPE": wape_micro(d["y_true"].values, d["y_pred"].values, eps=EPS),
    "MASE1": d["MASE1"].mean(),
    "MASE12": d["MASE12"].mean(),
    "n_series_mes": int(d.loc[d["cobertura_mes"]==1, [*PAIR_COLS]].drop_duplicates().shape[0]),
    "peso_gasto_mes": float(np.nansum(np.abs(d["y_true"]))),
})).reset_index(drop=True)
subset_portfolio_month["scope"] = "PORTFOLIO"

# Agregados por segmento del subset (reutilizamos función del bloque 6)
subset_seg_frames = []
for dim_col, dim_name in [("FM_COST_TYPE","FM_COST_TYPE"), ("PAIS","PAIS"), ("REGION","REGION"), ("TIPO_USO","TIPO_USO")]:
    if (dim_col == "FM_COST_TYPE") or (dim_col in subset_df.columns):
        uni = _universe_by_dim(stable_pairs, subset_df, dim_col)
        seg_df = _agg_segment_month(subset_df, dim_col, dim_name, uni)
        seg_df["scope"] = "SEGMENTO"
        subset_seg_frames.append(seg_df)

subset_seg_month = (pd.concat(subset_seg_frames, ignore_index=True)
                    if len(subset_seg_frames)>0 else
                    pd.DataFrame(columns=["nivel_segmento","clave_segmento",DATECOL,"MAE","SMAPE","WAPE","MASE1","MASE12",
                                          "n_series_mes","peso_gasto_mes","cobertura_relativa","scope"]))

# Guardamos en un único CSV con una columna 'scope' para distinguir vistas
# Para minimizar fricción, apilamos: primero portfolio, luego segmentos
subset_portfolio_month.to_csv(OUT_SUBSET, sep=CSV_SEP, index=False)
with open(OUT_SUBSET, "a", encoding="utf-8") as f:
    subset_seg_month.to_csv(f, sep=CSV_SEP, index=False, header=False)
log9(f"Guardado: {OUT_SUBSET}")

[2025-09-24T12:37:12] Subset estable: 2231 series.
[2025-09-24T12:37:16] Guardado: /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3/METRICAS/metrics_subset_estable.csv

# ============================================================
# Bloque 10) Control F: WAPE anual del portfolio vs baseline
# ============================================================

def _controlF_with_external_baseline(portfolio_out_csv: str, external_baseline_path: str | None):
    # Intentamos leer un baseline externo si se proporciona
    if external_baseline_path and os.path.exists(external_baseline_path):
        bdf = pd.read_csv(external_baseline_path, sep=CSV_SEP)
        # Intentamos localizar una columna razonable para el WAPE baseline anual
        cand_cols = [c for c in bdf.columns if "WAPE" in c.upper() and "BASELINE" in c.upper()]
        if len(cand_cols)==0 and "WAPE_portfolio_baseline" in bdf.columns:
            cand_cols = ["WAPE_portfolio_baseline"]
        if len(cand_cols)>0:
            # Tomamos el primer candidato
            baseline_val = float(pd.to_numeric(bdf[cand_cols[0]], errors="coerce").dropna().iloc[-1])
            return baseline_val
        log9("AVISO: no identificamos columna de baseline en el CSV externo; pasamos a SNaive12.")
    else:
        if external_baseline_path:
            log9(f"AVISO: baseline externo no encontrado en {external_baseline_path}; usamos SNaive12.")
    return None

def _controlF_compute_snaive12_baseline(train_df: pd.DataFrame,
                                         test_2024: pd.DataFrame,
                                         panel_pairs: pd.DataFrame) -> float:
    """
    Recalcula baseline SNaive12 en el mismo universo que se evalúa:
    - Solo series del panel (modelo_final).
    - Solo series con cobertura en 2024 (al menos un y_true no nulo).
    Devuelve el WAPE micro del portfolio para esas series.
    """
    # 1) Universo: pares del panel con cobertura en 2024
    base_eval = (panel_pairs
                 .merge(test_2024.rename(columns={VALUE_TRAINTEST: "y_true"}),
                        on=PAIR_COLS, how="left"))
    base_eval = base_eval[base_eval["y_true"].notna()]  # cobertura
    pares_cubiertos = base_eval[PAIR_COLS].drop_duplicates()
    if pares_cubiertos.empty:
        log9("[ControlF] AVISO: no hay pares con cobertura en 2024 para baseline SNaive12.")
        return np.nan

    # 2) Recalcular SNaive12 por serie (12 meses de 2024)
    train_idx = (_ensure_ms(train_df, DATECOL)
                 .sort_values(PAIR_COLS + [DATECOL])
                 .set_index(PAIR_COLS))

    preds_list = []
    for key, g in train_idx.groupby(level=[0, 1]):
        # Solo series que realmente vamos a evaluar
        if not ((pares_cubiertos[PAIR_COLS[0]] == key[0]) &
                (pares_cubiertos[PAIR_COLS[1]] == key[1])).any():
            continue
        y_hist = g[VALUE_TRAINTEST].astype(float).values
        d_hist = g[DATECOL].values
        yhat = snaive12_forecast(y_hist, d_hist, horizon=12)
        tmp = pd.DataFrame({
            PAIR_COLS[0]: [key[0]] * len(MONTHS_2024),
            PAIR_COLS[1]: [key[1]] * len(MONTHS_2024),
            DATECOL: MONTHS_2024,
            "yhat_snaive12": yhat
        })
        preds_list.append(tmp)

    if len(preds_list) == 0:
        log9("[ControlF] AVISO: no fue posible generar predicciones SNaive12 para ningún par del panel con cobertura.")
        return np.nan

    snaive_preds = pd.concat(preds_list, ignore_index=True)

    # 3) Ensamblar base de comparación y restringir a las filas con y_true
    base = (test_2024.rename(columns={VALUE_TRAINTEST: "y_true"})
            .merge(snaive_preds, on=PAIR_COLS + [DATECOL], how="inner"))
    base = base[base["y_true"].notna()]

    if base.empty:
        log9("[ControlF] AVISO: tras el merge, no quedan observaciones con y_true para evaluar baseline.")
        return np.nan

    # 4) WAPE micro del portfolio (baseline)
    return wape_micro(base["y_true"].values, base["yhat_snaive12"].values, eps=EPS)

# Leemos nuestro WAPE anual del portfolio ya calculado
WAPE_modelo_final = float(portfolio_out.loc[portfolio_out["WAPE"].notna(), "WAPE"].iloc[:-1].mean()) if "WAPE" in portfolio_out.columns else np.nan
# Para el valor anual, preferimos el campo "WAPE_2024" si está
if "WAPE_2024" in portfolio_out.columns and pd.notna(portfolio_out["WAPE_2024"]).any():
    WAPE_modelo_final = float(portfolio_out["WAPE_2024"].dropna().iloc[-1])

# Intentamos baseline externo; si no, recalculamos SNaive12 con cobertura adecuada
baseline_external = _controlF_with_external_baseline(OUT_PORTFOLIO, external_baseline_path=None)
if baseline_external is None:
    WAPE_baseline = _controlF_compute_snaive12_baseline(train_df, test_2024, panel_pairs)
    baseline_label = "SNaive12 (solo pares con cobertura 2024)"
else:
    WAPE_baseline = baseline_external
    baseline_label = "baseline externo"


ok_controlF = (WAPE_modelo_final <= WAPE_baseline) if (pd.notna(WAPE_modelo_final) and pd.notna(WAPE_baseline)) else False

log9(f"Control F — WAPE_portfolio_modelo_final={WAPE_modelo_final:.6f}")
log9(f"Control F — WAPE_portfolio_baseline ({baseline_label})={WAPE_baseline:.6f}")
log9(f"Control F — ok_controlF={ok_controlF}")

if not ok_controlF:
    log9("ADVERTENCIA Control F: el modelo final no mejora al baseline. Checklist rápido:")
    log9(" - ¿TSB activo en intermitentes?")
    log9(" - ¿Combo top-2 aplicado (weighted_by_inv_MASE12 o mean50_50)?")
    log9(" - ¿Fallback SNaive12 aplicado donde tocaba (solo test / fallos)?")

[2025-09-24T12:37:29] Control F — WAPE_portfolio_modelo_final=0.358801
[2025-09-24T12:37:29] Control F — WAPE_portfolio_baseline (SNaive12 (solo pares con cobertura 2024))=0.392322
[2025-09-24T12:37:29] Control F — ok_controlF=True

# ============================================================
# Bloque 11) Resumen final
# ============================================================

try:
    wape_anual = float(portfolio_out["WAPE_2024"].dropna().iloc[-1])
    mae_anual  = float(portfolio_out["MAE_2024"].dropna().iloc[-1])
    smape_anual= float(portfolio_out["SMAPE_2024"].dropna().iloc[-1])
    log9(f"Resumen — Portfolio 2024: WAPE={wape_anual:.6f}, MAE={mae_anual:.6f}, SMAPE={smape_anual:.2f}%")
except Exception:
    log9("Resumen — No fue posible imprimir métricas anuales del portfolio (faltan columnas).")

log9("Paso 9 finalizado.")

[2025-09-24T12:37:29] Resumen — Portfolio 2024: WAPE=0.358801, MAE=539.231234, SMAPE=78.77%
[2025-09-24T12:37:29] Paso 9 finalizado.

# ============================================================
# ESTRATEGIA 3 — PASO 10
# Reconciliación jerárquica ligera (MinT shrink) sobre 2024
# ============================================================
# En este paso:
#  - Leemos predicciones base 2024 por serie (bottom)
#  - Construimos un árbol jerárquico mínimo:
#       Bottom: (ID_BUILDING, FM_COST_TYPE)
#       Nivel 1: (FM_COST_TYPE, PAIS) si PAIS existe
#       Nivel 2 (opcional): (FM_COST_TYPE, REGION) si REGION existe
#       Top (opcional): (FM_COST_TYPE)
#  - Reconciliamos mes a mes mediante MinT (shrink simple) en forma WLS con restricciones:
#       y_b_rec = y_b + W_b A' (A W_b A')^{-1} (c - A y_b)
#       A es (n_agg x n_bottom); W_b diagonal (varianzas bottom)
#       c se toma de la verdad 2024 agregada cuando existe; si no, c = A y_b
#  - Clippamos a >= 0 solo en bottom
#  - Guardamos predicciones reconciliadas y comparamos métricas contra base
#  - Punto de control G: si la mejora en WAPE anual (portfolio) < 1 pp, recomendamos mantener base
# ============================================================


# -----------------------------
# Bloque 0 — Parámetros y rutas
# -----------------------------

np.random.seed(7)

ruta_base_3 = "/content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3"
CSV_SEP     = ";"

RUTA_RESULTADOS = os.path.join(ruta_base_3, "RESULTADOS")
RUTA_METRICAS   = os.path.join(ruta_base_3, "METRICAS")
RUTA_REPORTING  = os.path.join(ruta_base_3, "REPORTING")
RUTA_LOGS       = os.path.join(ruta_base_3, "LOGS")

os.makedirs(RUTA_RESULTADOS, exist_ok=True)
os.makedirs(RUTA_METRICAS,   exist_ok=True)
os.makedirs(RUTA_REPORTING,  exist_ok=True)
os.makedirs(RUTA_LOGS,       exist_ok=True)

# Entradas
PATH_PREDS  = os.path.join(RUTA_RESULTADOS, "preds_por_serie_2024.csv")   # del Paso 5
PATH_TEST   = os.path.join(RUTA_RESULTADOS, "test_full_2024.csv")         # verdad terreno 2024
PATH_DIM    = os.path.join(RUTA_METRICAS,   "dim_buildings.csv")          # del Paso 9
PATH_PASO8  = os.path.join(RUTA_METRICAS,   "paso8_metrics_2024_por_pareja.csv")  # opcional
PATH_TRAIN  = os.path.join(RUTA_RESULTADOS, "train_full_2021_2023.csv")   # opcional (escalas MASE)

# Salidas
OUT_PREDS_RECONC = os.path.join(RUTA_RESULTADOS, "preds_reconciliadas_2024.csv")
OUT_COMPARE      = os.path.join(RUTA_REPORTING,  "compare_metrics_reconc_vs_base.csv")
FIGS_DIR         = os.path.join(RUTA_REPORTING,  "figs_step10")
os.makedirs(FIGS_DIR, exist_ok=True)

# Fechas / claves
PAIR_COLS   = ["ID_BUILDING", "FM_COST_TYPE"]
DATECOL     = "FECHA"
VALCOL_TRUE = "cost_float_mod"
VALCOL_PRED = "yhat_combo"
H           = 12
MONTHS_2024 = pd.date_range("2024-01-01", periods=H, freq="MS")

# Logging
LOG_PATH = os.path.join(RUTA_LOGS, "estrategia3_step10.log")
def log10(msg: str):
    ts = datetime.now().isoformat(timespec="seconds")
    line = f"[{ts}] {msg}"
    print(line)
    with open(LOG_PATH, "a", encoding="utf-8") as f:
        f.write(line + "\n")

log10("Iniciamos Paso 10 — Reconciliación jerárquica MinT (shrink).")

[2025-09-24T12:37:30] Iniciamos Paso 10 — Reconciliación jerárquica MinT (shrink).

# -------------------------------------
# Bloque 1 — Utilidades generales/IO
# -------------------------------------

EPS = 1e-8

def _ensure_ms(df: pd.DataFrame, datecol: str) -> pd.DataFrame:
    df = df.copy()
    df[datecol] = pd.to_datetime(df[datecol]).dt.to_period("M").dt.to_timestamp()
    return df

def _read_csv(path: str, required: bool = True) -> pd.DataFrame:
    if os.path.exists(path):
        df = pd.read_csv(path, sep=CSV_SEP)
        df.columns = [str(c).strip().lstrip("\ufeff") for c in df.columns]
        return df
    if required:
        raise FileNotFoundError(f"Falta archivo requerido: {path}")
    log10(f"AVISO: no encontramos {os.path.basename(path)} (continuamos sin este artefacto).")
    return pd.DataFrame()

def _norm_pair_keys(df: pd.DataFrame) -> pd.DataFrame:
    if df is None or df.empty:
        return df
    out = df.copy()
    out.columns = [str(c).strip() for c in out.columns]
    if "ID_BUILDING" in out.columns:
        try:
            out["ID_BUILDING"] = pd.to_numeric(out["ID_BUILDING"], errors="coerce").astype("Int64")
        except Exception:
            pass
    if "FM_COST_TYPE" in out.columns:
        out["FM_COST_TYPE"] = out["FM_COST_TYPE"].astype(str).str.strip()
    return out

def _safe_div(num, den, eps=EPS):
    return num / (den + eps)

def wape_micro(y_true, y_pred, eps=EPS):
    y_true = np.asarray(y_true, dtype=float)
    y_pred = np.asarray(y_pred, dtype=float)
    return float(_safe_div(np.sum(np.abs(y_pred - y_true)), np.sum(np.abs(y_true)), eps))

def smape_vec(y_true, y_pred, eps=EPS):
    y_true = np.asarray(y_true, dtype=float)
    y_pred = np.asarray(y_pred, dtype=float)
    num = np.abs(y_pred - y_true)
    den = (np.abs(y_true) + np.abs(y_pred)) / 2.0
    return float(np.mean(_safe_div(num, den, eps)) * 100.0)

def _scale_diff(series: np.ndarray, lag: int):
    arr = np.asarray(series, dtype=float)
    if len(arr) <= lag:
        return np.nan
    diffs = np.abs(arr[lag:] - arr[:-lag])
    if diffs.size == 0:
        return np.nan
    return float(np.mean(diffs))

def mase_from_scale(errors: np.ndarray, scale: float):
    if scale is None or np.isnan(scale) or scale == 0:
        return np.nan
    errors = np.asarray(errors, dtype=float)
    return float(np.mean(np.abs(errors)) / float(scale))

# ------------------------------------------------
# Bloque 2 — Carga y preparación de las entradas
# ------------------------------------------------

preds_df = _ensure_ms(_read_csv(PATH_PREDS, required=True), DATECOL)
preds_df = _norm_pair_keys(preds_df)
if VALCOL_PRED not in preds_df.columns:
    raise ValueError(f"Esperábamos columna {VALCOL_PRED} en {os.path.basename(PATH_PREDS)}.")

# Nos quedamos con las columnas mínimas
keep_pred_cols = [c for c in [*PAIR_COLS, DATECOL, VALCOL_PRED, "route", "fallback_flag"] if c in preds_df.columns]
preds_df = preds_df[keep_pred_cols].copy()

# Aseguramos que tenemos una fila por (pair, mes); si hay duplicados, agregamos sum
preds_df = (preds_df.groupby(PAIR_COLS + [DATECOL], as_index=False)
                    .agg({VALCOL_PRED: "sum", **({"route":"last"} if "route" in preds_df.columns else {}),
                                         **({"fallback_flag":"max"} if "fallback_flag" in preds_df.columns else {})}))

# Verdad terreno 2024 (para métricas y, si queremos, para c)
test_df = _ensure_ms(_read_csv(PATH_TEST, required=True), DATECOL)
test_df  = _norm_pair_keys(test_df)
test_2024 = test_df[[*PAIR_COLS, DATECOL, VALCOL_TRUE]].copy()
# Reindexamos a malla 2024 (dejamos NaN donde no haya cobertura)
g = (test_2024.groupby(PAIR_COLS + [pd.Grouper(key=DATECOL, freq="MS")], as_index=False)[VALCOL_TRUE].sum())
panel_pairs = preds_df[PAIR_COLS].drop_duplicates()
reindexed = []
for a,b in panel_pairs.values:
    sub = g[(g[PAIR_COLS[0]]==a) & (g[PAIR_COLS[1]]==b)].set_index(DATECOL)
    sub = sub.reindex(MONTHS_2024).rename_axis(DATECOL).reset_index()
    sub[PAIR_COLS[0]] = a; sub[PAIR_COLS[1]] = b
    reindexed.append(sub)
test_2024 = pd.concat(reindexed, ignore_index=True)

# Dimensiones (opcional PAIS/REGION)
dim_df = _read_csv(PATH_DIM, required=False)
if not dim_df.empty:
    dim_cols_map = {}
    for c in dim_df.columns:
        cu = c.strip().upper()
        if cu == "PAIS":        dim_cols_map[c] = "PAIS"
        elif cu == "REGION":    dim_cols_map[c] = "REGION"
        elif cu == "ID_BUILDING": dim_cols_map[c] = "ID_BUILDING"
    dim_df = dim_df.rename(columns=dim_cols_map)
    dim_df = _norm_pair_keys(dim_df)
    dim_df = dim_df[[c for c in ["ID_BUILDING","PAIS","REGION"] if c in dim_df.columns]].drop_duplicates()
else:
    dim_df = pd.DataFrame(columns=["ID_BUILDING","PAIS","REGION"])

# Paso 8 (opcional) para proxy de W si no hay verdad
paso8_df = _read_csv(PATH_PASO8, required=False)

# Train (opcional) para escalas MASE
train_df = _read_csv(PATH_TRAIN, required=False)
if not train_df.empty:
    train_df = _ensure_ms(_norm_pair_keys(train_df), DATECOL)

log10(f"Cargadas entradas: preds={len(preds_df)} filas, test_2024={len(test_2024)} filas, dim={len(dim_df)} filas.")
log10(f"scipy.sparse disponible: {SCIPY_AVAILABLE}")

[2025-09-24T12:37:44] Cargadas entradas: preds=29148 filas, test_2024=29148 filas, dim=1650 filas.
[2025-09-24T12:37:44] scipy.sparse disponible: True

# ------------------------------------------------------------
# Bloque 3 — Preparación del panel y metadatos jerárquicos
# ------------------------------------------------------------

# Unimos PAIS/REGION a nivel bottom
bottom_df = preds_df.merge(dim_df, on="ID_BUILDING", how="left")
n_bottom = bottom_df[PAIR_COLS].drop_duplicates().shape[0]

# Diagnóstico de metadatos
pct_no_pais   = 1.0 - float(bottom_df["PAIS"].notna().mean()) if "PAIS" in bottom_df.columns else 1.0
pct_no_region = 1.0 - float(bottom_df["REGION"].notna().mean()) if "REGION" in bottom_df.columns else 1.0
log10(f"Metadatos: n_bottom={n_bottom} | % sin PAIS={100*pct_no_pais:.1f}% | % sin REGION={100*pct_no_region:.1f}%")

# Construimos claves agregadas presentes en el bottom real
def _unique_keys(df, cols):
    cols = [c for c in cols if c in df.columns]
    if len(cols)==0:
        return pd.DataFrame(columns=cols)
    u = df[cols].dropna().drop_duplicates()
    return u

keys_top        = _unique_keys(bottom_df, ["FM_COST_TYPE"])
keys_fm_pais    = _unique_keys(bottom_df, ["FM_COST_TYPE","PAIS"])
keys_fm_region  = _unique_keys(bottom_df, ["FM_COST_TYPE","REGION"])

# Avisamos si alguna capa queda vacía
if keys_top.empty:
    log10("AVISO: capa TOP (FM_COST_TYPE) vacía; continuamos sin top.")
if "PAIS" not in bottom_df.columns or keys_fm_pais.empty:
    log10("AVISO: capa (FM_COST_TYPE, PAIS) vacía o sin metadatos; continuamos sin esa capa.")
if "REGION" not in bottom_df.columns or keys_fm_region.empty:
    log10("AVISO: capa (FM_COST_TYPE, REGION) vacía o sin metadatos; continuamos sin esa capa.")

# Orden estable de claves
def _sorted_df(df, cols):
    return df.sort_values(cols).reset_index(drop=True)

keys_top       = _sorted_df(keys_top, ["FM_COST_TYPE"]) if not keys_top.empty else keys_top
keys_fm_pais   = _sorted_df(keys_fm_pais, ["FM_COST_TYPE","PAIS"]) if not keys_fm_pais.empty else keys_fm_pais
keys_fm_region = _sorted_df(keys_fm_region, ["FM_COST_TYPE","REGION"]) if not keys_fm_region.empty else keys_fm_region
keys_bottom    = _sorted_df(bottom_df[PAIR_COLS].drop_duplicates(), PAIR_COLS)

# Índice bottom y helpers de pertenencia
bottom_index = {tuple(r): i for i, r in keys_bottom.itertuples(index=False, name=None)}

def _child_idx_for_top(fm):
    mask = (keys_bottom["FM_COST_TYPE"]==fm)
    return keys_bottom.index[mask].tolist()

def _child_idx_for_fm_pais(fm, pais):
    mask = (keys_bottom["FM_COST_TYPE"]==fm)
    if "PAIS" not in bottom_df.columns:
        return []
    sub = bottom_df.merge(keys_bottom[mask][PAIR_COLS], on=PAIR_COLS, how="inner")
    sub = sub[sub["PAIS"]==pais]
    idx = keys_bottom.merge(sub[PAIR_COLS].drop_duplicates(), on=PAIR_COLS, how="inner").index.tolist()
    return idx

def _child_idx_for_fm_region(fm, region):
    mask = (keys_bottom["FM_COST_TYPE"]==fm)
    if "REGION" not in bottom_df.columns:
        return []
    sub = bottom_df.merge(keys_bottom[mask][PAIR_COLS], on=PAIR_COLS, how="inner")
    sub = sub[sub["REGION"]==region]
    idx = keys_bottom.merge(sub[PAIR_COLS].drop_duplicates(), on=PAIR_COLS, how="inner").index.tolist()
    return idx

# Reportamos tamaños por nivel
n_top       = 0 if keys_top.empty else len(keys_top)
n_fm_pais   = 0 if keys_fm_pais.empty else len(keys_fm_pais)
n_fm_region = 0 if keys_fm_region.empty else len(keys_fm_region)
total_nodes = n_top + n_fm_pais + n_fm_region + n_bottom
log10(f"Árbol: top={n_top} | fm×pais={n_fm_pais} | fm×region={n_fm_region} | bottom={n_bottom} | nodos totales={total_nodes}")

[2025-09-24T12:37:45] Metadatos: n_bottom=2429 | % sin PAIS=0.0% | % sin REGION=0.0%
[2025-09-24T12:37:45] Árbol: top=8 | fm×pais=49 | fm×region=573 | bottom=2429 | nodos totales=3059

# ---------------------------------------------
# Bloque 4 — Construcción de S (diag) y de A (solo agregados, sin filas bottom)
# ---------------------------------------------

def _build_S_matrix(n_bottom, keys_top, keys_fm_pais, keys_fm_region, keys_bottom):
    rows, cols, data = [], [], []
    row_id = 0

    if not keys_top.empty:
        for (fm,) in keys_top.itertuples(index=False, name=None):
            idxs = _child_idx_for_top(fm)
            for j in idxs:
                rows.append(row_id); cols.append(j); data.append(1.0)
            row_id += 1

    if not keys_fm_pais.empty:
        for (fm, pais) in keys_fm_pais.itertuples(index=False, name=None):
            idxs = _child_idx_for_fm_pais(fm, pais)
            for j in idxs:
                rows.append(row_id); cols.append(j); data.append(1.0)
            row_id += 1

    if not keys_fm_region.empty:
        for (fm, region) in keys_fm_region.itertuples(index=False, name=None):
            idxs = _child_idx_for_fm_region(fm, region)
            for j in idxs:
                rows.append(row_id); cols.append(j); data.append(1.0)
            row_id += 1

    start_bottom_row = row_id
    for j in range(n_bottom):
        rows.append(row_id); cols.append(j); data.append(1.0)
        row_id += 1

    m = row_id
    if SCIPY_AVAILABLE:
        S = sp.coo_matrix((data, (rows, cols)), shape=(m, n_bottom)).tocsr()
    else:
        S = np.zeros((m, n_bottom), dtype=float)
        for r, c, v in zip(rows, cols, data):
            S[r, c] = v
    return S, start_bottom_row

# S lo dejamos como diagnóstico
S, start_bottom_row = _build_S_matrix(n_bottom, keys_top, keys_fm_pais, keys_fm_region, keys_bottom)
if SCIPY_AVAILABLE:
    log10(f"Matriz S: shape={S.shape}, nnz={S.nnz}")
else:
    log10(f"Matriz S: shape={S.shape} (numpy densa)")

# A (solo agregados, sin filas bottom)
def _build_A_matrix(n_bottom, keys_top, keys_fm_pais, keys_fm_region):
    agg_rows = []

    def _add_row(child_idx_list):
        row = np.zeros(n_bottom, dtype=float)
        if len(child_idx_list) > 0:
            row[np.asarray(child_idx_list, dtype=int)] = 1.0
        agg_rows.append(row)

    # TOP
    if not keys_top.empty:
        for (fm,) in keys_top.itertuples(index=False, name=None):
            idxs = _child_idx_for_top(fm)
            _add_row(idxs)

    # FM×PAIS
    if not keys_fm_pais.empty:
        for (fm, pais) in keys_fm_pais.itertuples(index=False, name=None):
            idxs = _child_idx_for_fm_pais(fm, pais)
            _add_row(idxs)

    # FM×REGION
    if not keys_fm_region.empty:
        for (fm, region) in keys_fm_region.itertuples(index=False, name=None):
            idxs = _child_idx_for_fm_region(fm, region)
            _add_row(idxs)

    if len(agg_rows) == 0:
        A = np.zeros((0, n_bottom), dtype=float)
    else:
        A = np.vstack(agg_rows)
    return A

A = _build_A_matrix(n_bottom, keys_top, keys_fm_pais, keys_fm_region)
if SCIPY_AVAILABLE and A.size > 0:
    A_sparse = sp.csr_matrix(A)
else:
    A_sparse = None
log10(f"Matriz A (solo agregados): shape={A.shape}")

[2025-09-24T12:37:53] Matriz S: shape=(3059, 2429), nnz=9716
[2025-09-24T12:38:02] Matriz A (solo agregados): shape=(630, 2429)

# ----------------------------------------------------------
# Bloque 5 — Estimación de W (diagonal) para bottom
# ----------------------------------------------------------

# Errores 2024 bottom por serie para varianza: e = yhat_base - y_true
preds_bottom = preds_df.merge(keys_bottom, on=PAIR_COLS, how="inner")
preds_bottom = preds_bottom[[*PAIR_COLS, DATECOL, VALCOL_PRED]].copy()
preds_bottom = preds_bottom.set_index(PAIR_COLS + [DATECOL])[VALCOL_PRED].rename("yhat_base").reset_index()

truth = test_2024.set_index(PAIR_COLS + [DATECOL])[VALCOL_TRUE].rename("y_true").reset_index()
base_eval = preds_bottom.merge(truth, on=PAIR_COLS + [DATECOL], how="left")

var_err = {}
for (bid, ctype), g in base_eval.groupby(PAIR_COLS, as_index=False):
    e = (g["yhat_base"].values - g["y_true"].values) if g["y_true"].notna().any() else np.array([])
    if e.size > 0:
        v = float(np.nanvar(e, ddof=1)) if np.sum(~np.isnan(e)) >= 2 else float(np.nanvar(e))
        if not np.isfinite(v) or v <= 0:
            v = 1e-6
    else:
        v = np.nan
    var_err[(bid, ctype)] = v

proxy_from_paso8 = {}
if os.path.exists(PATH_PASO8):
    paso8 = _read_csv(PATH_PASO8, required=False)
    if not paso8.empty:
        paso8.columns = [str(c).strip() for c in paso8.columns]
        if all(c in paso8.columns for c in PAIR_COLS) and "MASE12" in paso8.columns:
            if "vista" in paso8.columns:
                paso8 = paso8[paso8["vista"].astype(str).str.lower().eq("all_months")]
            paso8 = paso8[[*PAIR_COLS, "MASE12"]].drop_duplicates(PAIR_COLS)
            for r in paso8.itertuples(index=False):
                k = (getattr(r, PAIR_COLS[0]), getattr(r, PAIR_COLS[1]))
                proxy_from_paso8[k] = float(pd.to_numeric(getattr(r, "MASE12"), errors="coerce"))

W_diag = np.zeros(n_bottom, dtype=float)
n_from_truth = 0
n_from_proxy = 0
for i, (bid, ctype) in enumerate(keys_bottom.itertuples(index=False, name=None)):
    k = (bid, ctype)
    v = var_err.get(k, np.nan)
    if np.isfinite(v):
        W_diag[i] = max(v, 1e-8)
        n_from_truth += 1
    else:
        m = proxy_from_paso8.get(k, np.nan)
        if np.isfinite(m):
            W_diag[i] = max(m**2, 1e-8)
            n_from_proxy += 1
        else:
            W_diag[i] = 1.0
W_inv_diag = 1.0 / np.maximum(W_diag, 1e-8)

log10(f"W diagonal: bottom={n_bottom} | de verdad_2024={n_from_truth} | de proxy_paso8={n_from_proxy} | resto=identity={n_bottom-n_from_truth-n_from_proxy}")

[2025-09-24T12:38:03] W diagonal: bottom=2429 | de verdad_2024=2231 | de proxy_paso8=0 | resto=identity=198

# ----------------------------------------------------------------
# Bloque 6 — Reconciliación mensual (WLS con A y c)
# ----------------------------------------------------------------

def _month_slice_pred(month_ts):
    sub = preds_df[preds_df[DATECOL].eq(month_ts)]
    merged = keys_bottom.merge(sub, on=PAIR_COLS, how="left")
    yb = merged[VALCOL_PRED].fillna(0.0).values.astype(float)
    return yb

def _month_slice_truth(month_ts):
    sub = test_2024[test_2024[DATECOL].eq(month_ts)]
    merged = keys_bottom.merge(sub, on=PAIR_COLS, how="left")
    yt = merged[VALCOL_TRUE].astype(float).values
    return yt

def _c_from_truth_or_base(month_ts, y_b):
    if A.shape[0] == 0:
        return np.zeros(0, dtype=float)
    y_true_b = _month_slice_truth(month_ts)
    if np.all(np.isnan(y_true_b)):
        return A @ y_b
    return A @ np.nan_to_num(y_true_b, nan=0.0)

def _mint_reconcile_for_month_bottomW(y_b, month_ts):
    m = n_bottom
    if m == 0 or A.shape[0] == 0:
        y_b_rec = np.maximum(y_b, 0.0)
        return y_b_rec

    # c: preferimos verdad agregada; si no hay, A y_b
    c = _c_from_truth_or_base(month_ts, y_b)

    # A W A'  con W diagonal
    if SCIPY_AVAILABLE and A_sparse is not None and A.shape[0] > 0:
        from scipy.sparse import diags, csr_matrix
        Wb = diags(W_diag, offsets=0, format="csr")          # (m x m)
        AW = (A_sparse @ Wb)                                 # (n_agg x m)
        AWAT = (AW @ A_sparse.T).toarray()                   # (n_agg x n_agg) denso para invertir
    else:
        # denso
        AW = A * W_diag                                      # broadcast por columnas (n_agg x m)
        AWAT = AW @ A.T

    # Inversa estable de AWAT
    try:
        AWAT_inv = np.linalg.pinv(AWAT, rcond=1e-10)
    except Exception:
        AWAT_inv = np.linalg.pinv(AWAT + 1e-8*np.eye(AWAT.shape[0]))

    # Residual en agregados
    r = c - (A @ y_b)                                        # (n_agg,)

    # Delta = W A' (A W A')^{-1} r
    v = AWAT_inv @ r                                         # (n_agg,)

    if SCIPY_AVAILABLE and A_sparse is not None and A.shape[0] > 0:
        tmp = (A_sparse.T @ v)                               # (m, 1) o (m,)
        tmp = np.asarray(tmp).ravel()
        delta = W_diag * tmp                                 # (m,)
    else:
        delta = ( (A.T * W_diag[:, None]) @ v )              # (m,)

    y_b_rec = y_b + delta
    # No negatividad en bottom
    y_b_rec = np.maximum(y_b_rec, 0.0)
    return y_b_rec

# Ejecutamos para los 12 meses
reconc_rows = []
for mth in MONTHS_2024:
    yb = _month_slice_pred(mth)
    yb_rec = _mint_reconcile_for_month_bottomW(yb, mth)
    df_month = keys_bottom.copy()
    df_month[DATECOL] = mth
    df_month["yhat_base"]   = yb
    df_month["yhat_reconc"] = yb_rec
    reconc_rows.append(df_month)

preds_reconc = pd.concat(reconc_rows, ignore_index=True)
preds_reconc = preds_reconc[[*PAIR_COLS, DATECOL, "yhat_reconc", "yhat_base"]].sort_values(PAIR_COLS + [DATECOL]).reset_index(drop=True)
preds_reconc.to_csv(OUT_PREDS_RECONC, sep=CSV_SEP, index=False)
log10(f"Guardado: {OUT_PREDS_RECONC} con {len(preds_reconc)} filas.")

[2025-09-24T12:38:06] Guardado: /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3/RESULTADOS/preds_reconciliadas_2024.csv con 29148 filas.

# -----------------------------------------------------------
# Bloque 7 — Métricas y comparación base vs reconciliado
# -----------------------------------------------------------

truth_all = test_2024[[*PAIR_COLS, DATECOL, VALCOL_TRUE]].rename(columns={VALCOL_TRUE:"y_true"}).copy()
base_all  = preds_reconc[[*PAIR_COLS, DATECOL, "yhat_base"]].copy()
reco_all  = preds_reconc[[*PAIR_COLS, DATECOL, "yhat_reconc"]].copy()
eval_df = truth_all.merge(base_all, on=PAIR_COLS+[DATECOL], how="left").merge(reco_all, on=PAIR_COLS+[DATECOL], how="left")

# Escalas MASE a partir de train si existe
scales = {}
if not train_df.empty and all(c in train_df.columns for c in [*PAIR_COLS, DATECOL, VALCOL_TRUE]):
    tr = _ensure_ms(train_df, DATECOL).sort_values(PAIR_COLS + [DATECOL]).set_index(PAIR_COLS)
    for key, g in tr.groupby(level=[0,1]):
        y_hist = g[VALCOL_TRUE].astype(float).values
        s1  = _scale_diff(y_hist, 1)
        s12 = _scale_diff(y_hist, 12)
        scales[key] = {"scale1": s1, "scale12": s12}

def _lookup_scales(row):
    key = (row[PAIR_COLS[0]], row[PAIR_COLS[1]])
    sc = scales.get(key, {"scale1": np.nan, "scale12": np.nan})
    return pd.Series({"scale1": sc["scale1"], "scale12": sc["scale12"]})

if len(scales)>0:
    eval_df = eval_df.merge(eval_df.apply(_lookup_scales, axis=1), left_index=True, right_index=True)
else:
    eval_df["scale1"]  = np.nan
    eval_df["scale12"] = np.nan

mask = eval_df["y_true"].notna()
work = eval_df.loc[mask].copy()
work["abs_err_base"]   = np.abs(work["yhat_base"]   - work["y_true"])
work["abs_err_reconc"] = np.abs(work["yhat_reconc"] - work["y_true"])

def _smape_point(y, yhat, eps=EPS):
    y = float(y) if pd.notna(y) else np.nan
    yhat = float(yhat) if pd.notna(yhat) else np.nan
    if np.isnan(y) or np.isnan(yhat):
        return np.nan
    den = max((abs(y) + abs(yhat))/2.0, eps)
    return 100.0 * abs(yhat - y) / den

work["SMAPE_base"]   = work.apply(lambda r: _smape_point(r["y_true"], r["yhat_base"]),   axis=1)
work["SMAPE_reconc"] = work.apply(lambda r: _smape_point(r["y_true"], r["yhat_reconc"]), axis=1)

def _mase_point(abs_err, scale):
    if scale is None or np.isnan(scale) or scale==0:
        return np.nan
    return float(abs_err / float(scale))

work["MASE12_base"]   = work.apply(lambda r: _mase_point(r["abs_err_base"],   r["scale12"]), axis=1)
work["MASE12_reconc"] = work.apply(lambda r: _mase_point(r["abs_err_reconc"], r["scale12"]), axis=1)

# ----------------------------
# Agregados de comparación
# ----------------------------

def _agg_portfolio_month(df):
    out = []
    for mth, g in df.groupby(DATECOL):
        MAE_b  = float(g["abs_err_base"].mean())
        MAE_r  = float(g["abs_err_reconc"].mean())
        WAPE_b = wape_micro(g["y_true"].values, g["yhat_base"].values)
        WAPE_r = wape_micro(g["y_true"].values, g["yhat_reconc"].values)
        SM_b   = float(g["SMAPE_base"].mean())
        SM_r   = float(g["SMAPE_reconc"].mean())
        M12_b  = float(g["MASE12_base"].mean())
        M12_r  = float(g["MASE12_reconc"].mean())
        out.append({
            "scope":"PORTFOLIO_MENSUAL","clave":None,DATECOL:mth,
            "MAE_base":MAE_b,"MAE_reconc":MAE_r,"ΔMAE":MAE_r-MAE_b,
            "WAPE_base":WAPE_b,"WAPE_reconc":WAPE_r,"ΔWAPE":WAPE_r-WAPE_b,
            "SMAPE_base":SM_b,"SMAPE_reconc":SM_r,"ΔSMAPE":SM_r-SM_b,
            "MASE12_base":M12_b,"MASE12_reconc":M12_r,"ΔMASE12":M12_r-M12_b
        })
    return pd.DataFrame(out)

def _agg_portfolio_annual(df):
    MAE_b  = float(df["abs_err_base"].mean())
    MAE_r  = float(df["abs_err_reconc"].mean())
    WAPE_b = wape_micro(df["y_true"].values, df["yhat_base"].values)
    WAPE_r = wape_micro(df["y_true"].values, df["yhat_reconc"].values)
    SM_b   = float(df["SMAPE_base"].mean())
    SM_r   = float(df["SMAPE_reconc"].mean())
    M12_b  = float(df["MASE12_base"].mean())
    M12_r  = float(df["MASE12_reconc"].mean())
    return pd.DataFrame([{
        "scope":"PORTFOLIO_ANUAL","clave":"2024",DATECOL:pd.Timestamp("2024-12-01"),
        "MAE_base":MAE_b,"MAE_reconc":MAE_r,"ΔMAE":MAE_r-MAE_b,
        "WAPE_base":WAPE_b,"WAPE_reconc":WAPE_r,"ΔWAPE":WAPE_r-WAPE_b,
        "SMAPE_base":SM_b,"SMAPE_reconc":SM_r,"ΔSMAPE":SM_r-SM_b,
        "MASE12_base":M12_b,"MASE12_reconc":M12_r,"ΔMASE12":M12_r-M12_b
    }])

def _agg_segment(df, group_cols, scope_name):
    out = []
    df2 = df.dropna(subset=[c for c in group_cols if c!="FM_COST_TYPE"], how="any") if len(group_cols)>1 else df
    if df2.empty:
        return pd.DataFrame(columns=["scope","clave",DATECOL,"MAE_base","MAE_reconc","ΔMAE",
                                     "WAPE_base","WAPE_reconc","ΔWAPE",
                                     "SMAPE_base","SMAPE_reconc","ΔSMAPE",
                                     "MASE12_base","MASE12_reconc","ΔMASE12"])
    for keys, g in df2.groupby(group_cols + [DATECOL]):
        *seg, mth = keys
        MAE_b  = float(g["abs_err_base"].mean())
        MAE_r  = float(g["abs_err_reconc"].mean())
        WAPE_b = wape_micro(g["y_true"].values, g["yhat_base"].values)
        WAPE_r = wape_micro(g["y_true"].values, g["yhat_reconc"].values)
        SM_b   = float(g["SMAPE_base"].mean())
        SM_r   = float(g["SMAPE_reconc"].mean())
        M12_b  = float(g["MASE12_base"].mean())
        M12_r  = float(g["MASE12_reconc"].mean())
        clave = "|".join(map(str, seg))
        out.append({
            "scope":scope_name,"clave":clave,DATECOL:mth,
            "MAE_base":MAE_b,"MAE_reconc":MAE_r,"ΔMAE":MAE_r-MAE_b,
            "WAPE_base":WAPE_b,"WAPE_reconc":WAPE_r,"ΔWAPE":WAPE_r-WAPE_b,
            "SMAPE_base":SM_b,"SMAPE_reconc":SM_r,"ΔSMAPE":SM_r-SM_b,
            "MASE12_base":M12_b,"MASE12_reconc":M12_r,"ΔMASE12":M12_r-M12_b
        })
    return pd.DataFrame(out)

# Enriquecemos work con PAIS/REGION para agregados de segmento
work = work.merge(dim_df, on="ID_BUILDING", how="left")

# Portfolio mensual y anual
cmp_port_month = _agg_portfolio_month(work)
cmp_port_annual= _agg_portfolio_annual(work)

# Segmentos
seg_fmcost    = _agg_segment(work, ["FM_COST_TYPE"], "SEGMENTO_FMCOST")
seg_fm_pais   = _agg_segment(work, ["FM_COST_TYPE","PAIS"], "SEGMENTO_FMCOST_PAIS") if "PAIS" in work.columns else pd.DataFrame()
seg_fm_region = _agg_segment(work, ["FM_COST_TYPE","REGION"], "SEGMENTO_FMCOST_REGION") if "REGION" in work.columns else pd.DataFrame()

compare_out = pd.concat([cmp_port_month, cmp_port_annual, seg_fmcost, seg_fm_pais, seg_fm_region], ignore_index=True)
compare_out = compare_out.sort_values([ "scope", "clave", DATECOL ]).reset_index(drop=True)
compare_out.to_csv(OUT_COMPARE, sep=CSV_SEP, index=False)
log10(f"Guardado: {OUT_COMPARE} con {len(compare_out)} filas.")

[2025-09-24T12:38:20] Guardado: /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3/REPORTING/compare_metrics_reconc_vs_base.csv con 7021 filas.

# ---------------------------------------------------------
# Bloque 8 — Punto de control G y diagnósticos opcionales
# ---------------------------------------------------------

# Punto G: mejora WAPE anual del portfolio (absoluta en puntos porcentuales)
try:
    row_ann = compare_out[compare_out["scope"]=="PORTFOLIO_ANUAL"].iloc[0]
    wape_b = float(row_ann["WAPE_base"])
    wape_r = float(row_ann["WAPE_reconc"])
    delta_pp = 100.0 * (wape_r - wape_b)
    log10(f"Punto G — WAPE anual portfolio (base={wape_b:.6f}, reconc={wape_r:.6f}, Δ={delta_pp:+.2f} pp)")
    if abs(delta_pp) < 1.0:
        log10("Punto G — Mejora marginal (< 1 pp). Recomendamos conservar versión no reconciliada como entrega principal y dejar esta como anexo.")
    else:
        log10("Punto G — Hay cambio apreciable (>= 1 pp). Podemos considerar entregar la reconciliada si no penaliza métricas clave.")
except Exception as e:
    log10(f"AVISO Punto G — no fue posible calcular WAPE anual del portfolio: {e}")

# Histograma opcional de deltas en bottom (reconc - base) por mes
try:
    import matplotlib.pyplot as plt
    deltas = (preds_reconc["yhat_reconc"] - preds_reconc["yhat_base"]).astype(float)
    plt.figure()
    deltas.plot(kind="hist", bins=30)
    plt.title("Distribución de deltas bottom (yhat_reconc - yhat_base)")
    plt.xlabel("Delta")
    plt.ylabel("Frecuencia")
    fig_path = os.path.join(FIGS_DIR, "hist_deltas_bottom.png")
    plt.tight_layout(); plt.savefig(fig_path); plt.close()
    log10(f"Guardado histograma: {fig_path}")
except Exception as e:
    log10(f"AVISO: no fue posible guardar histograma de deltas: {e}")

log10("Paso 10 finalizado.")

[2025-09-24T12:38:20] Punto G — WAPE anual portfolio (base=0.358801, reconc=0.333307, Δ=-2.55 pp)
[2025-09-24T12:38:20] Punto G — Hay cambio apreciable (>= 1 pp). Podemos considerar entregar la reconciliada si no penaliza métricas clave.
[2025-09-24T12:38:20] Guardado histograma: /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3/REPORTING/figs_step10/hist_deltas_bottom.png
[2025-09-24T12:38:20] Paso 10 finalizado.

# ============================================================
# Bloque 9 - Visualización resultados del Paso 10 — Efecto de la reconciliación (con real 2024)
# Requiere que en el Paso 10 hayamos generado:
# - RESULTADOS/preds_reconciliadas_2024.csv
# - REPORTING/compare_metrics_reconc_vs_base.csv
# - RESULTADOS/test_full_2024.csv
# - METRICAS/dim_buildings.csv (opcional: para PAIS/REGION)
# ============================================================


# Rutas base (mismas del proyecto)
ruta_base_3 = "/content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3"
CSV_SEP = ";"

RUTA_RESULTADOS = os.path.join(ruta_base_3, "RESULTADOS")
RUTA_METRICAS   = os.path.join(ruta_base_3, "METRICAS")
RUTA_REPORTING  = os.path.join(ruta_base_3, "REPORTING")
FIGS_DIR        = os.path.join(RUTA_REPORTING, "figs_step10")
os.makedirs(FIGS_DIR, exist_ok=True)

# Entradas
PATH_PREDS_RECONC = os.path.join(RUTA_RESULTADOS, "preds_reconciliadas_2024.csv")
PATH_COMPARE      = os.path.join(RUTA_REPORTING,  "compare_metrics_reconc_vs_base.csv")
PATH_TEST         = os.path.join(RUTA_RESULTADOS, "test_full_2024.csv")
PATH_DIM          = os.path.join(RUTA_METRICAS,   "dim_buildings.csv")

PAIR_COLS = ["ID_BUILDING","FM_COST_TYPE"]
DATECOL   = "FECHA"
VAL_TRUE  = "cost_float_mod"

def _ensure_ms(df, datecol):
    df = df.copy()
    df[datecol] = pd.to_datetime(df[datecol]).dt.to_period("M").dt.to_timestamp()
    return df

# Carga
preds_reconc = pd.read_csv(PATH_PREDS_RECONC, sep=CSV_SEP)
preds_reconc[DATECOL] = pd.to_datetime(preds_reconc[DATECOL])

compare_df = pd.read_csv(PATH_COMPARE, sep=CSV_SEP)
compare_df[DATECOL] = pd.to_datetime(compare_df[DATECOL], errors="coerce")

truth = pd.read_csv(PATH_TEST, sep=CSV_SEP)
truth = _ensure_ms(truth, DATECOL)[[*PAIR_COLS, DATECOL, VAL_TRUE]]

# Dim opcional
dim = pd.read_csv(PATH_DIM, sep=CSV_SEP) if os.path.exists(PATH_DIM) else pd.DataFrame(columns=["ID_BUILDING","PAIS","REGION"])
if "PAIS" not in dim.columns and "COUNTRY_DEF" in dim.columns:
    dim = dim.rename(columns={"COUNTRY_DEF":"PAIS"})
if "REGION" not in dim.columns and "ID_REGION_GRUPO" in dim.columns:
    dim = dim.rename(columns={"ID_REGION_GRUPO":"REGION"})
dim = dim[[c for c in ["ID_BUILDING","PAIS","REGION"] if c in dim.columns]].drop_duplicates()

# --- 1) TABLA de deltas en bottom (reconc - base) ---
tmp = preds_reconc.copy()
tmp["delta"] = tmp["yhat_reconc"] - tmp["yhat_base"]

# Top ±20 ajustes en absoluto (para inspección)
top_up   = tmp.sort_values("delta", ascending=False).head(20)
top_down = tmp.sort_values("delta", ascending=True).head(20)

out_table = pd.concat([
    top_up.assign(tag="TOP_POS"),
    top_down.assign(tag="TOP_NEG")
]).merge(dim, on="ID_BUILDING", how="left")

out_table_path = os.path.join(RUTA_REPORTING, "compare_vistas_visual.csv")
out_table.to_csv(out_table_path, sep=CSV_SEP, index=False)

# --- 2) AGREGADO PORTFOLIO: antes vs después vs verdad ---
portfolio = (tmp
             .merge(truth, on=PAIR_COLS+[DATECOL], how="left")
             .groupby(DATECOL, as_index=False)
             .agg(y_base=("yhat_base","sum"),
                  y_reco=("yhat_reconc","sum"),
                  y_true=(VAL_TRUE,"sum")))

# --- 3) AGREGADO FM_COST_TYPE (mensual) ---
by_fm = (tmp
         .merge(truth, on=PAIR_COLS+[DATECOL], how="left")
         .groupby(["FM_COST_TYPE", DATECOL], as_index=False)
         .agg(y_base=("yhat_base","sum"),
              y_reco=("yhat_reconc","sum"),
              y_true=(VAL_TRUE,"sum")))

# --- 4) PLOTS ---

# 4.1 Histograma de deltas bottom
plt.figure()
tmp["delta"].plot(kind="hist", bins=40)
plt.title("Deltas bottom (yhat_reconc - yhat_base)")
plt.xlabel("Delta"); plt.ylabel("Frecuencia")
plt.tight_layout(); plt.savefig(os.path.join(FIGS_DIR, "hist_deltas_bottom.png")); plt.close()

# 4.2 Boxplot de deltas por FM_COST_TYPE
plt.figure(figsize=(10,5))
tmp_box = tmp.groupby(["FM_COST_TYPE"]).apply(lambda d: d["delta"].values)
# Convertimos para boxplot
data_box = [tmp[tmp["FM_COST_TYPE"]==k]["delta"].values for k in sorted(tmp["FM_COST_TYPE"].unique())]
plt.boxplot(data_box, labels=sorted(tmp["FM_COST_TYPE"].unique()), showfliers=False)
plt.title("Distribución de deltas por FM_COST_TYPE")
plt.xlabel("FM_COST_TYPE"); plt.ylabel("Delta (reconc - base)")
plt.xticks(rotation=45, ha="right")
plt.tight_layout(); plt.savefig(os.path.join(FIGS_DIR, "box_deltas_por_fmcost.png")); plt.close()

# 4.3 Scatter y_base vs y_reco (portfolio mensual) con línea y=x
plt.figure()
plt.scatter(portfolio["y_base"], portfolio["y_reco"])
minv = float(np.nanmin([portfolio["y_base"].min(), portfolio["y_reco"].min()]))
maxv = float(np.nanmax([portfolio["y_base"].max(), portfolio["y_reco"].max()]))
plt.plot([minv, maxv], [minv, maxv])
plt.title("Portfolio mensual: y_base vs y_reco")
plt.xlabel("y_base (suma mes)"); plt.ylabel("y_reco (suma mes)")
plt.tight_layout(); plt.savefig(os.path.join(FIGS_DIR, "scatter_portfolio_base_vs_reco.png")); plt.close()

# 4.4 Líneas en el tiempo (portfolio): base vs reco vs true
plt.figure(figsize=(10,5))
plt.plot(portfolio[DATECOL], portfolio["y_base"], label="base")
plt.plot(portfolio[DATECOL], portfolio["y_reco"], label="reconc")
if portfolio["y_true"].notna().any():
    plt.plot(portfolio[DATECOL], portfolio["y_true"], label="true")
plt.title("Portfolio mensual — total (base vs reconc vs true)")
plt.xlabel("Mes"); plt.ylabel("Suma mensual")
plt.legend()
plt.tight_layout(); plt.savefig(os.path.join(FIGS_DIR, "line_portfolio_base_reco_true.png")); plt.close()

# 4.5 Barras apiladas por FM_COST_TYPE (anual): base vs reco vs true
by_fm_year = by_fm.groupby("FM_COST_TYPE", as_index=False).agg(y_base=("y_base","sum"),
                                                               y_reco=("y_reco","sum"),
                                                               y_true=("y_true","sum"))
by_fm_year = by_fm_year.sort_values("y_true", ascending=False)

x = np.arange(len(by_fm_year))
w = 0.35
plt.figure(figsize=(12,5))
plt.bar(x - w/2, by_fm_year["y_base"], width=w, label="base")
plt.bar(x + w/2, by_fm_year["y_reco"], width=w, label="reconc")
plt.xticks(x, by_fm_year["FM_COST_TYPE"], rotation=45, ha="right")
plt.title("Total 2024 por FM_COST_TYPE — base vs reconc")
plt.ylabel("Suma anual")
plt.legend()
plt.tight_layout(); plt.savefig(os.path.join(FIGS_DIR, "bar_fmcost_anual_base_vs_reco.png")); plt.close()

# 4.6 Efecto en error relativo mensual (WAPE) a nivel portfolio
def wape_micro(y_true, y_pred, eps=1e-8):
    y_true = np.asarray(y_true, dtype=float)
    y_pred = np.asarray(y_pred, dtype=float)
    return float(np.sum(np.abs(y_pred - y_true)) / (np.sum(np.abs(y_true)) + eps))

wapes = []
for m, g in (tmp.merge(truth, on=PAIR_COLS+[DATECOL], how="left")
               .groupby(DATECOL)):
    y_true = g[VAL_TRUE].values
    yb = g["yhat_base"].values
    yr = g["yhat_reconc"].values
    if np.isfinite(y_true).sum() == 0:
        continue
    wapes.append((m, wape_micro(y_true, yb), wape_micro(y_true, yr)))

wapes = pd.DataFrame(wapes, columns=[DATECOL,"WAPE_base","WAPE_reconc"]).sort_values(DATECOL)
plt.figure(figsize=(10,5))
plt.plot(wapes[DATECOL], wapes["WAPE_base"], label="WAPE base")
plt.plot(wapes[DATECOL], wapes["WAPE_reconc"], label="WAPE reconc")
plt.title("Portfolio — WAPE mensual (menor es mejor)")
plt.xlabel("Mes"); plt.ylabel("WAPE")
plt.legend()
plt.tight_layout(); plt.savefig(os.path.join(FIGS_DIR, "line_portfolio_WAPE_mensual.png")); plt.close()

# 4.7 Top 20 ajustes absolutos (barras horizontales)
tops = (tmp.assign(abs_delta=tmp["delta"].abs())
          .sort_values("abs_delta", ascending=False)
          .head(20)
          .merge(dim, on="ID_BUILDING", how="left"))
labs = tops.apply(lambda r: f"{r['ID_BUILDING']}|{r['FM_COST_TYPE']}|{str(r.get('PAIS',''))}", axis=1)
plt.figure(figsize=(10,8))
plt.barh(range(len(tops)), tops["delta"])
plt.yticks(range(len(tops)), labs)
plt.gca().invert_yaxis()
plt.title("Top 20 ajustes absolutos en bottom (reconc - base)")
plt.xlabel("Delta")
plt.tight_layout(); plt.savefig(os.path.join(FIGS_DIR, "barh_top20_deltas_abs.png")); plt.close()

print("Listo. Archivos generados:")
print(" - Tabla de inspección:", out_table_path)
print(" - Figuras en:", FIGS_DIR)

Listo. Archivos generados:
 - Tabla de inspección: /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3/REPORTING/compare_vistas_visual.csv
 - Figuras en: /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3/REPORTING/figs_step10

# Visualizamos los resultados

# Rutas base
ruta_base_3 = "/content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3"
CSV_SEP = ";"

RUTA_REPORTING  = os.path.join(ruta_base_3, "REPORTING")
FIGS_DIR        = os.path.join(RUTA_REPORTING, "figs_step10")
TABLE_VISUAL    = os.path.join(RUTA_REPORTING, "compare_vistas_visual.csv")

# 1) Cargar tabla de inspección
if os.path.exists(TABLE_VISUAL):
    df_visual = pd.read_csv(TABLE_VISUAL, sep=CSV_SEP)
    print("Tabla de inspección (primeras filas):")
    display(df_visual.head(10))
else:
    print("No se encontró compare_vistas_visual.csv")

# 2) Mostrar imágenes de la carpeta de figuras
if os.path.exists(FIGS_DIR):
    figs = [f for f in os.listdir(FIGS_DIR) if f.lower().endswith((".png",".jpg",".jpeg"))]
    for f in sorted(figs):
        img_path = os.path.join(FIGS_DIR, f)
        img = mpimg.imread(img_path)
        plt.figure(figsize=(8,5))
        plt.imshow(img)
        plt.axis("off")
        plt.title(f)
        plt.show()
else:
    print("No se encontró la carpeta de figuras:", FIGS_DIR)

Tabla de inspección (primeras filas):

dim_df.head(20)

# ============================================================
# ESTRATEGIA 3 — PASO 11 (PRODUCCIÓN)
# Reconciliación jerárquica MinT (shrink) sin datos reales en test
# ============================================================
# En este paso:
#  - Leemos predicciones base 2024 por serie (bottom)
#  - Leemos anclas/targets por nivel agregado si existen (p.ej. planes)
#  - Construimos jerarquía mínima:
#       Bottom: (ID_BUILDING, FM_COST_TYPE)
#       Nivel 1 opcional: (FM_COST_TYPE, PAIS) si PAIS existe
#       Nivel 2 opcional: (FM_COST_TYPE, REGION) si REGION existe
#       Top opcional: (FM_COST_TYPE)
#  - Reconciliamos mes a mes mediante MinT (shrink simple) en forma WLS con restricciones:
#       y_b_rec = y_b + W_b A' (A W_b A')^{-1} (c - A y_b)
#       A es (n_agg x n_bottom); W_b diagonal (varianzas bottom)
#       c es el vector de objetivos agregados:
#          * si tenemos anclas (planes) para ese mes y nivel -> las usamos
#          * si no, usamos c = A y_b (autoconsistencia)
#  - Clippamos a >= 0 en bottom
#  - Dado que no tenemos verdad, construimos comparativos estructurales
#    (consistencia por nivel, magnitud de ajustes, distribución de deltas).
#  - Dejamos salidas en RESULTADOS/ y REPORTING/ para auditoría.
# ============================================================

# -----------------------------
# Bloque 0 — Parámetros y rutas
# -----------------------------

# Fijamos semillas
np.random.seed(7)

# Rutas base (mismas convenciones del proyecto)
ruta_base_3 = "/content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3"
CSV_SEP     = ";"

RUTA_RESULTADOS = os.path.join(ruta_base_3, "RESULTADOS")
RUTA_METRICAS   = os.path.join(ruta_base_3, "METRICAS")
RUTA_REPORTING  = os.path.join(ruta_base_3, "REPORTING")
RUTA_LOGS       = os.path.join(ruta_base_3, "LOGS")

os.makedirs(RUTA_RESULTADOS, exist_ok=True)
os.makedirs(RUTA_METRICAS,   exist_ok=True)
os.makedirs(RUTA_REPORTING,  exist_ok=True)
os.makedirs(RUTA_LOGS,       exist_ok=True)

# Entradas obligatorias
PATH_PREDS = os.path.join(RUTA_RESULTADOS, "preds_por_serie_2024.csv")  # del Paso 5

# Entradas opcionales
PATH_DIM     = os.path.join(RUTA_METRICAS, "dim_buildings.csv")         # del Paso 9 (PAIS/REGION)
PATH_PASO8   = os.path.join(RUTA_METRICAS, "paso8_metrics_2024_por_pareja.csv")  # para proxy de W
PATH_TRAIN   = os.path.join(RUTA_RESULTADOS, "train_full_2021_2023.csv")         # para escalas/var proxy
PATH_ANCLA   = os.path.join(RUTA_METRICAS, "anclas_reconciliacion_2024.csv")     # OPCIONAL: objetivos por nivel

# Salidas
OUT_PREDS_RECONC = os.path.join(RUTA_RESULTADOS, "preds_reconciliadas_2024_prod.csv")
OUT_COMPARE      = os.path.join(RUTA_REPORTING,  "compare_estructural_reconc_vs_base_prod.csv")
FIGS_DIR         = os.path.join(RUTA_REPORTING,  "figs_step11")
os.makedirs(FIGS_DIR, exist_ok=True)

# Claves/fechas
PAIR_COLS   = ["ID_BUILDING", "FM_COST_TYPE"]
DATECOL     = "FECHA"
VALCOL_PRED = "yhat_combo"
H           = 12
MONTHS_2024 = pd.date_range("2024-01-01", periods=H, freq="MS")

# Logging simple
LOG_PATH = os.path.join(RUTA_LOGS, "estrategia3_step11.log")
def log11(msg: str):
    ts = datetime.now().isoformat(timespec="seconds")
    line = f"[{ts}] {msg}"
    print(line)
    with open(LOG_PATH, "a", encoding="utf-8") as f:
        f.write(line + "\n")

# -------------------------------------
# Bloque 1 — Utilidades generales/IO
# -------------------------------------

EPS = 1e-8

def _ensure_ms(df: pd.DataFrame, datecol: str) -> pd.DataFrame:
    # Normalizamos a comienzo de mes (MS) para alinear malla
    df = df.copy()
    df[datecol] = pd.to_datetime(df[datecol]).dt.to_period("M").dt.to_timestamp()
    return df

def _read_csv(path: str, required: bool=True) -> pd.DataFrame:
    # Leemos CSV con limpieza de columnas; si required=False y no existe, devolvemos vacío
    if os.path.exists(path):
        df = pd.read_csv(path, sep=CSV_SEP)
        df.columns = [str(c).strip().lstrip("\ufeff") for c in df.columns]
        return df
    if required:
        raise FileNotFoundError(f"Falta archivo requerido: {path}")
    log11(f"AVISO: no encontramos {os.path.basename(path)} (continuamos sin este artefacto).")
    return pd.DataFrame()

def _norm_pair_keys(df: pd.DataFrame) -> pd.DataFrame:
    # Normalizamos claves de par y tipamos lo que se pueda
    if df is None or df.empty:
        return df
    out = df.copy()
    out.columns = [str(c).strip() for c in out.columns]
    if "ID_BUILDING" in out.columns:
        try:
            out["ID_BUILDING"] = pd.to_numeric(out["ID_BUILDING"], errors="coerce").astype("Int64")
        except Exception:
            pass
    if "FM_COST_TYPE" in out.columns:
        out["FM_COST_TYPE"] = out["FM_COST_TYPE"].astype(str).str.strip()
    return out

def _safe_div(num, den, eps=EPS):
    return num / (den + eps)

# -----------------------------------------------------------------
# Bloque 2 — Carga de entradas y preparación (sin verdad terreno)
# -----------------------------------------------------------------

log11("Iniciamos Paso 11 — Reconciliación MinT (producción, sin verdad).")

# Predicciones bottom (obligatorias)
preds_df = _ensure_ms(_read_csv(PATH_PREDS, required=True), DATECOL)
preds_df = _norm_pair_keys(preds_df)
if VALCOL_PRED not in preds_df.columns:
    raise ValueError(f"Esperábamos columna {VALCOL_PRED} en {os.path.basename(PATH_PREDS)}.")

# Nos quedamos con las columnas mínimas más trazabilidad (si existen)
keep_pred_cols = [c for c in [*PAIR_COLS, DATECOL, VALCOL_PRED, "route", "fallback_flag"] if c in preds_df.columns]
preds_df = preds_df[keep_pred_cols].copy()

# Agregamos por si hubiera duplicados de (pair, mes)
preds_df = (preds_df
            .groupby(PAIR_COLS + [DATECOL], as_index=False)
            .agg({VALCOL_PRED: "sum",
                  **({"route":"last"} if "route" in preds_df.columns else {}),
                  **({"fallback_flag":"max"} if "fallback_flag" in preds_df.columns else {})}))

# Dimensiones opcionales
dim_df = _read_csv(PATH_DIM, required=False)
if not dim_df.empty:
    # Normalizamos nombres esperados
    dim_cols_map = {}
    for c in dim_df.columns:
        cu = c.strip().upper()
        if cu == "PAIS":          dim_cols_map[c] = "PAIS"
        elif cu == "REGION":      dim_cols_map[c] = "REGION"
        elif cu == "ID_BUILDING": dim_cols_map[c] = "ID_BUILDING"
    dim_df = dim_df.rename(columns=dim_cols_map)
    dim_df = _norm_pair_keys(dim_df)
    dim_df = dim_df[[c for c in ["ID_BUILDING","PAIS","REGION"] if c in dim_df.columns]].drop_duplicates()
else:
    dim_df = pd.DataFrame(columns=["ID_BUILDING","PAIS","REGION"])

# Posibles anclas (opcional)
# Esperamos un archivo con columnas:
# - nivel: uno de {"FM_COST_TYPE","FM_COST_TYPE|PAIS","FM_COST_TYPE|REGION"}
# - clave: concatenación de las llaves del nivel (p.ej. "ENERGIA|ESPAÑA")
# - FECHA: primer día de mes (MS)
# - target: valor objetivo para ese nivel y mes
ancla_df = _read_csv(PATH_ANCLA, required=False)
if not ancla_df.empty:
    # Normalizamos fechas y nombres
    if DATECOL in ancla_df.columns:
        ancla_df = _ensure_ms(ancla_df, DATECOL)
    else:
        raise KeyError("El CSV de anclas debe incluir columna FECHA.")
    # Limpiamos
    ancla_df.columns = [str(c).strip() for c in ancla_df.columns]
    needed = {"nivel","clave",DATECOL,"target"}
    if not needed.issubset(set(ancla_df.columns)):
        raise KeyError(f"Anclas: faltan columnas {list(needed - set(ancla_df.columns))}")
    # Filtramos a 2024 y reindexamos por seguridad
    ancla_df = ancla_df[ancla_df[DATECOL].isin(MONTHS_2024)].copy()

log11(f"Cargadas entradas: preds={len(preds_df)} filas, dim={len(dim_df)} filas, anclas={'sí' if not ancla_df.empty else 'no'}")
log11(f"scipy.sparse disponible: {SCIPY_AVAILABLE}")

[2025-09-26T06:02:22] Iniciamos Paso 11 — Reconciliación MinT (producción, sin verdad).
[2025-09-26T06:02:23] AVISO: no encontramos anclas_reconciliacion_2024.csv (continuamos sin este artefacto).
[2025-09-26T06:02:23] Cargadas entradas: preds=29148 filas, dim=1650 filas, anclas=no
[2025-09-26T06:02:23] scipy.sparse disponible: True

# ------------------------------------------------------------
# Bloque 3 — Preparación del panel y metadatos jerárquicos
# ------------------------------------------------------------

# Unimos PAIS/REGION al bottom
bottom_df = preds_df.merge(dim_df, on="ID_BUILDING", how="left")
n_bottom = bottom_df[PAIR_COLS].drop_duplicates().shape[0]

# Diagnóstico metadatos
pct_no_pais   = 1.0 - float(bottom_df["PAIS"].notna().mean()) if "PAIS" in bottom_df.columns else 1.0
pct_no_region = 1.0 - float(bottom_df["REGION"].notna().mean()) if "REGION" in bottom_df.columns else 1.0
log11(f"Metadatos: n_bottom={n_bottom} | % sin PAIS={100*pct_no_pais:.1f}% | % sin REGION={100*pct_no_region:.1f}%")

# Claves por nivel presentes
def _unique_keys(df, cols):
    cols = [c for c in cols if c in df.columns]
    if len(cols)==0:
        return pd.DataFrame(columns=cols)
    return df[cols].dropna().drop_duplicates()

keys_top       = _unique_keys(bottom_df, ["FM_COST_TYPE"])
keys_fm_pais   = _unique_keys(bottom_df, ["FM_COST_TYPE","PAIS"])
keys_fm_region = _unique_keys(bottom_df, ["FM_COST_TYPE","REGION"])
keys_bottom    = bottom_df[PAIR_COLS].drop_duplicates().sort_values(PAIR_COLS).reset_index(drop=True)

if keys_top.empty:
    log11("AVISO: capa TOP (FM_COST_TYPE) vacía; seguimos sin top.")
if "PAIS" not in bottom_df.columns or keys_fm_pais.empty:
    log11("AVISO: capa (FM_COST_TYPE, PAIS) no disponible; seguimos sin esa capa.")
if "REGION" not in bottom_df.columns or keys_fm_region.empty:
    log11("AVISO: capa (FM_COST_TYPE, REGION) no disponible; seguimos sin esa capa.")

# Índices auxiliares
def _child_idx_for_top(fm):
    mask = (keys_bottom["FM_COST_TYPE"]==fm)
    return keys_bottom.index[mask].tolist()

def _child_idx_for_fm_pais(fm, pais):
    if "PAIS" not in bottom_df.columns:
        return []
    sub = bottom_df.merge(keys_bottom[[*PAIR_COLS]], on=PAIR_COLS, how="inner")
    sub = sub[(sub["FM_COST_TYPE"]==fm) & (sub["PAIS"]==pais)]
    idx = keys_bottom.merge(sub[PAIR_COLS].drop_duplicates(), on=PAIR_COLS, how="inner").index.tolist()
    return idx

def _child_idx_for_fm_region(fm, region):
    if "REGION" not in bottom_df.columns:
        return []
    sub = bottom_df.merge(keys_bottom[[*PAIR_COLS]], on=PAIR_COLS, how="inner")
    sub = sub[(sub["FM_COST_TYPE"]==fm) & (sub["REGION"]==region)]
    idx = keys_bottom.merge(sub[PAIR_COLS].drop_duplicates(), on=PAIR_COLS, how="inner").index.tolist()
    return idx

n_top       = len(keys_top) if not keys_top.empty else 0
n_fm_pais   = len(keys_fm_pais) if not keys_fm_pais.empty else 0
n_fm_region = len(keys_fm_region) if not keys_fm_region.empty else 0
log11(f"Árbol: top={n_top} | fm×pais={n_fm_pais} | fm×region={n_fm_region} | bottom={n_bottom}")

[2025-09-26T06:02:27] Metadatos: n_bottom=2429 | % sin PAIS=0.0% | % sin REGION=0.0%
[2025-09-26T06:02:27] Árbol: top=8 | fm×pais=49 | fm×region=573 | bottom=2429

# -----------------------------------------------------
# Bloque 4 — Construcción de A (solo niveles agregados)
# -----------------------------------------------------

def _build_A_matrix(n_bottom, keys_top, keys_fm_pais, keys_fm_region):
    agg_rows = []

    def _add_row(child_idx_list):
        if not child_idx_list:      # evita fila cero
            return
        row = np.zeros(n_bottom, dtype=float)
        row[np.asarray(child_idx_list, dtype=int)] = 1.0
        agg_rows.append(row)

    if not keys_top.empty:
        for (fm,) in keys_top.itertuples(index=False, name=None):
            _add_row(_child_idx_for_top(fm))

    if not keys_fm_pais.empty:
        for (fm, pais) in keys_fm_pais.itertuples(index=False, name=None):
            _add_row(_child_idx_for_fm_pais(fm, pais))

    if not keys_fm_region.empty:
        for (fm, region) in keys_fm_region.itertuples(index=False, name=None):
            _add_row(_child_idx_for_fm_region(fm, region))

    return np.vstack(agg_rows) if agg_rows else np.zeros((0, n_bottom), dtype=float)

A = _build_A_matrix(n_bottom, keys_top, keys_fm_pais, keys_fm_region)
A_sparse = (sp.csr_matrix(A) if (SCIPY_AVAILABLE and A.size>0) else None)
log11(f"Matriz A (agregados): shape={A.shape}")

[2025-09-26T06:03:37] Matriz A (agregados): shape=(630, 2429)

# ---------------------------------------------------------
# Bloque 5 — Estimación de W (diagonal) bottom sin verdad
# ---------------------------------------------------------
# Sin verdad, nos apoyamos en proxies:
#  - Si tenemos Paso 8 con MASE12 por pareja, usamos MASE12^2 como proxy de varianza
#  - Si tenemos train, podemos usar dispersión histórica como proxy (por ejemplo, var de primeras diferencias)
#  - Fallback: identidad

# Proxy desde Paso 8
proxy_mase = {}
paso8_df = _read_csv(PATH_PASO8, required=False)
if not paso8_df.empty:
    paso8_df.columns = [str(c).strip() for c in paso8_df.columns]
    if set(PAIR_COLS).issubset(paso8_df.columns) and "MASE12" in paso8_df.columns:
        if "vista" in paso8_df.columns:
            paso8_df = paso8_df[paso8_df["vista"].astype(str).str.lower().eq("all_months")]
        paso8_df = paso8_df[[*PAIR_COLS, "MASE12"]].drop_duplicates(PAIR_COLS)
        for r in paso8_df.itertuples(index=False):
            proxy_mase[(getattr(r, PAIR_COLS[0]), getattr(r, PAIR_COLS[1]))] = float(pd.to_numeric(getattr(r, "MASE12"), errors="coerce"))

# Proxy desde train (dispersión histórica de primeras diferencias)
proxy_hist = {}
train_df = _read_csv(PATH_TRAIN, required=False)
if not train_df.empty and all(c in train_df.columns for c in [*PAIR_COLS, DATECOL, "cost_float_mod"]):
    tr = _ensure_ms(_norm_pair_keys(train_df), DATECOL)
    for (bid, ctype), g in tr.groupby(PAIR_COLS, as_index=False):
        y = pd.to_numeric(g["cost_float_mod"], errors="coerce").astype(float).values
        if len(y) >= 13:
            diffs = np.diff(y)
            v = float(np.nanvar(diffs, ddof=1)) if np.sum(~np.isnan(diffs))>=2 else float(np.nanvar(diffs))
            if not np.isfinite(v) or v<=0: v = np.nan
        else:
            v = np.nan
        proxy_hist[(bid, ctype)] = v

# Combinamos proxies para W_diag
keys_bottom_tuples = [tuple(x) for x in keys_bottom[PAIR_COLS].itertuples(index=False, name=None)]
W_diag = np.zeros(n_bottom, dtype=float)
n_from_mase, n_from_hist = 0, 0

for i, k in enumerate(keys_bottom_tuples):
    v = np.nan
    # 1) MASE12^2
    m = proxy_mase.get(k, np.nan)
    if np.isfinite(m):
        v = max(m**2, 1e-8)
        n_from_mase += 1
    else:
        # 2) Var de diffs históricas
        h = proxy_hist.get(k, np.nan)
        if np.isfinite(h) and h>0:
            v = max(h, 1e-8)
            n_from_hist += 1
        else:
            # 3) Fallback identidad (peso uniforme)
            v = 1.0
    W_diag[i] = v

W_diag = np.maximum(W_diag, 1e-8)
W_inv_diag = 1.0 / np.maximum(W_diag, 1e-8)
log11(f"W diagonal: bottom={n_bottom} | de MASE12={n_from_mase} | de hist_diffs={n_from_hist} | resto=identity={n_bottom - n_from_mase - n_from_hist}")

[2025-09-26T06:03:40] W diagonal: bottom=2429 | de MASE12=2171 | de hist_diffs=153 | resto=identity=105

# ----------------------------------------------------------------
# Bloque 6 — c (objetivos por nivel): referencia o ancla externa o autoconsistencia
# ----------------------------------------------------------------

# Construimos helpers para obtener y_b del mes y el vector c del mes
def _month_slice_pred(month_ts):
    sub = preds_df[preds_df[DATECOL].eq(month_ts)]
    merged = keys_bottom.merge(sub, on=PAIR_COLS, how="left")
    yb = merged[VALCOL_PRED].fillna(0.0).values.astype(float)
    return yb

def _c_from_anchor_or_base(month_ts, y_b):
    # Si no hay A (sin agregados), devolvemos vector vacío
    if A.shape[0] == 0:
        return np.zeros(0, dtype=float)

    # Si hay anclas y coinciden con nuestra capa, las usamos
    if not ancla_df.empty:
        # Construimos A y orden de filas: primero TOP, luego FM×PAIS, luego FM×REGION (mismo orden aplicado en A)
        targets = []

        # TOP: "FM_COST_TYPE"
        if not keys_top.empty:
            for (fm,) in keys_top.itertuples(index=False, name=None):
                clave = f"{fm}"
                row = ancla_df[(ancla_df["nivel"]=="FM_COST_TYPE") &
                               (ancla_df["clave"]==clave) &
                               (ancla_df[DATECOL].eq(month_ts))]
                if len(row)==1:
                    targets.append(float(row["target"].iloc[0]))
                else:
                    targets.append(np.nan)  # si no hay ancora, ponemos NaN para rellenar luego

        # FM×PAIS
        if not keys_fm_pais.empty:
            for (fm, pais) in keys_fm_pais.itertuples(index=False, name=None):
                clave = f"{fm}|{pais}"
                row = ancla_df[(ancla_df["nivel"]=="FM_COST_TYPE|PAIS") &
                               (ancla_df["clave"]==clave) &
                               (ancla_df[DATECOL].eq(month_ts))]
                targets.append(float(row["target"].iloc[0]) if len(row)==1 else np.nan)

        # FM×REGION
        if not keys_fm_region.empty:
            for (fm, region) in keys_fm_region.itertuples(index=False, name=None):
                clave = f"{fm}|{region}"
                row = ancla_df[(ancla_df["nivel"]=="FM_COST_TYPE|REGION") &
                               (ancla_df["clave"]==clave) &
                               (ancla_df[DATECOL].eq(month_ts))]
                targets.append(float(row["target"].iloc[0]) if len(row)==1 else np.nan)

        c = np.asarray(targets, dtype=float)
        # Donde no tengamos ancora, nos autoconsistimos con A y_b
        if np.any(np.isnan(c)):
            c_base = A @ y_b
            c = np.where(np.isnan(c), c_base, c)
        return c

    # Sin anclas, usamos autoconsistencia
    return A @ y_b

# ----------------------------------------------------------------
# Bloque 7 — Reconciliación mensual (WLS con A y c sin verdad)
# ----------------------------------------------------------------

def _mint_reconcile_for_month_bottomW(y_b, month_ts):
    # Si no hay agregados, devolvemos bottom clippeado
    if (n_bottom == 0) or (A.shape[0] == 0):
        y_b_rec = np.maximum(y_b, 0.0)
        return y_b_rec

    # Objetivo c desde ancla o base
    c = _c_from_anchor_or_base(month_ts, y_b)

    # A W A' con W diagonal
    if SCIPY_AVAILABLE and A_sparse is not None and A.shape[0] > 0:
        from scipy.sparse import diags
        Wb = diags(W_diag, offsets=0, format="csr")
        AW = (A_sparse @ Wb)            # (n_agg x m)
        AWAT = (AW @ A_sparse.T).toarray()
    else:
        AW = A * W_diag                 # (n_agg x m)
        AWAT = AW @ A.T                 # (n_agg x n_agg)

    # Inversa estable
    try:
        AWAT_inv = np.linalg.pinv(AWAT, rcond=1e-10)
    except Exception:
        AWAT_inv = np.linalg.pinv(AWAT + 1e-8*np.eye(AWAT.shape[0]))

    # Residual en agregados
    r = c - (A @ y_b)

    # Delta = W A' (A W A')^{-1} r
    v = AWAT_inv @ r
    if SCIPY_AVAILABLE and A_sparse is not None and A.shape[0] > 0:
        tmp = (A_sparse.T @ v)
        tmp = np.asarray(tmp).ravel()
        delta = W_diag * tmp
    else:
        delta = ((A.T * W_diag[:, None]) @ v)

    y_b_rec = y_b + delta
    # No negatividad
    y_b_rec = np.maximum(y_b_rec, 0.0)
    return y_b_rec

# Ejecutamos reconciliación para los 12 meses
reconc_rows = []
for mth in MONTHS_2024:
    yb = _month_slice_pred(mth)
    yb_rec = _mint_reconcile_for_month_bottomW(yb, mth)
    df_month = keys_bottom.copy()
    df_month[DATECOL] = mth
    df_month["yhat_base"]   = yb
    df_month["yhat_reconc"] = yb_rec
    reconc_rows.append(df_month)

preds_reconc = pd.concat(reconc_rows, ignore_index=True)
preds_reconc = preds_reconc[[*PAIR_COLS, DATECOL, "yhat_reconc", "yhat_base"]].sort_values(PAIR_COLS + [DATECOL]).reset_index(drop=True)
preds_reconc.to_csv(OUT_PREDS_RECONC, sep=CSV_SEP, index=False)
log11(f"Guardado: {OUT_PREDS_RECONC} con {len(preds_reconc)} filas.")

[2025-09-26T06:03:55] Guardado: /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3/RESULTADOS/preds_reconciliadas_2024_prod.csv con 29148 filas.

# ----------------------------------------------------------------
# Bloque 8 — Comparaciones estructurales (sin verdad)
# ----------------------------------------------------------------
# Como no tenemos verdad, generamos diagnósticos que nos permitan:
#  - Ver la magnitud de ajustes por mes y por nivel
#  - Verificar consistencia de sumas en cada capa
#  - Medir cuánto nos alejamos de anclas cuando existen

def _agg_sum(df, valcol, group_cols):
    return (df.groupby(group_cols, as_index=False)
              .agg(total=(valcol, "sum")))

# Construimos panel base/reconc
tmp = preds_reconc.copy()
tmp["delta"] = tmp["yhat_reconc"] - tmp["yhat_base"]

# Portfolio mensual base vs reconc
port_base  = _agg_sum(tmp, "yhat_base",   [DATECOL])
port_reco  = _agg_sum(tmp, "yhat_reconc", [DATECOL])
port = port_base.merge(port_reco, on=[DATECOL], suffixes=("_base","_reco"))
port["delta_portfolio"] = port["total_reco"] - port["total_base"]

# Agregado por FM_COST_TYPE
fm_base = _agg_sum(tmp, "yhat_base",   ["FM_COST_TYPE", DATECOL])
fm_reco = _agg_sum(tmp, "yhat_reconc", ["FM_COST_TYPE", DATECOL])
fm = fm_base.merge(fm_reco, on=["FM_COST_TYPE", DATECOL], suffixes=("_base","_reco"))
fm["delta_fm"] = fm["total_reco"] - fm["total_base"]

# Agregado por FM_COST_TYPE×PAIS (si existe)
if "PAIS" in bottom_df.columns and bottom_df["PAIS"].notna().any():
    df2 = tmp.merge(dim_df, on="ID_BUILDING", how="left")
    fmp_base = _agg_sum(df2, "yhat_base",   ["FM_COST_TYPE","PAIS", DATECOL])
    fmp_reco = _agg_sum(df2, "yhat_reconc", ["FM_COST_TYPE","PAIS", DATECOL])
    fmp = fmp_base.merge(fmp_reco, on=["FM_COST_TYPE","PAIS",DATECOL], suffixes=("_base","_reco"))
    fmp["delta_fm_pais"] = fmp["total_reco"] - fmp["total_base"]
else:
    fmp = pd.DataFrame(columns=["FM_COST_TYPE","PAIS",DATECOL,"total_base","total_reco","delta_fm_pais"])

# Agregado por FM_COST_TYPE×REGION (si existe)
if "REGION" in bottom_df.columns and bottom_df["REGION"].notna().any():
    df3 = tmp.merge(dim_df, on="ID_BUILDING", how="left")
    fmr_base = _agg_sum(df3, "yhat_base",   ["FM_COST_TYPE","REGION", DATECOL])
    fmr_reco = _agg_sum(df3, "yhat_reconc", ["FM_COST_TYPE","REGION", DATECOL])
    fmr = fmr_base.merge(fmr_reco, on=["FM_COST_TYPE","REGION",DATECOL], suffixes=("_base","_reco"))
    fmr["delta_fm_region"] = fmr["total_reco"] - fmr["total_base"]
else:
    fmr = pd.DataFrame(columns=["FM_COST_TYPE","REGION",DATECOL,"total_base","total_reco","delta_fm_region"])

# Si hay anclas, medimos gap vs target
gaps_rows = []
if not ancla_df.empty and A.shape[0] > 0:
    # TOP
    if not keys_top.empty:
        agg_top = fm.groupby(["FM_COST_TYPE", DATECOL], as_index=False).agg(y_reco=("total_reco","sum"))
        agg_top["nivel"] = "FM_COST_TYPE"
        agg_top["clave"] = agg_top["FM_COST_TYPE"].astype(str)
        gaps_rows.append(agg_top[["nivel","clave",DATECOL,"y_reco"]].rename(columns={"y_reco":"valor"}))

    # FM×PAIS
    if not fmp.empty:
        agg_fmp = fmp.rename(columns={"total_reco":"y_reco"})
        agg_fmp["nivel"] = "FM_COST_TYPE|PAIS"
        agg_fmp["clave"] = agg_fmp["FM_COST_TYPE"].astype(str) + "|" + agg_fmp["PAIS"].astype(str)  # <- usar agg_fmp
        gaps_rows.append(agg_fmp[["nivel","clave",DATECOL,"y_reco"]].rename(columns={"y_reco":"valor"}))

    # FM×REGION
    if not fmr.empty:
        agg_fmr = fmr.rename(columns={"total_reco":"y_reco"})
        agg_fmr["nivel"] = "FM_COST_TYPE|REGION"
        agg_fmr["clave"] = agg_fmr["FM_COST_TYPE"].astype(str) + "|" + agg_fmr["REGION"].astype(str)  # <- usar agg_fmr
        gaps_rows.append(agg_fmr[["nivel","clave",DATECOL,"y_reco"]].rename(columns={"y_reco":"valor"}))

    vals = pd.concat(gaps_rows, ignore_index=True) if len(gaps_rows)>0 else pd.DataFrame(columns=["nivel","clave",DATECOL,"valor"])
    gaps = ancla_df.merge(vals, on=["nivel","clave",DATECOL], how="left")
    gaps["gap_reco_vs_target"] = gaps["valor"] - gaps["target"]
else:
    gaps = pd.DataFrame(columns=["nivel","clave",DATECOL,"target","valor","gap_reco_vs_target"])

# Guardamos comparativos estructurales en un único CSV (apilamos con una columna 'scope')
comparativos = []

port2 = port.copy();  port2["scope"] = "PORTFOLIO_MENSUAL"
comparativos.append(port2.rename(columns={"total_base":"base","total_reco":"reconc","delta_portfolio":"delta"}))

fm2 = fm.copy();      fm2["scope"]   = "FM_COST_TYPE_MENSUAL"
fm2 = fm2.rename(columns={"total_base":"base","total_reco":"reconc","delta_fm":"delta"})
comparativos.append(fm2[["scope","FM_COST_TYPE",DATECOL,"base","reconc","delta"]])

if not fmp.empty:
    fmp2 = fmp.copy(); fmp2["scope"] = "FM_COST_TYPE|PAIS_MENSUAL"
    fmp2 = fmp2.rename(columns={"total_base":"base","total_reco":"reconc","delta_fm_pais":"delta"})
    comparativos.append(fmp2[["scope","FM_COST_TYPE","PAIS",DATECOL,"base","reconc","delta"]])

if not fmr.empty:
    fmr2 = fmr.copy(); fmr2["scope"] = "FM_COST_TYPE|REGION_MENSUAL"
    fmr2 = fmr2.rename(columns={"total_base":"base","total_reco":"reconc","delta_fm_region":"delta"})
    comparativos.append(fmr2[["scope","FM_COST_TYPE","REGION",DATECOL,"base","reconc","delta"]])

if not gaps.empty:
    gaps2 = gaps.copy(); gaps2["scope"] = "GAP_RECO_VS_TARGET"
    comparativos.append(gaps2[["scope","nivel","clave",DATECOL,"target","valor","gap_reco_vs_target"]])

compare_out = pd.concat(comparativos, ignore_index=True) if len(comparativos)>0 else pd.DataFrame()
compare_out.to_csv(OUT_COMPARE, sep=CSV_SEP, index=False)
log11(f"Guardado: {OUT_COMPARE} con {len(compare_out)} filas.")

[2025-09-26T06:03:56] Guardado: /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3/REPORTING/compare_estructural_reconc_vs_base_prod.csv con 7572 filas.

# Bloque 9 - Diagnósticos gráficos mínimos (sin verdad)

try:

    # Histograma de deltas bottom
    plt.figure()
    tmp["delta"].plot(kind="hist", bins=40)
    plt.title("Deltas bottom (yhat_reconc - yhat_base)")
    plt.xlabel("Delta"); plt.ylabel("Frecuencia")
    plt.tight_layout(); plt.savefig(os.path.join(FIGS_DIR, "hist_deltas_bottom_step11.png")); plt.close()

    # Serie temporal portfolio base vs reconc
    plt.figure(figsize=(10,5))
    plt.plot(port[DATECOL], port["total_base"], label="base")
    plt.plot(port[DATECOL], port["total_reco"], label="reconc")
    plt.title("Portfolio mensual — base vs reconc (sin verdad)")
    plt.xlabel("Mes"); plt.ylabel("Suma mensual")
    plt.legend()
    plt.tight_layout(); plt.savefig(os.path.join(FIGS_DIR, "line_portfolio_base_reco_step11.png")); plt.close()

    log11(f"Guardadas figuras en {FIGS_DIR}")
except Exception as e:
    log11(f"AVISO: no fue posible generar figuras: {e}")

log11("Paso 11 finalizado (producción).")

[2025-09-26T06:03:58] Guardadas figuras en /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3/REPORTING/figs_step11
[2025-09-26T06:03:58] Paso 11 finalizado (producción).

# ============================================================
# Puente de comparación — Paso 10 vs Paso 11
# Comparamos las reconciliaciones entre sí (sin verdad)
# ============================================================

# -----------------------------
# Bloque A — Parámetros y rutas
# -----------------------------

ruta_base_3 = "/content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3"
CSV_SEP     = ";"

RUTA_RESULTADOS = os.path.join(ruta_base_3, "RESULTADOS")
RUTA_METRICAS   = os.path.join(ruta_base_3, "METRICAS")
RUTA_REPORTING  = os.path.join(ruta_base_3, "REPORTING")

PATH_STEP10 = os.path.join(RUTA_RESULTADOS, "preds_reconciliadas_2024.csv")
PATH_STEP11 = os.path.join(RUTA_RESULTADOS, "preds_reconciliadas_2024_prod.csv")
PATH_DIM    = os.path.join(RUTA_METRICAS,   "dim_buildings.csv")

OUT_BRIDGE  = os.path.join(RUTA_REPORTING,  "compare_step10_vs_step11.csv")

PAIR_COLS = ["ID_BUILDING","FM_COST_TYPE"]
DATECOL   = "FECHA"

# --------------------------------
# Bloque B — Carga y normalización
# --------------------------------
def _ensure_ms(df, datecol):
    # Normalizamos a MS para asegurar alineación
    df = df.copy()
    df[datecol] = pd.to_datetime(df[datecol]).dt.to_period("M").dt.to_timestamp()
    return df

def _norm_pair(df):
    # Normalizamos claves
    out = df.copy()
    out["ID_BUILDING"] = pd.to_numeric(out["ID_BUILDING"], errors="coerce").astype("Int64")
    out["FM_COST_TYPE"] = out["FM_COST_TYPE"].astype(str).str.strip()
    return out

step10 = pd.read_csv(PATH_STEP10, sep=CSV_SEP)
step11 = pd.read_csv(PATH_STEP11, sep=CSV_SEP)
step10[DATECOL] = pd.to_datetime(step10[DATECOL])
step11[DATECOL] = pd.to_datetime(step11[DATECOL])
step10 = _norm_pair(step10); step11 = _norm_pair(step11)

dim = pd.read_csv(PATH_DIM, sep=CSV_SEP) if os.path.exists(PATH_DIM) else pd.DataFrame()
if not dim.empty:
    # Nos quedamos con PAIS y REGION si existen
    keep = [c for c in ["ID_BUILDING","PAIS","REGION"] if c in dim.columns]
    dim = dim[keep].drop_duplicates()

# ------------------------------------------------------------
# Bloque C — Ensamble bottom y deltas entre reconciliaciones
# ------------------------------------------------------------
# Unimos por par y fecha; renombramos para distinguir
st10 = step10.rename(columns={"yhat_reconc":"yhat_reconc_step10"})
st11 = step11.rename(columns={"yhat_reconc":"yhat_reconc_step11"})

cols_merge = [*PAIR_COLS, DATECOL]
merged = st10[cols_merge + ["yhat_reconc_step10"]].merge(
           st11[cols_merge + ["yhat_reconc_step11"]],
           on=cols_merge, how="inner"
         )

# Calculamos delta entre métodos
merged["delta_step11_vs_step10"] = merged["yhat_reconc_step11"] - merged["yhat_reconc_step10"]
merged["abs_delta"] = merged["delta_step11_vs_step10"].abs()

# Adjuntamos dimensiones si están disponibles
if not dim.empty:
    merged = merged.merge(dim, on="ID_BUILDING", how="left")

# ------------------------------------------------------------
# Bloque D — Agregados por nivel para entender diferencias
# ------------------------------------------------------------
def _sum_by(gcols, label):
    # Agregamos por grupo y fecha para comparar totales por método
    g = merged.groupby(gcols + [DATECOL], as_index=False).agg(
        y_step10=("yhat_reconc_step10","sum"),
        y_step11=("yhat_reconc_step11","sum")
    )
    g["delta"] = g["y_step11"] - g["y_step10"]
    g["scope"] = label
    return g

# Portfolio
port = _sum_by([], "PORTFOLIO_MENSUAL")

# FM_COST_TYPE
by_fm = _sum_by(["FM_COST_TYPE"], "FM_COST_TYPE_MENSUAL")

# FM_COST_TYPE×PAIS si existe
if "PAIS" in merged.columns and merged["PAIS"].notna().any():
    by_fm_pais = _sum_by(["FM_COST_TYPE","PAIS"], "FM_COST_TYPE|PAIS_MENSUAL")
else:
    by_fm_pais = pd.DataFrame(columns=["scope","FM_COST_TYPE","PAIS",DATECOL,"y_step10","y_step11","delta"])

# FM_COST_TYPE×REGION si existe
if "REGION" in merged.columns and merged["REGION"].notna().any():
    by_fm_region = _sum_by(["FM_COST_TYPE","REGION"], "FM_COST_TYPE|REGION_MENSUAL")
else:
    by_fm_region = pd.DataFrame(columns=["scope","FM_COST_TYPE","REGION",DATECOL,"y_step10","y_step11","delta"])

# Apilamos salidas
bridge_top = port.copy()
bridge_fm  = by_fm.copy()
frames = [bridge_top, bridge_fm]
if not by_fm_pais.empty:   frames.append(by_fm_pais)
if not by_fm_region.empty: frames.append(by_fm_region)
bridge = pd.concat(frames, ignore_index=True)

# Guardamos resultado maestro y también anexamos bottom con deltas
# Para minimizar fricción, guardamos dos bloques en un mismo CSV
bridge.to_csv(OUT_BRIDGE, sep=CSV_SEP, index=False)
with open(OUT_BRIDGE, "a", encoding="utf-8") as f:
    f.write("\n")  # separador lógico
    merged.to_csv(f, sep=CSV_SEP, index=False, header=True)

print(f"Guardado: {OUT_BRIDGE}")

Guardado: /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3/REPORTING/compare_step10_vs_step11.csv

# ============================================================
# Comparación visual — Paso 10 vs Paso 11 (sin verdad)
# En este notebook comparamos las predicciones reconciliadas
# generadas por el Paso 10 (con verdad) y el Paso 11 (sin verdad)
# ============================================================

# -----------------------------
# Bloque A — Parámetros y rutas
# -----------------------------

# Definimos rutas base del proyecto
ruta_base_3 = "/content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3"
CSV_SEP     = ";"

# Carpetas estándar
RUTA_RESULTADOS = os.path.join(ruta_base_3, "RESULTADOS")
RUTA_METRICAS   = os.path.join(ruta_base_3, "METRICAS")
RUTA_REPORTING  = os.path.join(ruta_base_3, "REPORTING")
FIGS_DIR        = os.path.join(RUTA_REPORTING, "figs_step10_vs_step11")
os.makedirs(FIGS_DIR, exist_ok=True)

# Entradas necesarias
PATH_STEP10 = os.path.join(RUTA_RESULTADOS, "preds_reconciliadas_2024.csv")       # Paso 10
PATH_STEP11 = os.path.join(RUTA_RESULTADOS, "preds_reconciliadas_2024_prod.csv")  # Paso 11
PATH_DIM    = os.path.join(RUTA_METRICAS,   "dim_buildings.csv")                   # Opcional

# Constantes de clave
PAIR_COLS = ["ID_BUILDING","FM_COST_TYPE"]
DATECOL   = "FECHA"

# --------------------------------
# Bloque B — Funciones utilitarias
# --------------------------------
def _ensure_ms(df: pd.DataFrame, datecol: str) -> pd.DataFrame:
    # Normalizamos las fechas a comienzo de mes para alinear a malla MS
    out = df.copy()
    out[datecol] = pd.to_datetime(out[datecol]).dt.to_period("M").dt.to_timestamp()
    return out

def _norm_pair(df: pd.DataFrame) -> pd.DataFrame:
    # Normalizamos claves y tipamos
    out = df.copy()
    if "ID_BUILDING" in out.columns:
        out["ID_BUILDING"] = pd.to_numeric(out["ID_BUILDING"], errors="coerce").astype("Int64")
    if "FM_COST_TYPE" in out.columns:
        out["FM_COST_TYPE"] = out["FM_COST_TYPE"].astype(str).str.strip()
    return out

def _load_csv(path: str, sep: str = CSV_SEP, required: bool = True) -> pd.DataFrame:
    # Cargamos un CSV con validación opcional
    if os.path.exists(path):
        df = pd.read_csv(path, sep=sep)
        df.columns = [str(c).strip().lstrip("\ufeff") for c in df.columns]
        return df
    if required:
        raise FileNotFoundError(f"Falta el archivo requerido: {path}")
    print(f"[AVISO] No encontramos {os.path.basename(path)} (continuamos sin este artefacto).")
    return pd.DataFrame()

def _sum_by(df: pd.DataFrame, gcols: list[str], datecol: str, c10: str, c11: str, label: str) -> pd.DataFrame:
    # Agregamos por grupo y fecha para comparar totales por método
    g = df.groupby(gcols + [datecol], as_index=False).agg(
        y_step10=(c10,"sum"),
        y_step11=(c11,"sum")
    )
    g["delta"] = g["y_step11"] - g["y_step10"]
    g["scope"] = label
    return g

def _annual_total(df: pd.DataFrame, key_cols: list[str], c10: str, c11: str, label: str) -> pd.DataFrame:
    # Agregamos anual total por clave para comparar magnitudes
    g = df.groupby(key_cols, as_index=False).agg(
        y_step10=(c10,"sum"),
        y_step11=(c11,"sum")
    )
    g["delta"] = g["y_step11"] - g["y_step10"]
    g["scope"] = label
    return g

# ---------------------------------------
# Bloque C — Carga, joins y transformados
# ---------------------------------------
# Cargamos reconciliaciones
step10 = _load_csv(PATH_STEP10)
step11 = _load_csv(PATH_STEP11)

# Normalizamos columnas de fecha y clave
for df in (step10, step11):
    if DATECOL in df.columns:
        df[DATECOL] = pd.to_datetime(df[DATECOL], errors="coerce")
    df[:] = df  # noop para dejar claro que mantenemos esquema

step10 = _norm_pair(step10)
step11 = _norm_pair(step11)

# Renombramos columnas para distinguir métodos
if "yhat_reconc" not in step10.columns or "yhat_reconc" not in step11.columns:
    raise KeyError("Esperábamos columna 'yhat_reconc' en ambas entradas de reconciliación.")

st10 = step10.rename(columns={"yhat_reconc":"yhat_reconc_step10"})
st11 = step11.rename(columns={"yhat_reconc":"yhat_reconc_step11"})

# Merge de bottom por par y fecha
cols_merge = [*PAIR_COLS, DATECOL]
merged = (st10[cols_merge + ["yhat_reconc_step10"]]
          .merge(st11[cols_merge + ["yhat_reconc_step11"]],
                 on=cols_merge, how="inner"))

# Calculamos delta entre métodos a nivel bottom
merged["delta_step11_vs_step10"] = merged["yhat_reconc_step11"] - merged["yhat_reconc_step10"]
merged["abs_delta"] = merged["delta_step11_vs_step10"].abs()

# Dimensiones opcionales (PAIS/REGION)
dim = _load_csv(PATH_DIM, required=False)
if not dim.empty:
    keep = [c for c in ["ID_BUILDING","PAIS","REGION"] if c in dim.columns]
    if keep:
        dim = dim[keep].drop_duplicates()
        merged = merged.merge(dim, on="ID_BUILDING", how="left")

# Agregados por nivel y fecha
port = _sum_by(merged, [], DATECOL, "yhat_reconc_step10", "yhat_reconc_step11", "PORTFOLIO_MENSUAL")
by_fm = _sum_by(merged, ["FM_COST_TYPE"], DATECOL, "yhat_reconc_step10", "yhat_reconc_step11", "FM_COST_TYPE_MENSUAL")
if "PAIS" in merged.columns and merged["PAIS"].notna().any():
    by_fm_pais = _sum_by(merged, ["FM_COST_TYPE","PAIS"], DATECOL, "yhat_reconc_step10", "yhat_reconc_step11", "FM_COST_TYPE|PAIS_MENSUAL")
else:
    by_fm_pais = pd.DataFrame(columns=["scope","FM_COST_TYPE","PAIS",DATECOL,"y_step10","y_step11","delta"])
if "REGION" in merged.columns and merged["REGION"].notna().any():
    by_fm_region = _sum_by(merged, ["FM_COST_TYPE","REGION"], DATECOL, "yhat_reconc_step10", "yhat_reconc_step11", "FM_COST_TYPE|REGION_MENSUAL")
else:
    by_fm_region = pd.DataFrame(columns=["scope","FM_COST_TYPE","REGION",DATECOL,"y_step10","y_step11","delta"])

# Totales anuales por FM_COST_TYPE y por capa territorial si procede
annual_fm       = _annual_total(merged, ["FM_COST_TYPE"], "yhat_reconc_step10", "yhat_reconc_step11", "FM_COST_TYPE_ANUAL")
annual_fm_pais  = _annual_total(merged, ["FM_COST_TYPE","PAIS"], "yhat_reconc_step10", "yhat_reconc_step11", "FM_COST_TYPE|PAIS_ANUAL")   if "PAIS"   in merged.columns else pd.DataFrame()
annual_fm_region= _annual_total(merged, ["FM_COST_TYPE","REGION"], "yhat_reconc_step10", "yhat_reconc_step11", "FM_COST_TYPE|REGION_ANUAL") if "REGION" in merged.columns else pd.DataFrame()

# ------------------------------------------------
# Bloque D — Visualizaciones (cada plot independiente)
# ------------------------------------------------

# 1) Portfolio mensual: líneas base vs prod y delta
plt.figure(figsize=(10,5))
plt.plot(port[DATECOL], port["y_step10"], label="Paso 10 — reconc")
plt.plot(port[DATECOL], port["y_step11"], label="Paso 11 — reconc")
plt.title("Portfolio mensual — Paso 10 vs Paso 11")
plt.xlabel("Mes"); plt.ylabel("Suma mensual")
plt.legend()
p1 = os.path.join(FIGS_DIR, "line_portfolio_step10_vs_step11.png")
plt.tight_layout(); plt.savefig(p1); plt.show()

plt.figure(figsize=(10,4))
plt.plot(port[DATECOL], port["delta"])
plt.axhline(0)
plt.title("Portfolio mensual — Delta (Paso 11 - Paso 10)")
plt.xlabel("Mes"); plt.ylabel("Delta")
p2 = os.path.join(FIGS_DIR, "line_portfolio_delta.png")
plt.tight_layout(); plt.savefig(p2); plt.show()

# 2) Barras anuales por FM_COST_TYPE: Paso 10 vs Paso 11
by_fm_year = annual_fm.sort_values("y_step10", ascending=False).reset_index(drop=True)
x = np.arange(len(by_fm_year))
w = 0.35
plt.figure(figsize=(12,5))
plt.bar(x - w/2, by_fm_year["y_step10"], width=w, label="Paso 10 — anual")
plt.bar(x + w/2, by_fm_year["y_step11"], width=w, label="Paso 11 — anual")
plt.xticks(x, by_fm_year["FM_COST_TYPE"], rotation=45, ha="right")
plt.title("Total anual por FM_COST_TYPE — Paso 10 vs Paso 11")
plt.ylabel("Suma anual")
plt.legend()
p3 = os.path.join(FIGS_DIR, "bar_anual_fmcost_step10_vs_step11.png")
plt.tight_layout(); plt.savefig(p3); plt.show()

# 3) Boxplot de deltas a nivel bottom por FM_COST_TYPE
plt.figure(figsize=(10,5))
cats = sorted(merged["FM_COST_TYPE"].dropna().unique())
data_box = [merged.loc[merged["FM_COST_TYPE"]==k, "delta_step11_vs_step10"].values for k in cats]
if len(cats) > 0:
    plt.boxplot(data_box, labels=cats, showfliers=False)
    plt.title("Distribución de deltas bottom por FM_COST_TYPE (Paso 11 - Paso 10)")
    plt.xlabel("FM_COST_TYPE"); plt.ylabel("Delta")
    plt.xticks(rotation=45, ha="right")
    p4 = os.path.join(FIGS_DIR, "box_deltas_bottom_por_fmcost.png")
    plt.tight_layout(); plt.savefig(p4); plt.show()
else:
    plt.text(0.5, 0.5, "No hay FM_COST_TYPE disponibles para boxplot", ha="center")
    plt.show()

# 4) Top 20 ajustes absolutos bottom (barras horizontales)
tops = (merged.assign(abs_delta=merged["abs_delta"].astype(float))
        .sort_values("abs_delta", ascending=False)
        .head(20))
if not tops.empty:
    if "PAIS" in tops.columns:
        labs = tops.apply(lambda r: f"{r['ID_BUILDING']}|{r['FM_COST_TYPE']}|{str(r.get('PAIS',''))}", axis=1)
    else:
        labs = tops.apply(lambda r: f"{r['ID_BUILDING']}|{r['FM_COST_TYPE']}", axis=1)
    plt.figure(figsize=(10,8))
    plt.barh(range(len(tops)), tops["delta_step11_vs_step10"])
    plt.yticks(range(len(tops)), labs)
    plt.gca().invert_yaxis()
    plt.title("Top 20 ajustes absolutos en bottom (Paso 11 - Paso 10)")
    plt.xlabel("Delta")
    p5 = os.path.join(FIGS_DIR, "barh_top20_deltas_abs.png")
    plt.tight_layout(); plt.savefig(p5); plt.show()

# 5) Heatmap simple FM_COST_TYPE × mes de deltas agregados (si hay variedad)
pivot_fm = by_fm.pivot(index="FM_COST_TYPE", columns=DATECOL, values="delta") if not by_fm.empty else pd.DataFrame()
if not pivot_fm.empty:
    plt.figure(figsize=(12,6))
    plt.imshow(pivot_fm.values, aspect="auto")
    plt.colorbar()
    plt.yticks(range(pivot_fm.shape[0]), pivot_fm.index)
    plt.xticks(range(pivot_fm.shape[1]), [d.strftime("%Y-%m") for d in pivot_fm.columns], rotation=45, ha="right")
    plt.title("Mapa de calor — Delta mensual por FM_COST_TYPE (Paso 11 - Paso 10)")
    p6 = os.path.join(FIGS_DIR, "heatmap_delta_fmcost_mes.png")
    plt.tight_layout(); plt.savefig(p6); plt.show()

# 6) Scatter mensual portfolio: y_step10 vs y_step11 con línea y=x
if not port.empty:
    plt.figure()
    plt.scatter(port["y_step10"], port["y_step11"])
    minv = float(np.nanmin([port["y_step10"].min(), port["y_step11"].min()]))
    maxv = float(np.nanmax([port["y_step10"].max(), port["y_step11"].max()]))
    plt.plot([minv, maxv], [minv, maxv])
    plt.title("Portfolio mensual — y(P10) vs y(P11)")
    plt.xlabel("Paso 10"); plt.ylabel("Paso 11")
    p7 = os.path.join(FIGS_DIR, "scatter_portfolio_p10_vs_p11.png")
    plt.tight_layout(); plt.savefig(p7); plt.show()

print("Listo. Guardamos las figuras en:", FIGS_DIR)

Listo. Guardamos las figuras en: /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3/REPORTING/figs_step10_vs_step11

# ============================================================
# Verificación de base Paso 10 vs Paso 11 y comparación base P10 vs reconc P11
# ============================================================

# -----------------------------
# Bloque A — Parámetros y rutas
# -----------------------------


ruta_base_3 = "/content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3"
CSV_SEP     = ";"

RUTA_RESULTADOS = os.path.join(ruta_base_3, "RESULTADOS")
RUTA_REPORTING  = os.path.join(ruta_base_3, "REPORTING")
os.makedirs(RUTA_REPORTING, exist_ok=True)

PATH_STEP10_RECONC = os.path.join(RUTA_RESULTADOS, "preds_reconciliadas_2024.csv")        # generado en Paso 10
PATH_STEP11_RECONC = os.path.join(RUTA_RESULTADOS, "preds_reconciliadas_2024_prod.csv")   # generado en Paso 11

PAIR_COLS = ["ID_BUILDING","FM_COST_TYPE"]
DATECOL   = "FECHA"

# --------------------------------
# Bloque B — Funciones utilitarias
# --------------------------------
def _load(path: str, required: bool=True) -> pd.DataFrame:
    # Cargamos CSV con limpieza mínima de columnas
    if os.path.exists(path):
        df = pd.read_csv(path, sep=CSV_SEP)
        df.columns = [str(c).strip().lstrip("\ufeff") for c in df.columns]
        if DATECOL in df.columns:
            df[DATECOL] = pd.to_datetime(df[DATECOL], errors="coerce")
        if "ID_BUILDING" in df.columns:
            df["ID_BUILDING"] = pd.to_numeric(df["ID_BUILDING"], errors="coerce").astype("Int64")
        if "FM_COST_TYPE" in df.columns:
            df["FM_COST_TYPE"] = df["FM_COST_TYPE"].astype(str).str.strip()
        return df
    if required:
        raise FileNotFoundError(f"Falta archivo: {path}")
    return pd.DataFrame()

# ---------------------------------------
# Bloque C — Carga y alineación de bases
# ---------------------------------------
p10 = _load(PATH_STEP10_RECONC, required=True)
p11 = _load(PATH_STEP11_RECONC, required=True)

# Verificamos que tengamos columnas esperadas
need10 = set(PAIR_COLS + [DATECOL, "yhat_base", "yhat_reconc"])
need11 = set(PAIR_COLS + [DATECOL, "yhat_base", "yhat_reconc"])
if not need10.issubset(set(p10.columns)):
    raise KeyError(f"Paso 10 sin columnas requeridas: faltan {list(need10 - set(p10.columns))}")
if not need11.issubset(set(p11.columns)):
    raise KeyError(f"Paso 11 sin columnas requeridas: faltan {list(need11 - set(p11.columns))}")

# Alineamos por clave y fecha; renombramos para distinguir
m = (p10[PAIR_COLS + [DATECOL, "yhat_base", "yhat_reconc"]]
     .rename(columns={"yhat_base":"yhat_base_p10", "yhat_reconc":"yhat_reconc_p10"})
     .merge(
         p11[PAIR_COLS + [DATECOL, "yhat_base", "yhat_reconc"]]
         .rename(columns={"yhat_base":"yhat_base_p11", "yhat_reconc":"yhat_reconc_p11"}),
         on=PAIR_COLS + [DATECOL], how="inner"
     ))

# ---------------------------------------------
# Bloque D — Chequeo de igualdad de la “base”
# ---------------------------------------------
# Comprobamos si la base es exactamente igual (mismo universo y valores)
if m.empty:
    raise ValueError("No hay intersección entre Paso 10 y Paso 11 por (ID_BUILDING, FM_COST_TYPE, FECHA).")

m["diff_base"] = (m["yhat_base_p11"] - m["yhat_base_p10"]).astype(float)
n_total = len(m)
n_iguales = int((m["diff_base"].abs() < 1e-9).sum())  # tolerancia numérica pequeña
n_distintos = n_total - n_iguales

print(f"[Base] Observaciones comparadas: {n_total}")
print(f"[Base] Iguales (|Δ|<1e-9): {n_iguales}  |  Distintas: {n_distintos}")

if n_distintos > 0:
    print("[Base] Muestra de diferencias (5 filas):")
    display(m.loc[m["diff_base"].abs()>=1e-9, PAIR_COLS + [DATECOL, "yhat_base_p10", "yhat_base_p11", "diff_base"]]
            .sort_values("diff_base").head(5))

# -------------------------------------------------------
# Bloque E — Comparación base(P10) vs reconciliación(P11)
# -------------------------------------------------------
# Calculamos delta entre la base del P10 y la reconciliación del P11
m["delta_baseP10_vs_recoP11"] = (m["yhat_reconc_p11"] - m["yhat_base_p10"]).astype(float)
m["abs_delta_baseP10_vs_recoP11"] = m["delta_baseP10_vs_recoP11"].abs()

# Agregamos por portfolio mensual
portfolio = (m.groupby([DATECOL], as_index=False)
               .agg(y_base_p10=("yhat_base_p10","sum"),
                    y_reco_p11=("yhat_reconc_p11","sum")))
portfolio["delta"] = portfolio["y_reco_p11"] - portfolio["y_base_p10"]

print("\n[Comparación] Portfolio mensual — primeras filas:")
display(portfolio.head())

# Top 20 ajustes absolutos por serie-mes entre base P10 vs reco P11
top20 = (m.sort_values("abs_delta_baseP10_vs_recoP11", ascending=False)
           .head(20)[PAIR_COLS + [DATECOL, "yhat_base_p10", "yhat_reconc_p11", "delta_baseP10_vs_recoP11"]])
print("\n[Comparación] Top 20 ajustes absolutos base(P10) vs reconc(P11):")
display(top20)

[Base] Observaciones comparadas: 29148
[Base] Iguales (|Δ|<1e-9): 29148  |  Distintas: 0

[Comparación] Portfolio mensual — primeras filas:

[Comparación] Top 20 ajustes absolutos base(P10) vs reconc(P11):

# ============================================================
# Extracción de países únicos desde dim_buildings.csv
# ============================================================

# -------------------------------
# Bloque 0) Parámetros y rutas
# -------------------------------


# Definimos las rutas coherentes con el proyecto
ruta_base_3 = "/content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3"
CSV_SEP     = ";"

RUTA_METRICAS = os.path.join(ruta_base_3, "METRICAS")
os.makedirs(RUTA_METRICAS, exist_ok=True)

PATH_DIM   = os.path.join(RUTA_METRICAS, "dim_buildings.csv")
PATH_OUT   = os.path.join(RUTA_METRICAS, "paises_unicos_detectados.csv")


# ------------------------------------------
# Bloque 1) Helpers (normalización simple)
# ------------------------------------------
def _norm_str(s):
    """Normalizamos strings: convertimos a str, recortamos y vacíos a NaN."""
    if s is None or (isinstance(s, float) and np.isnan(s)):
        return np.nan
    s = str(s).strip()
    return np.nan if s == "" else s


# -------------------------------
# Bloque 2) Carga y extracción
# -------------------------------
# Leemos dim_buildings y localizamos la columna de país.
if not os.path.exists(PATH_DIM):
    raise FileNotFoundError(f"No encontramos {PATH_DIM}. Asegurémonos de ejecutar el Paso 9 antes.")

dim = pd.read_csv(PATH_DIM, sep=CSV_SEP)

# Intentamos detectar la columna de país con nombres comunes
cols_up = {c.upper().strip(): c for c in dim.columns}
if "PAIS" in cols_up:
    col_pais = cols_up["PAIS"]
elif "COUNTRY" in cols_up:
    col_pais = cols_up["COUNTRY"]
elif "COUNTRY_DEF" in cols_up:
    col_pais = cols_up["COUNTRY_DEF"]
else:
    raise KeyError("No encontramos columna de país (PAIS/COUNTRY/COUNTRY_DEF) en dim_buildings.csv")

# Obtenemos países únicos normalizados
paises = (
    dim[[col_pais]]
    .rename(columns={col_pais: "PAIS"})
    .assign(PAIS=lambda d: d["PAIS"].map(_norm_str))
    .dropna(subset=["PAIS"])
    .drop_duplicates()
    .sort_values("PAIS")
    .reset_index(drop=True)
)

# Guardamos y mostramos
paises.to_csv(PATH_OUT, sep=CSV_SEP, index=False)

print(f"[Info] Países únicos detectados: {len(paises)}")
print("[Info] Los listamos uno por línea para que nos los puedas pegar tal cual:")
for p in paises["PAIS"].tolist():
    print(p)

print(f"\n[Info] También los guardamos en: {PATH_OUT}")

[Info] Países únicos detectados: 13
[Info] Los listamos uno por línea para que nos los puedas pegar tal cual:
COLOMBIA
COSTA RICA
DESCONOCIDO
ESPAÑA
ITALIA
MALTA
MARRUECOS
MÉXICO
PANAMÁ
PERÚ
PUERTO RICO
REPÚBLICA DOMINICANA
VENEZUELA

[Info] También los guardamos en: /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3/METRICAS/paises_unicos_detectados.csv

# ============================================================
# Construcción de ipc_por_pais_fechaRef.csv
# ============================================================
# - Obtenemos países únicos desde METRICAS/dim_buildings.csv (columna PAIS)
# - Deducimos fecha_ref = mes anterior al último FECHA del TRAIN
# - Consultamos IMF SDMX JSON (dataset PCPI_IX, frecuencia mensual)
# - Calculamos IPC interanual (YoY %) en fecha_ref por país
# - Guardamos METRICAS/ipc_por_pais_fechaRef.csv
# ============================================================

# -----------------------------
# Bloque 0) Imports y constantes
# -----------------------------

# Rutas base del proyecto (ajustamos a tu estructura)
ruta_base_3 = "/content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3"
CSV_SEP     = ";"

RUTA_RESULTADOS = os.path.join(ruta_base_3, "RESULTADOS")
RUTA_METRICAS   = os.path.join(ruta_base_3, "METRICAS")
RUTA_LOGS       = os.path.join(ruta_base_3, "LOGS")
os.makedirs(RUTA_RESULTADOS, exist_ok=True)
os.makedirs(RUTA_METRICAS,   exist_ok=True)
os.makedirs(RUTA_LOGS,       exist_ok=True)

PATH_TRAIN = os.path.join(RUTA_RESULTADOS, "train_full_2021_2023.csv")  # o el TRAIN vigente
PATH_DIM   = os.path.join(RUTA_METRICAS,   "dim_buildings.csv")
OUT_IPC    = os.path.join(RUTA_METRICAS,   "ipc_por_pais_fechaRef.csv")
LOG_PATH   = os.path.join(RUTA_LOGS,       "ipc_fetch.log")

# IMF SDMX (PCPI_IX mensual). Usamos CompactData para respuesta más simple.
IMF_BASE = "https://dataservices.imf.org/REST/SDMX_JSON.svc/CompactData"
IMF_DATASET = "PCPI_IX"   # Índice de Precios al Consumidor (2010=100 ó base IMF), mensual
# CompactData estructura: /CompactData/PCPI_IX/M.<REF_AREA>.PCPI_IX?startPeriod=YYYY-MM&endPeriod=YYYY-MM

# Mapeo de nombres de país (en dim_buildings) → REF_AREA IMF (ISO2)
# Afinamos con códigos ISO2 estándar, que el IMF suele usar como REF_AREA
PAIS_TO_IMF_REF = {
    "COLOMBIA": "CO",
    "COSTA RICA": "CR",
    "ESPAÑA": "ES",
    "ITALIA": "IT",
    "MALTA": "MT",
    "MARRUECOS": "MA",
    "MÉXICO": "MX",
    "MEXICO": "MX",
    "PANAMÁ": "PA",
    "PANAMA": "PA",
    "PERÚ": "PE",
    "PERU": "PE",
    "PUERTO RICO": "PR",
    "REPÚBLICA DOMINICANA": "DO",
    "REPUBLICA DOMINICANA": "DO",
    "VENEZUELA": "VE",
    # “DESCONOCIDO” lo omitimos explícitamente más abajo
}

# -----------------------------
# Bloque 1) Logging y utilidades
# -----------------------------
def log_ipc(msg: str):
    ts = datetime.now().isoformat(timespec="seconds")
    line = f"[{ts}] {msg}"
    print(line)
    with open(LOG_PATH, "a", encoding="utf-8") as f:
        f.write(line + "\n")

def _ensure_ms(df: pd.DataFrame, datecol: str) -> pd.DataFrame:
    # Normalizamos FECHA al primer día de mes para evitar desalineaciones
    df = df.copy()
    df[datecol] = pd.to_datetime(df[datecol]).dt.to_period("M").dt.to_timestamp()
    return df

def _read_csv(path: str, required: bool=True, sep: str=CSV_SEP) -> pd.DataFrame:
    if os.path.exists(path):
        df = pd.read_csv(path, sep=sep)
        df.columns = [str(c).strip().lstrip("\ufeff") for c in df.columns]
        return df
    if required:
        raise FileNotFoundError(f"Falta el archivo requerido: {path}")
    log_ipc(f"AVISO: no encontramos {os.path.basename(path)} (continuamos sin este artefacto).")
    return pd.DataFrame()

def _prev_month(ts: pd.Timestamp) -> pd.Timestamp:
    # Restamos un mes respetando frecuencia mensual
    ts = pd.Timestamp(ts).to_period("M").to_timestamp()
    return (ts - pd.offsets.MonthBegin(1))

def _as_ym(ts: pd.Timestamp) -> str:
    # Devolvemos YYYY-MM
    return pd.Timestamp(ts).strftime("%Y-%m")

# -----------------------------
# Bloque 2) Países únicos desde dim_buildings y fecha_ref desde TRAIN
# -----------------------------
def _get_unique_paises_from_dim(path_dim: str) -> List[str]:
    dim = _read_csv(path_dim, required=True)
    # Normalizamos nombre de la columna y de valores
    cols_upper = {c.upper(): c for c in dim.columns}
    if "PAIS" not in cols_upper:
        raise KeyError("No encontramos columna 'PAIS' en dim_buildings.csv")
    pais_col = cols_upper["PAIS"]
    paises = (dim[pais_col]
              .astype(str)
              .str.strip()
              .str.upper()
              .replace({"": np.nan, "NAN": np.nan}))
    # Omitimos explícitamente DESCONOCIDO
    out = sorted([p for p in paises.dropna().unique().tolist() if p != "DESCONOCIDO"])
    return out

def _get_fecha_ref_from_train(path_train: str, datecol: str="FECHA") -> pd.Timestamp:
    train = _read_csv(path_train, required=True)
    if datecol not in train.columns:
        # Intentamos alternativas comunes
        cand = [c for c in train.columns if c.strip().upper() in ("FECHA","DATE","PERIOD","DS")]
        if not cand:
            raise KeyError(f"No encontramos columna de fecha en TRAIN (esperábamos '{datecol}')")
        datecol = cand[0]
    train = _ensure_ms(train, datecol)
    last_ts = pd.to_datetime(train[datecol]).max()
    # Usamos el mes anterior al último del TRAIN, como definimos en el requerimiento
    fecha_ref = _prev_month(last_ts)
    return fecha_ref

# -----------------------------
# Bloque 3) Consulta IMF SDMX (PCPI_IX) y cálculo YoY
# -----------------------------
def _imf_compact_series(ref_area: str, start_ym: str, end_ym: str) -> Optional[Dict[str, Any]]:
    # Construimos endpoint: CompactData/PCPI_IX/M.<REF_AREA>.PCPI_IX?startPeriod=YYYY-MM&endPeriod=YYYY-MM
    url = f"{IMF_BASE}/{IMF_DATASET}/M.{ref_area}.PCPI_IX?startPeriod={start_ym}&endPeriod={end_ym}"
    try:
        r = requests.get(url, timeout=30)
        r.raise_for_status()
        return r.json()
    except Exception as e:
        log_ipc(f"ERROR IMF fetch {ref_area}: {e}")
        return None

def _parse_compact_to_ts(json_obj: Dict[str, Any]) -> pd.DataFrame:
    # Extraemos pares (time,value) desde CompactData → DataSet → Series → Obs
    # Devolvemos DataFrame con columnas: DATE(YYYY-MM-01), VALUE(float)
    try:
        ds = json_obj.get("CompactData", {}).get("DataSet", {})
        series = ds.get("Series")
        if series is None:
            return pd.DataFrame(columns=["DATE","VALUE"])
        # En CompactData, Series puede ser dict o list; normalizamos a lista
        if isinstance(series, dict):
            series = [series]
        rows = []
        for s in series:
            obs = s.get("Obs", [])
            if isinstance(obs, dict):
                obs = [obs]
            for o in obs:
                t = o.get("@TIME_PERIOD")
                v = o.get("@OBS_VALUE")
                if t is None or v is None:
                    continue
                # Convertimos YYYY-MM a timestamp MS
                try:
                    dt = pd.to_datetime(f"{t}-01")
                    val = float(v)
                    rows.append((dt, val))
                except Exception:
                    continue
        if not rows:
            return pd.DataFrame(columns=["DATE","VALUE"])
        df = pd.DataFrame(rows, columns=["DATE","VALUE"]).sort_values("DATE")
        return df
    except Exception:
        return pd.DataFrame(columns=["DATE","VALUE"])

def _compute_yoy_at(df: pd.DataFrame, fecha_ref: pd.Timestamp) -> Tuple[Optional[float], Optional[float], Optional[float]]:
    # Calculamos valor en fecha_ref y en fecha_ref - 12m; devolvemos (val_ref, val_lag12, yoy_pct)
    if df.empty:
        return (None, None, None)
    df = df.copy()
    df["DATE"] = pd.to_datetime(df["DATE"]).dt.to_period("M").dt.to_timestamp()
    ref = fecha_ref.to_period("M").to_timestamp()
    lag = (ref - pd.offsets.DateOffset(years=1)).to_period("M").to_timestamp()
    v_ref = df.loc[df["DATE"].eq(ref), "VALUE"]
    v_lag = df.loc[df["DATE"].eq(lag), "VALUE"]
    if v_ref.empty or v_lag.empty:
        return (float(v_ref.iloc[0]) if not v_ref.empty else None,
                float(v_lag.iloc[0]) if not v_lag.empty else None,
                None)
    a = float(v_ref.iloc[0]); b = float(v_lag.iloc[0])
    if b == 0 or not math.isfinite(a) or not math.isfinite(b):
        return (a, b, None)
    yoy = 100.0 * (a / b - 1.0)
    return (a, b, yoy)

# -----------------------------
# Bloque 4) Flujo principal
# -----------------------------
def main_build_ipc_csv_no_pandasdmx(
    path_dim: str = PATH_DIM,
    path_train: str = PATH_TRAIN,
    out_csv: str = OUT_IPC
):
    # 1) Países únicos desde dim_buildings
    paises = _get_unique_paises_from_dim(path_dim)
    log_ipc(f"Paises únicos en dim_buildings (sin 'DESCONOCIDO'): {paises}")

    # 2) Fecha de referencia desde TRAIN (mes anterior al último)
    fecha_ref = _get_fecha_ref_from_train(path_train, datecol="FECHA")
    ref_ym = _as_ym(fecha_ref)
    # Para robustez, pedimos un rango amplio que cubra al menos ref y ref-12
    start_ym = _as_ym((fecha_ref - pd.offsets.DateOffset(years=1)) - pd.offsets.MonthBegin(1))
    end_ym   = ref_ym
    log_ipc(f"Fecha de referencia para IPC (YoY): {ref_ym} (lag de 12 meses incluido desde {start_ym})")

    # 3) Iteramos países → IMF → YoY
    rows = []
    for pais in paises:
        pais_norm = pais.strip().upper()
        ref_area = PAIS_TO_IMF_REF.get(pais_norm)
        if ref_area is None:
            rows.append({
                "PAIS": pais_norm,
                "REF_AREA_IMF": None,
                "FECHA_REF": ref_ym,
                "CPI_VALUE": None,
                "CPI_VALUE_LAG12": None,
                "IPC_YOY_PCT": None,
                "STATUS": "SIN_MAPEO_REF_AREA"
            })
            continue

        js = _imf_compact_series(ref_area, start_ym, end_ym)
        if js is None:
            rows.append({
                "PAIS": pais_norm,
                "REF_AREA_IMF": ref_area,
                "FECHA_REF": ref_ym,
                "CPI_VALUE": None,
                "CPI_VALUE_LAG12": None,
                "IPC_YOY_PCT": None,
                "STATUS": "ERROR_FETCH"
            })
            continue

        df = _parse_compact_to_ts(js)
        val_ref, val_lag, yoy = _compute_yoy_at(df, fecha_ref)
        status = "OK" if yoy is not None else ("FALTAN_MESES" if (val_ref is None or val_lag is None) else "DIV_ZERO_O_NAN")
        rows.append({
            "PAIS": pais_norm,
            "REF_AREA_IMF": ref_area,
            "FECHA_REF": ref_ym,
            "CPI_VALUE": val_ref,
            "CPI_VALUE_LAG12": val_lag,
            "IPC_YOY_PCT": yoy,
            "STATUS": status
        })
        log_ipc(f"{pais_norm}: {status} | YoY={yoy}")

    # 4) Guardamos CSV
    out = pd.DataFrame(rows, columns=["PAIS","REF_AREA_IMF","FECHA_REF","CPI_VALUE","CPI_VALUE_LAG12","IPC_YOY_PCT","STATUS"])
    out.to_csv(out_csv, sep=CSV_SEP, index=False)
    log_ipc(f"Guardado: {out_csv} con {len(out)} filas.")

# -----------------------------
# Bloque 5) Ejecutamos
# -----------------------------
main_build_ipc_csv_no_pandasdmx()

[2025-09-26T06:04:38] Paises únicos en dim_buildings (sin 'DESCONOCIDO'): ['COLOMBIA', 'COSTA RICA', 'ESPAÑA', 'ITALIA', 'MALTA', 'MARRUECOS', 'MÉXICO', 'PANAMÁ', 'PERÚ', 'PUERTO RICO', 'REPÚBLICA DOMINICANA', 'VENEZUELA']
[2025-09-26T06:04:38] Fecha de referencia para IPC (YoY): 2023-11 (lag de 12 meses incluido desde 2022-10)
[2025-09-26T06:05:08] ERROR IMF fetch CO: HTTPSConnectionPool(host='dataservices.imf.org', port=443): Max retries exceeded with url: /REST/SDMX_JSON.svc/CompactData/PCPI_IX/M.CO.PCPI_IX?startPeriod=2022-10&endPeriod=2023-11 (Caused by ConnectTimeoutError(<urllib3.connection.HTTPSConnection object at 0x7eae7ed156d0>, 'Connection to dataservices.imf.org timed out. (connect timeout=30)'))
[2025-09-26T06:05:38] ERROR IMF fetch CR: HTTPSConnectionPool(host='dataservices.imf.org', port=443): Max retries exceeded with url: /REST/SDMX_JSON.svc/CompactData/PCPI_IX/M.CR.PCPI_IX?startPeriod=2022-10&endPeriod=2023-11 (Caused by ConnectTimeoutError(<urllib3.connection.HTTPSConnection object at 0x7eae81ec70e0>, 'Connection to dataservices.imf.org timed out. (connect timeout=30)'))
[2025-09-26T06:06:08] ERROR IMF fetch ES: HTTPSConnectionPool(host='dataservices.imf.org', port=443): Max retries exceeded with url: /REST/SDMX_JSON.svc/CompactData/PCPI_IX/M.ES.PCPI_IX?startPeriod=2022-10&endPeriod=2023-11 (Caused by ConnectTimeoutError(<urllib3.connection.HTTPSConnection object at 0x7eae82e465a0>, 'Connection to dataservices.imf.org timed out. (connect timeout=30)'))
[2025-09-26T06:06:38] ERROR IMF fetch IT: HTTPSConnectionPool(host='dataservices.imf.org', port=443): Max retries exceeded with url: /REST/SDMX_JSON.svc/CompactData/PCPI_IX/M.IT.PCPI_IX?startPeriod=2022-10&endPeriod=2023-11 (Caused by ConnectTimeoutError(<urllib3.connection.HTTPSConnection object at 0x7eae81dae060>, 'Connection to dataservices.imf.org timed out. (connect timeout=30)'))
[2025-09-26T06:07:08] ERROR IMF fetch MT: HTTPSConnectionPool(host='dataservices.imf.org', port=443): Max retries exceeded with url: /REST/SDMX_JSON.svc/CompactData/PCPI_IX/M.MT.PCPI_IX?startPeriod=2022-10&endPeriod=2023-11 (Caused by ConnectTimeoutError(<urllib3.connection.HTTPSConnection object at 0x7eae7eeb7e90>, 'Connection to dataservices.imf.org timed out. (connect timeout=30)'))
[2025-09-26T06:07:38] ERROR IMF fetch MA: HTTPSConnectionPool(host='dataservices.imf.org', port=443): Max retries exceeded with url: /REST/SDMX_JSON.svc/CompactData/PCPI_IX/M.MA.PCPI_IX?startPeriod=2022-10&endPeriod=2023-11 (Caused by ConnectTimeoutError(<urllib3.connection.HTTPSConnection object at 0x7eae7eeb5640>, 'Connection to dataservices.imf.org timed out. (connect timeout=30)'))
[2025-09-26T06:08:08] ERROR IMF fetch MX: HTTPSConnectionPool(host='dataservices.imf.org', port=443): Max retries exceeded with url: /REST/SDMX_JSON.svc/CompactData/PCPI_IX/M.MX.PCPI_IX?startPeriod=2022-10&endPeriod=2023-11 (Caused by ConnectTimeoutError(<urllib3.connection.HTTPSConnection object at 0x7eae81dafc80>, 'Connection to dataservices.imf.org timed out. (connect timeout=30)'))
[2025-09-26T06:08:38] ERROR IMF fetch PA: HTTPSConnectionPool(host='dataservices.imf.org', port=443): Max retries exceeded with url: /REST/SDMX_JSON.svc/CompactData/PCPI_IX/M.PA.PCPI_IX?startPeriod=2022-10&endPeriod=2023-11 (Caused by ConnectTimeoutError(<urllib3.connection.HTTPSConnection object at 0x7eae82e46a80>, 'Connection to dataservices.imf.org timed out. (connect timeout=30)'))
[2025-09-26T06:09:08] ERROR IMF fetch PE: HTTPSConnectionPool(host='dataservices.imf.org', port=443): Max retries exceeded with url: /REST/SDMX_JSON.svc/CompactData/PCPI_IX/M.PE.PCPI_IX?startPeriod=2022-10&endPeriod=2023-11 (Caused by ConnectTimeoutError(<urllib3.connection.HTTPSConnection object at 0x7eae7ee692e0>, 'Connection to dataservices.imf.org timed out. (connect timeout=30)'))
[2025-09-26T06:09:39] ERROR IMF fetch PR: HTTPSConnectionPool(host='dataservices.imf.org', port=443): Max retries exceeded with url: /REST/SDMX_JSON.svc/CompactData/PCPI_IX/M.PR.PCPI_IX?startPeriod=2022-10&endPeriod=2023-11 (Caused by ConnectTimeoutError(<urllib3.connection.HTTPSConnection object at 0x7eae8203fad0>, 'Connection to dataservices.imf.org timed out. (connect timeout=30)'))
[2025-09-26T06:10:09] ERROR IMF fetch DO: HTTPSConnectionPool(host='dataservices.imf.org', port=443): Max retries exceeded with url: /REST/SDMX_JSON.svc/CompactData/PCPI_IX/M.DO.PCPI_IX?startPeriod=2022-10&endPeriod=2023-11 (Caused by ConnectTimeoutError(<urllib3.connection.HTTPSConnection object at 0x7eae7eebaa20>, 'Connection to dataservices.imf.org timed out. (connect timeout=30)'))
[2025-09-26T06:10:39] ERROR IMF fetch VE: HTTPSConnectionPool(host='dataservices.imf.org', port=443): Max retries exceeded with url: /REST/SDMX_JSON.svc/CompactData/PCPI_IX/M.VE.PCPI_IX?startPeriod=2022-10&endPeriod=2023-11 (Caused by ConnectTimeoutError(<urllib3.connection.HTTPSConnection object at 0x7eae7d30c170>, 'Connection to dataservices.imf.org timed out. (connect timeout=30)'))
[2025-09-26T06:10:39] Guardado: /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3/METRICAS/ipc_por_pais_fechaRef.csv con 12 filas.

# ============================================================
# IPC mensual (YoY %) por país — sin internet, con parche diciembre
# ============================================================
# En este bloque:
#  - Leemos países únicos desde dim_buildings.csv (excluimos DESCONOCIDO)
#  - Inferimos el mes objetivo a partir de la última fecha del train (mes-1)
#  - Creamos una tabla mensual de IPC YoY por país
#  - Aplicamos un parche con los IPC de diciembre 2023/2024 que nos diste
#  - Opción: forzar siempre el IPC anual de diciembre del mismo año
#  - Guardamos METRICAS/ipc_monthly_lookup.csv
# ============================================================

# ------------------------------
# Bloque 0 · Parámetros y rutas
# ------------------------------

# Definimos base del proyecto (ajustar si fuera necesario)
ruta_base_3 = "/content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3"
CSV_SEP     = ";"

RUTA_RESULTADOS = os.path.join(ruta_base_3, "RESULTADOS")
RUTA_METRICAS   = os.path.join(ruta_base_3, "METRICAS")
RUTA_LOGS       = os.path.join(ruta_base_3, "LOGS")

os.makedirs(RUTA_RESULTADOS, exist_ok=True)
os.makedirs(RUTA_METRICAS,   exist_ok=True)
os.makedirs(RUTA_LOGS,       exist_ok=True)

PATH_DIM     = os.path.join(RUTA_METRICAS,   "dim_buildings.csv")           # entrada
PATH_TRAIN   = os.path.join(RUTA_RESULTADOS, "train_full_2021_2023.csv")    # entrada (puede ser 2021..2025)
OUT_IPC_CSV  = os.path.join(RUTA_METRICAS,   "ipc_monthly_lookup.csv")      # salida

LOG_PATH = os.path.join(RUTA_LOGS, "ipc_monthly_build.log")

def log_ipc(msg: str):
    # Registramos mensajes en consola y en un log simple
    stamp = datetime.now().isoformat(timespec="seconds")
    line  = f"[{stamp}] {msg}"
    print(line)
    with open(LOG_PATH, "a", encoding="utf-8") as f:
        f.write(line + "\n")

# -----------------------------------------------------
# Bloque 1 · Tabla de IPC de diciembre (parche manual)
# -----------------------------------------------------
# Aquí ponemos los valores que nos diste. Mantendremos una estructura ampliable
# por años. Los valores están en porcentaje (p.ej. 3.10 = 3.10%).

IPC_DICIEMBRE = {
    2023: {
        "COLOMBIA": 9.28,
        "COSTA RICA": -1.77,
        "ESPAÑA": 3.10,
        "ITALIA": 0.60,
        "MALTA": 3.70,
        "MARRUECOS": 3.40,
        "MÉXICO": 4.66,
        "PANAMÁ": 1.90,
        "PERÚ": 3.24,
        "PUERTO RICO": 3.10,
        "REPÚBLICA DOMINICANA": 3.57,
        "VENEZUELA": 189.80
    },
    2024: {
        "COLOMBIA": 3.33,
        "COSTA RICA": 1.96,
        "ESPAÑA": 2.80,
        "ITALIA": 1.30,
        "MALTA": 1.80,
        "MARRUECOS": 0.70,
        "MÉXICO": 4.21,
        "PANAMÁ": -0.10,
        "PERÚ": 1.97,
        "PUERTO RICO": 1.90,
        "REPÚBLICA DOMINICANA": 3.35,
        "VENEZUELA": 189.80
    },
    # Lo dejamos preparado para añadir 2025 y futuros años
    # 2025: { "PAIS": valor, ... }
}

# Bandera de operación: si la activamos, forzamos SIEMPRE diciembre del año correspondiente
FORZAR_DICIEMBRE_SIEMPRE = False  # podemos activar cuando queramos

# -------------------------------------------
# Bloque 2 · Helpers de fechas y normalizado
# -------------------------------------------

def _ensure_ms(dt_series: pd.Series) -> pd.Series:
    # Normalizamos una serie de fechas al primer día de mes (frecuencia MS)
    s = pd.to_datetime(dt_series)
    return s.dt.to_period("M").dt.to_timestamp()

def _ult_mes_menos_1(df: pd.DataFrame, datecol: str) -> pd.Timestamp:
    # Obtenemos el último mes del train, lo normalizamos a MS, y restamos un mes
    if df.empty or datecol not in df.columns:
        raise ValueError("No podemos determinar la fecha objetivo: el train está vacío o falta la columna FECHA.")
    fechas = _ensure_ms(df[datecol])
    last  = fechas.max()
    # Restamos un mes (si last es 2025-08-01 => objetivo 2025-07-01)
    objetivo = (last.to_period("M") - 1).to_timestamp()
    return objetivo

def _build_base_table(paises: list[str], target_month: pd.Timestamp) -> pd.DataFrame:
    # Creamos una tabla base con una fila por país y el mes objetivo
    out = pd.DataFrame({
        "PAIS": paises,
        "FECHA": [target_month] * len(paises),
    })
    out["ANIO"] = out["FECHA"].dt.year
    out["MES"]  = out["FECHA"].dt.month
    # Inicializamos columnas de IPC y trazabilidad
    out["IPC_YOY_PCT"]          = np.nan
    out["IPC_YOY_PCT_DECEMBER"] = np.nan
    out["ESTIMADO_DEC_ANUAL"]   = 0
    out["STATUS"]               = "INIT"
    out["FUENTE_IPC"]           = "INIT"
    return out

def _apply_december_patch(df: pd.DataFrame) -> pd.DataFrame:
    # Aplicamos el parche de diciembre 2023/2024 a la columna IPC_YOY_PCT_DECEMBER
    df = df.copy()
    # Asignamos el diciembre anual disponible para el ANIO de cada fila
    mask_year_in_patch = df["ANIO"].isin(IPC_DICIEMBRE.keys())
    for idx in df.index[mask_year_in_patch]:
        year = int(df.at[idx, "ANIO"])
        pais = str(df.at[idx, "PAIS"]).upper().strip()
        if pais in IPC_DICIEMBRE.get(year, {}):
            df.at[idx, "IPC_YOY_PCT_DECEMBER"] = float(IPC_DICIEMBRE[year][pais])
    return df

def _fill_from_december_when_needed(df: pd.DataFrame) -> pd.DataFrame:
    # Rellenamos IPC_YOY_PCT con el diciembre anual si el mes objetivo no es diciembre
    # y si hay valor disponible
    df = df.copy()
    is_december_row = df["MES"].eq(12)
    # Caso 1: si el mes objetivo ES diciembre y tenemos valor, usamos directamente diciembre
    mask_use_direct_dec = is_december_row & df["IPC_YOY_PCT_DECEMBER"].notna()
    df.loc[mask_use_direct_dec, "IPC_YOY_PCT"]        = df.loc[mask_use_direct_dec, "IPC_YOY_PCT_DECEMBER"]
    df.loc[mask_use_direct_dec, "STATUS"]             = "OK_DEC_DIRECT"
    df.loc[mask_use_direct_dec, "FUENTE_IPC"]         = "USER_DECEMBER"
    df.loc[mask_use_direct_dec, "ESTIMADO_DEC_ANUAL"] = 0

    # Caso 2: si el mes objetivo NO es diciembre, estimamos con diciembre del mismo año si existe
    mask_estimate = (~is_december_row) & df["IPC_YOY_PCT"].isna() & df["IPC_YOY_PCT_DECEMBER"].notna()
    df.loc[mask_estimate, "IPC_YOY_PCT"]        = df.loc[mask_estimate, "IPC_YOY_PCT_DECEMBER"]
    df.loc[mask_estimate, "STATUS"]             = "OK_EST_DEC"
    df.loc[mask_estimate, "FUENTE_IPC"]         = "USER_DECEMBER"
    df.loc[mask_estimate, "ESTIMADO_DEC_ANUAL"] = 1

    return df

def _force_december_override(df: pd.DataFrame) -> pd.DataFrame:
    # Si activamos la bandera, sobreescribimos SIEMPRE con diciembre del mismo año cuando exista
    df = df.copy()
    if FORZAR_DICIEMBRE_SIEMPRE:
        mask_force = df["IPC_YOY_PCT_DECEMBER"].notna()
        df.loc[mask_force, "IPC_YOY_PCT"]        = df.loc[mask_force, "IPC_YOY_PCT_DECEMBER"]
        df.loc[mask_force, "STATUS"]             = "OK_OVERRIDE_DEC"
        df.loc[mask_force, "FUENTE_IPC"]         = "USER_DECEMBER"
        df.loc[mask_force, "ESTIMADO_DEC_ANUAL"] = 1
        log_ipc(f"Forzamos diciembre anual en {int(mask_force.sum())} filas (FORZAR_DICIEMBRE_SIEMPRE=True).")
    else:
        log_ipc("No forzamos diciembre anual (FORZAR_DICIEMBRE_SIEMPRE=False).")
    return df

# ---------------------------------------------------------
# Bloque 3 · Cargamos entradas y detectamos mes objetivo
# ---------------------------------------------------------

# 3.1 Países únicos desde dim_buildings
if not os.path.exists(PATH_DIM):
    raise FileNotFoundError(f"No encontramos dim_buildings en {PATH_DIM}")

dim = pd.read_csv(PATH_DIM, sep=CSV_SEP)
if "PAIS" not in dim.columns:
    raise KeyError("Esperábamos columna 'PAIS' en dim_buildings.csv")

paises_all = (dim["PAIS"].astype(str).str.strip().str.upper()
              .replace({"": np.nan})
              .dropna()
              .unique()
              .tolist())
paises = [p for p in paises_all if p != "DESCONOCIDO"]

log_ipc(f"Paises únicos en dim_buildings (sin 'DESCONOCIDO'): {paises}")

# 3.2 Mes objetivo desde train (último mes - 1)
if not os.path.exists(PATH_TRAIN):
    raise FileNotFoundError(f"No encontramos train en {PATH_TRAIN}")

train = pd.read_csv(PATH_TRAIN, sep=CSV_SEP)
if "FECHA" not in train.columns:
    raise KeyError("Esperábamos columna 'FECHA' en train.")

# Normalizamos la fecha
train["FECHA"] = _ensure_ms(train["FECHA"])
target_month = _ult_mes_menos_1(train, "FECHA")
log_ipc(f"Mes objetivo IPC (YoY) inferido: {target_month.strftime('%Y-%m')} (último train - 1 mes)")

# ---------------------------------------------------------
# Bloque 4 · Construcción y parche del IPC mensual
# ---------------------------------------------------------

# 4.1 Base por país
ipc_df = _build_base_table(paises, target_month)

# 4.2 Parche: valores de diciembre disponibles
ipc_df = _apply_december_patch(ipc_df)

# 4.3 Relleno: si no es diciembre, usamos diciembre del mismo año (estimación)
ipc_df = _fill_from_december_when_needed(ipc_df)

# 4.4 Opción: forzar diciembre siempre (si activamos la bandera)
ipc_df = _force_december_override(ipc_df)

# 4.5 Completar trazabilidad por defecto donde no asignamos IPC
mask_empty = ipc_df["IPC_YOY_PCT"].isna()
ipc_df.loc[mask_empty, "STATUS"]     = "MISSING"
ipc_df.loc[mask_empty, "FUENTE_IPC"] = ipc_df.loc[mask_empty, "FUENTE_IPC"].replace("INIT", "NO_SOURCE")

# Orden de columnas y tipos
cols_order = [
    "PAIS", "FECHA", "ANIO", "MES",
    "IPC_YOY_PCT", "IPC_YOY_PCT_DECEMBER",
    "ESTIMADO_DEC_ANUAL", "STATUS", "FUENTE_IPC"
]
ipc_df = ipc_df[cols_order].sort_values(["PAIS", "FECHA"]).reset_index(drop=True)

# Guardamos CSV
ipc_df.to_csv(OUT_IPC_CSV, sep=CSV_SEP, index=False)
log_ipc(f"Guardado: {OUT_IPC_CSV} con {len(ipc_df)} filas.")

# ---------------------------------------------------------
# Bloque 5 · Resumen operativo
# ---------------------------------------------------------
n_ok  = int((ipc_df["STATUS"].isin(["OK_DEC_DIRECT","OK_EST_DEC","OK_OVERRIDE_DEC"])).sum())
n_miss= int((ipc_df["STATUS"]=="MISSING").sum())
log_ipc(f"Resumen — filas OK={n_ok} | faltantes={n_miss}")

# Mostramos una vista rápida
display(ipc_df.head(10))

[2025-09-26T05:11:00] Paises únicos en dim_buildings (sin 'DESCONOCIDO'): ['ESPAÑA', 'PERÚ', 'PANAMÁ', 'MÉXICO', 'COSTA RICA', 'COLOMBIA', 'ITALIA', 'REPÚBLICA DOMINICANA', 'MALTA', 'PUERTO RICO', 'MARRUECOS', 'VENEZUELA']
[2025-09-26T05:11:00] Mes objetivo IPC (YoY) inferido: 2023-11 (último train - 1 mes)
[2025-09-26T05:11:00] No forzamos diciembre anual (FORZAR_DICIEMBRE_SIEMPRE=False).
[2025-09-26T05:11:00] Guardado: /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3/METRICAS/ipc_monthly_lookup.csv con 12 filas.
[2025-09-26T05:11:00] Resumen — filas OK=12 | faltantes=0

# ============================================================
# PASO 12 (PRODUCCIÓN con anclas)
# Reconciliación MinT usando:
#   - Real 2023 por segmento (desde TRAIN 2021-2023)
#   - IPC (YoY %) por país desde METRICAS/ipc_monthly_lookup.csv
#   - Distribución mensual proporcional al share del pronóstico base
#   - Robustez con W distinto a identidad
#     con referencias o anclas (TOP desde FM×PAIS y REGION por share base)
#     y un export para auditoría
# Salidas:
#   RESULTADOS/preds_reconciliadas_step12.csv
#   REPORTING/compare_estructural_step12.csv
#   REPORTING/figs_step12/*.png
# ============================================================
# ============================================================
# Robustez con W distinto a identidad
# con referencias o anclas (TOP desde FM×PAIS y REGION por share base)
# y un export para auditoría.
# dos versiones de reconciliación la estandar STD y la ACTIVE
# ============================================================

# --- Parámetros/Rutas (ajusta si hace falta) ---
ruta_base_3   = "/content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3"
CSV_SEP       = ";"

RUTA_RESULTADOS = os.path.join(ruta_base_3, "RESULTADOS")
RUTA_METRICAS   = os.path.join(ruta_base_3, "METRICAS")
RUTA_REPORTING  = os.path.join(ruta_base_3, "REPORTING")
RUTA_LOGS       = os.path.join(ruta_base_3, "LOGS")
for _d in [RUTA_RESULTADOS, RUTA_METRICAS, RUTA_REPORTING, RUTA_LOGS]:
    os.makedirs(_d, exist_ok=True)

# Entradas principales
PATH_PREDS   = os.path.join(RUTA_RESULTADOS, "preds_por_serie_2024.csv")         # base 2024 por serie
PATH_DIM     = os.path.join(RUTA_METRICAS,   "dim_buildings.csv")                # PAIS/REGION
PATH_TRAIN   = os.path.join(RUTA_RESULTADOS, "train_full_2021_2023.csv")         # real 2023
PATH_IPC     = os.path.join(RUTA_METRICAS,   "ipc_monthly_lookup.csv")           # IPC YoY % por país (mes objetivo)
PATH_PASO8   = os.path.join(RUTA_METRICAS,   "paso8_metrics_2024_por_pareja.csv")# para W MASE12²

# Salidas
OUT_PREDS_RECONC = os.path.join(RUTA_RESULTADOS, "preds_reconciliadas_step12.csv")
# OUT_PREDS_RECONC_BEFORE = os.path.join(RUTA_RESULTADOS, "preds_reconciliadas_step12_beforeclip.csv")
OUT_COMPARE      = os.path.join(RUTA_REPORTING,  "compare_estructural_step12.csv")
FIGS_DIR         = os.path.join(RUTA_REPORTING,  "figs_step12")
os.makedirs(FIGS_DIR, exist_ok=True)

# Export anclas (auditoría)
TOP_CSV = os.path.join(RUTA_METRICAS, "target_top_month_step12.csv")
FMP_CSV = os.path.join(RUTA_METRICAS, "target_fmp_month_step12.csv")
FMR_CSV = os.path.join(RUTA_METRICAS, "target_fmr_month_step12.csv")

# Claves/Frecuencia
PAIR_COLS   = ["ID_BUILDING","FM_COST_TYPE"]
DATECOL     = "FECHA"
VALCOL_PRED = "yhat_combo"
MONTHS_2024 = pd.date_range("2024-01-01", periods=12, freq="MS")
EPS = 1e-8

def log12(msg: str):
    ts = datetime.now().isoformat(timespec="seconds")
    line = f"[{ts}] {msg}"
    print(line)
    with open(os.path.join(RUTA_LOGS, "estrategia3_step12.log"), "a", encoding="utf-8") as f:
        f.write(line + "\n")

def _ensure_ms(df: pd.DataFrame, datecol: str) -> pd.DataFrame:
    out = df.copy()
    out[datecol] = pd.to_datetime(out[datecol]).dt.to_period("M").dt.to_timestamp()
    return out

def _read_csv(path: str, required=True) -> pd.DataFrame:
    if os.path.exists(path):
        df = pd.read_csv(path, sep=CSV_SEP)
        df.columns = [str(c).strip().lstrip("\ufeff") for c in df.columns]
        return df
    if required:
        raise FileNotFoundError(f"Falta archivo requerido: {path}")
    log12(f"AVISO: no existe {path}, seguimos.")
    return pd.DataFrame()

def _norm_pair_keys(df: pd.DataFrame) -> pd.DataFrame:
    if df is None or df.empty: return df
    out = df.copy()
    out.columns = [str(c).strip() for c in out.columns]
    if "ID_BUILDING" in out.columns:
        try:
            out["ID_BUILDING"] = pd.to_numeric(out["ID_BUILDING"], errors="coerce").astype("Int64")
        except Exception:
            pass
    if "FM_COST_TYPE" in out.columns:
        out["FM_COST_TYPE"] = out["FM_COST_TYPE"].astype(str).str.strip()
    return out

# ------------------------------
# 0) Carga entradas
# ------------------------------
log12("Paso 12 — Inicio.")
preds_df = _ensure_ms(_norm_pair_keys(_read_csv(PATH_PREDS, required=True)), DATECOL)
if VALCOL_PRED not in preds_df.columns:
    raise KeyError(f"Esperábamos columna {VALCOL_PRED} en {os.path.basename(PATH_PREDS)}.")
keep_pred_cols = [c for c in [*PAIR_COLS, DATECOL, VALCOL_PRED, "route", "fallback_flag"] if c in preds_df.columns]
preds_df = preds_df[keep_pred_cols].copy()
# agregamos por si hay duplicados de (pair,mes)
preds_df = (preds_df.groupby(PAIR_COLS+[DATECOL], as_index=False)
                   .agg({VALCOL_PRED:"sum",
                         **({"route":"last"} if "route" in preds_df.columns else {}),
                         **({"fallback_flag":"max"} if "fallback_flag" in preds_df.columns else {})}))

dim_df   = _norm_pair_keys(_read_csv(PATH_DIM, required=True))
ipc_df   = _read_csv(PATH_IPC, required=True)
train_df = _ensure_ms(_norm_pair_keys(_read_csv(PATH_TRAIN, required=True)), DATECOL)

# normalizamos dim (solo lo necesario)
dim_keep = [c for c in ["ID_BUILDING","PAIS","REGION"] if c in dim_df.columns]
dim_slim = dim_df[dim_keep].drop_duplicates() if len(dim_keep) else pd.DataFrame(columns=["ID_BUILDING","PAIS","REGION"])

# ------------------------------
# 1) Metadatos jerárquicos
# ------------------------------
bottom_df    = preds_df.merge(dim_slim, on="ID_BUILDING", how="left")
keys_bottom  = bottom_df[PAIR_COLS].drop_duplicates().sort_values(PAIR_COLS).reset_index(drop=True)
keys_top     = bottom_df[["FM_COST_TYPE"]].drop_duplicates().dropna().reset_index(drop=True)
keys_fm_pais = bottom_df[["FM_COST_TYPE","PAIS"]].dropna().drop_duplicates().reset_index(drop=True)
keys_fm_reg  = bottom_df[["FM_COST_TYPE","REGION"]].dropna().drop_duplicates().reset_index(drop=True)

n_bottom = len(keys_bottom)
log12(f"Nodos: bottom={n_bottom} | top={len(keys_top)} | fm×pais={len(keys_fm_pais)} | fm×region={len(keys_fm_reg)}")

# helpers de hijos
def _child_idx_for_top(fm):
    mask = (keys_bottom["FM_COST_TYPE"]==fm)
    return keys_bottom.index[mask].tolist()

def _child_idx_for_fm_pais(fm, pais):
    sub = bottom_df.merge(keys_bottom, on=PAIR_COLS, how="inner")
    sub = sub[(sub["FM_COST_TYPE"]==fm) & (sub["PAIS"]==pais)]
    return keys_bottom.merge(sub[PAIR_COLS].drop_duplicates(), on=PAIR_COLS, how="inner").index.tolist()

def _child_idx_for_fm_region(fm, region):
    sub = bottom_df.merge(keys_bottom, on=PAIR_COLS, how="inner")
    sub = sub[(sub["FM_COST_TYPE"]==fm) & (sub["REGION"]==region)]
    return keys_bottom.merge(sub[PAIR_COLS].drop_duplicates(), on=PAIR_COLS, how="inner").index.tolist()

# ------------------------------
# 2) Matriz A (agregados)
#     Orden: TOP → FM×PAIS → FM×REGION
# ------------------------------
def _build_A_matrix(n_bottom, keys_top, keys_fm_pais, keys_fm_reg):
    rows = []
    # TOP
    if not keys_top.empty:
        for (fm,) in keys_top.itertuples(index=False, name=None):
            idxs = _child_idx_for_top(fm)
            row = np.zeros(n_bottom, dtype=float)
            if idxs: row[np.asarray(idxs, dtype=int)] = 1.0
            rows.append(row)
    # FM×PAIS
    if not keys_fm_pais.empty:
        for (fm, pais) in keys_fm_pais.itertuples(index=False, name=None):
            idxs = _child_idx_for_fm_pais(fm, pais)
            row = np.zeros(n_bottom, dtype=float)
            if idxs: row[np.asarray(idxs, dtype=int)] = 1.0
            rows.append(row)
    # FM×REGION
    if not keys_fm_reg.empty:
        for (fm, region) in keys_fm_reg.itertuples(index=False, name=None):
            idxs = _child_idx_for_fm_region(fm, region)
            row = np.zeros(n_bottom, dtype=float)
            if idxs: row[np.asarray(idxs, dtype=int)] = 1.0
            rows.append(row)

    if len(rows)==0:
        A = np.zeros((0, n_bottom), dtype=float)
    else:
        A = np.vstack(rows)
    return A

A = _build_A_matrix(n_bottom, keys_top, keys_fm_pais, keys_fm_reg)
log12(f"Matriz A: shape={A.shape}")

# ------------------------------
# 3) W diagonal (MASE12² de Paso 8; fallback identidad)
# ------------------------------
W_diag = np.ones(n_bottom, dtype=float)  # fallback
try:
    if os.path.exists(PATH_PASO8):
        paso8 = pd.read_csv(PATH_PASO8, sep=CSV_SEP)
        paso8.columns = [str(c).strip() for c in paso8.columns]
        if "vista" in paso8.columns:
            paso8 = paso8[paso8["vista"].astype(str).str.lower().eq("all_months")]
        assert set(PAIR_COLS).issubset(paso8.columns) and ("MASE12" in paso8.columns)
        # mapa MASE
        mase_map = {
            (int(pd.to_numeric(r[PAIR_COLS[0]], errors="coerce")), str(r[PAIR_COLS[1]]).strip()):
            float(pd.to_numeric(r["MASE12"], errors="coerce"))
            for _, r in paso8[[*PAIR_COLS,"MASE12"]].drop_duplicates(PAIR_COLS).iterrows()
        }
        cnt_mase = 0
        for i, (bid, ctype) in enumerate(keys_bottom[PAIR_COLS].itertuples(index=False, name=None)):
            key = (int(bid) if pd.notna(bid) else bid, str(ctype).strip())
            m = mase_map.get(key, np.nan)
            if np.isfinite(m):
                W_diag[i] = max(m**2, 1e-8)
                cnt_mase += 1
            else:
                W_diag[i] = 1.0
        log12(f"W diagonal: MASE12² disponibles={cnt_mase} | identidad={n_bottom - cnt_mase}")
    else:
        log12(f"AVISO: no existe {PATH_PASO8}. W=identidad.")
except Exception as e:
    log12(f"AVISO: no se pudo construir W desde paso8 ({e}). W=identidad.")
W_inv_diag = 1.0 / np.maximum(W_diag, 1e-8)

# ------------------------------
# 4) Anclas anuales (FM×PAIS) y mensualización por share base
#     - real 2023 desde TRAIN (sum 2023)
#     - factor IPC por país (último IPC en ipc_monthly_lookup)
#     - target anual = real_2023 * (1 + IPC/100)
#     - reparto mensual por shares del pronóstico base (sin IPC extra)
#     - TOP = suma por FM,mes; REGION = share dentro de FM×PAIS
# ------------------------------
# real 2023 por FM×PAIS
train_with_dim = train_df.merge(dim_slim, on="ID_BUILDING", how="left")
real_2023_fmp = (train_with_dim[(train_with_dim[DATECOL]>=pd.Timestamp("2023-01-01")) &
                                (train_with_dim[DATECOL]<=pd.Timestamp("2023-12-01"))]
                 .dropna(subset=["PAIS"])
                 .groupby(["FM_COST_TYPE","PAIS"], as_index=False)["cost_float_mod"].sum()
                 .rename(columns={"cost_float_mod":"real_2023"}))

# IPC país (tomamos el último registro por país del CSV)
ipc_df[DATECOL] = pd.to_datetime(ipc_df[DATECOL], errors="coerce")
ipc_latest = (ipc_df.sort_values(["PAIS", DATECOL])
                   .groupby("PAIS", as_index=False).tail(1)[["PAIS","IPC_YOY_PCT"]])
ipc_latest["factor"] = 1.0 + pd.to_numeric(ipc_latest["IPC_YOY_PCT"], errors="coerce").fillna(0.0)/100.0

# target anual fmp = real_2023 * factor
target_annual_fmp = (real_2023_fmp
                     .merge(ipc_latest[["PAIS","factor"]], on="PAIS", how="left")
                     .assign(factor=lambda d: d["factor"].fillna(1.0))
                     .assign(target_annual=lambda d: d["real_2023"] * d["factor"]))

# shares mensuales desde el base 2024 por FM×PAIS×mes
preds_with_dim = preds_df.merge(dim_slim, on="ID_BUILDING", how="left")
base_fmp_month = (preds_with_dim.dropna(subset=["PAIS"])
                  .groupby(["FM_COST_TYPE","PAIS",DATECOL], as_index=False)[VALCOL_PRED]
                  .sum().rename(columns={VALCOL_PRED:"base_sum"}))

sum_base_fmp = (base_fmp_month.groupby(["FM_COST_TYPE","PAIS"], as_index=False)["base_sum"].sum()
                .rename(columns={"base_sum":"base_sum_annual"}))
shares_fmp = base_fmp_month.merge(sum_base_fmp, on=["FM_COST_TYPE","PAIS"], how="left")
shares_fmp["share"] = shares_fmp["base_sum"] / shares_fmp["base_sum_annual"].replace(0.0, np.nan)

# target_fmp_month = share * target_annual
target_fmp_month = (shares_fmp
                    .merge(target_annual_fmp[["FM_COST_TYPE","PAIS","target_annual"]],
                           on=["FM_COST_TYPE","PAIS"], how="left"))
target_fmp_month["target"] = target_fmp_month["share"].fillna(0.0) * target_fmp_month["target_annual"].fillna(0.0)
target_fmp_month = target_fmp_month[["FM_COST_TYPE","PAIS",DATECOL,"target"]].copy()

# TOP mensual = suma de fmp por FM,mes
target_top_month = (target_fmp_month.groupby(["FM_COST_TYPE",DATECOL], as_index=False)["target"].sum())

# FM×REGION: share mensual dentro de cada (FM,PAIS,mes) sobre base
base_fmr_month = (preds_with_dim.dropna(subset=["PAIS","REGION"])
                  .groupby(["FM_COST_TYPE","PAIS","REGION",DATECOL], as_index=False)[VALCOL_PRED]
                  .sum().rename(columns={VALCOL_PRED:"base_sum"}))
shares_fmr = base_fmr_month.merge(
    base_fmp_month.rename(columns={"base_sum":"base_sum_fmp"}),
    on=["FM_COST_TYPE","PAIS",DATECOL], how="left"
)
shares_fmr["share"] = shares_fmr["base_sum"] / shares_fmr["base_sum_fmp"].replace(0.0, np.nan)

target_fmr_month = shares_fmr.merge(
    target_fmp_month.rename(columns={"target":"target_fmp"}),
    on=["FM_COST_TYPE","PAIS",DATECOL], how="left"
)
target_fmr_month["target"] = target_fmr_month["share"].fillna(0.0) * target_fmr_month["target_fmp"].fillna(0.0)
target_fmr_month = target_fmr_month[["FM_COST_TYPE","PAIS","REGION",DATECOL,"target"]].copy()

# ------------------------------
# 5) Export anclas para auditoría
# ------------------------------
topexp = target_top_month.copy()
topexp["nivel"] = "FM_COST_TYPE"
topexp["clave"] = topexp["FM_COST_TYPE"].astype(str)
topexp = topexp[["nivel","clave",DATECOL,"target"]].sort_values([DATECOL,"clave"])
topexp.to_csv(TOP_CSV, sep=CSV_SEP, index=False)

fmpexp = target_fmp_month.copy()
fmpexp["nivel"] = "FM_COST_TYPE|PAIS"
fmpexp["clave"] = fmpexp["FM_COST_TYPE"].astype(str) + "|" + fmpexp["PAIS"].astype(str)
fmpexp = fmpexp[["nivel","clave",DATECOL,"target"]].sort_values([DATECOL,"clave"])
fmpexp.to_csv(FMP_CSV, sep=CSV_SEP, index=False)

fmrexp = target_fmr_month.copy()
fmrexp["nivel"] = "FM_COST_TYPE|REGION"
fmrexp["clave"] = fmrexp["FM_COST_TYPE"].astype(str) + "|" + fmrexp["REGION"].astype(str)
fmrexp = fmrexp[["nivel","clave",DATECOL,"target"]].sort_values([DATECOL,"clave"])
fmrexp.to_csv(FMR_CSV, sep=CSV_SEP, index=False)

log12(f"Exportadas anclas mensuales usadas: {TOP_CSV} | {FMP_CSV} | {FMR_CSV}")

# ============================================================
# 6) Reconciliación P12 — variante estándar y variante SOLO ACTIVOS
#     - estándar: reconcilia todos los bottom
#     - solo_activos: congela pares con base <= ZERO_THR
#     Guardamos ANTES y DESPUÉS del clip para ambas variantes
# ============================================================

ZERO_THR = 1e-9  # umbral para decidir “cero estructural”

# Salidas específicas de variantes
OUT_P12_STD          = os.path.join(RUTA_RESULTADOS, "preds_reconciliadas_step12.csv")  # mantiene nombre histórico
OUT_P12_STD_BEFORE   = os.path.join(RUTA_RESULTADOS, "preds_reconciliadas_step12_beforeclip.csv")
OUT_P12_ACTIVE       = os.path.join(RUTA_RESULTADOS, "preds_reconciliadas_step12_activeonly.csv")
OUT_P12_ACTIVE_BEFORE= os.path.join(RUTA_RESULTADOS, "preds_reconciliadas_step12_activeonly_beforeclip.csv")

# Comparativas
OUT_CMP_BOTTOM = os.path.join(RUTA_REPORTING, "compare_step12_std_vs_active_bottom.csv")
OUT_CMP_PORT   = os.path.join(RUTA_REPORTING, "compare_step12_std_vs_active_portfolio.csv")
OUT_FIG_DIR    = os.path.join(RUTA_REPORTING, "figs_step12_compare_active")
os.makedirs(OUT_FIG_DIR, exist_ok=True)

def _yb_from_preds(month_ts):
    sub = preds_df[preds_df[DATECOL].eq(month_ts)]
    merged = keys_bottom.merge(sub, on=PAIR_COLS, how="left")
    return merged[VALCOL_PRED].fillna(0.0).values.astype(float)

def _c_from_targets(month_ts):
    m = pd.Timestamp(month_ts).to_period("M").to_timestamp()
    parts = []

    # TOP
    if not keys_top.empty:
        want = (keys_top.assign(**{DATECOL:m})
                .merge(target_top_month.rename(columns={"target":"t"}),
                       on=["FM_COST_TYPE",DATECOL], how="left")
                .sort_values(["FM_COST_TYPE"]))
        parts.append(want["t"].fillna(0.0).values.astype(float))

    # FM×PAIS
    if not keys_fm_pais.empty:
        want = (keys_fm_pais.assign(**{DATECOL:m})
                .merge(target_fmp_month.rename(columns={"target":"t"}),
                       on=["FM_COST_TYPE","PAIS",DATECOL], how="left")
                .sort_values(["FM_COST_TYPE","PAIS"]))
        parts.append(want["t"].fillna(0.0).values.astype(float))

    # FM×REGION
    if not keys_fm_reg.empty:
        want = (keys_fm_reg.assign(**{DATECOL:m})
                .merge(target_fmr_month.rename(columns={"target":"t"}),
                       on=["FM_COST_TYPE","REGION",DATECOL], how="left")
                .sort_values(["FM_COST_TYPE","REGION"]))
        parts.append(want["t"].fillna(0.0).values.astype(float))

    if len(parts)==0:
        return np.zeros(0, dtype=float)
    return np.concatenate(parts, axis=0)

def _mint_reconcile_full(yb, c):
    """Reconciliación estándar sobre todos los nodos."""
    if A.shape[0]==0 or yb.size==0:
        y_pre = yb.copy()
        y_pos = np.maximum(y_pre, 0.0)
        return y_pre, y_pos
    AW   = A * W_diag
    AWAT = AW @ A.T
    try:
        AWAT_inv = np.linalg.pinv(AWAT, rcond=1e-10)
    except Exception:
        AWAT_inv = np.linalg.pinv(AWAT + 1e-8*np.eye(AWAT.shape[0]))
    r     = c - (A @ yb)
    v     = AWAT_inv @ r
    delta = (A.T * W_diag[:, None]) @ v
    y_pre = yb + delta
    y_pos = np.maximum(y_pre, 0.0)
    return y_pre, y_pos

def _mint_reconcile_activeonly(yb, c, zero_thr=ZERO_THR):
    """
    Reconciliación SOLO en nodos activos S = {i: yb_i > zero_thr}.
    F (resto) se fija. Relaja filas sin hijos activos.
    """
    m = yb.size
    if A.shape[0] == 0 or m == 0:
        y_pre = yb.copy()
        y_pos = np.maximum(y_pre, 0.0)
        return y_pre, y_pos

    active_mask = (yb > zero_thr)
    idx_S = np.where(active_mask)[0]
    idx_F = np.where(~active_mask)[0]

    if idx_S.size == 0:
        y_pre = yb.copy()
        y_pos = np.maximum(y_pre, 0.0)
        return y_pre, y_pos

    A_S = A[:, idx_S]
    A_F = A[:, idx_F] if idx_F.size>0 else np.zeros((A.shape[0],0))
    yS  = yb[idx_S]
    yF  = yb[idx_F]
    W_S = W_diag[idx_S]

    r = c - (A_S @ yS) - (A_F @ yF)

    if A_S.size > 0:
        no_dof = (np.abs(A_S).sum(axis=1) <= 0)
        if np.any(no_dof):
            r = r.copy()
            r[no_dof] = 0.0

    if A_S.size == 0:
        y_pre = yb.copy()
        y_pos = np.maximum(y_pre, 0.0)
        return y_pre, y_pos

    AW   = A_S * W_S
    AWAT = AW @ A_S.T
    try:
        AWAT_inv = np.linalg.pinv(AWAT, rcond=1e-10)
    except Exception:
        AWAT_inv = np.linalg.pinv(AWAT + 1e-8*np.eye(AWAT.shape[0]))

    v      = AWAT_inv @ r
    deltaS = (A_S.T * W_S[:, None]) @ v

    y_pre = yb.copy()
    y_pre[idx_S] = yS + deltaS
    y_pos = np.maximum(y_pre, 0.0)
    return y_pre, y_pos

def _run_variant(name, reconcile_fn, out_after, out_before):
    rows_after, rows_before = [], []
    for mth in MONTHS_2024:
        yb = _yb_from_preds(mth)
        c  = _c_from_targets(mth)
        y_pre, y_pos = reconcile_fn(yb, c)

        df_after = keys_bottom.copy()
        df_after[DATECOL]       = mth
        df_after["yhat_base"]   = yb
        df_after["yhat_reconc"] = y_pos
        rows_after.append(df_after)

        df_before = keys_bottom.copy()
        df_before[DATECOL]       = mth
        df_before["yhat_base"]   = yb
        df_before["yhat_reconc"] = y_pre
        rows_before.append(df_before)

    after  = (pd.concat(rows_after,  ignore_index=True)
              .sort_values(PAIR_COLS+[DATECOL]).reset_index(drop=True))
    before = (pd.concat(rows_before, ignore_index=True)
              .sort_values(PAIR_COLS+[DATECOL]).reset_index(drop=True))

    after.to_csv(out_after,   sep=CSV_SEP, index=False)
    before.to_csv(out_before, sep=CSV_SEP, index=False)
    negs = int((after["yhat_reconc"] < 0).sum())
    log12(f"[{name}] Guardado AFTER:  {out_after} ({len(after)} filas) | negativos={negs}")
    log12(f"[{name}] Guardado BEFORE: {out_before} ({len(before)} filas)")
    return after, before

log12(f"Meses a reconciliar: {len(MONTHS_2024)} ({MONTHS_2024[0].date()} .. {MONTHS_2024[-1].date()})")

# Variante estándar (todos los nodos)
p12_std_after, p12_std_before = _run_variant(
    name="P12_STD",
    reconcile_fn=_mint_reconcile_full,
    out_after=OUT_P12_STD,
    out_before=OUT_P12_STD_BEFORE
)

# Variante solo-activos (congelando base≈0)
p12_act_after, p12_act_before = _run_variant(
    name="P12_ACTIVE",
    reconcile_fn=lambda yb, c: _mint_reconcile_activeonly(yb, c, ZERO_THR),
    out_after=OUT_P12_ACTIVE,
    out_before=OUT_P12_ACTIVE_BEFORE
)
# ------------------------------
# 8) Figuras rápidas (self-contained, no depende de variables previas)
# ------------------------------
try:
    import matplotlib.pyplot as plt

    # Aseguramos un DataFrame base para las figuras
    _std_df = p12_std_after if 'p12_std_after' in locals() else pd.read_csv(OUT_P12_STD, sep=CSV_SEP, parse_dates=[DATECOL])

    # Histograma de deltas bottom (STD)
    tmp_fig = _std_df.copy()
    tmp_fig["delta"] = tmp_fig["yhat_reconc"] - tmp_fig["yhat_base"]

    plt.figure(figsize=(8,5))
    tmp_fig["delta"].plot(kind="hist", bins=40)
    plt.title("Paso 12 (STD) — Deltas bottom (yhat_reconc - yhat_base)")
    plt.xlabel("Delta"); plt.ylabel("Frecuencia")
    plt.tight_layout()
    plt.savefig(os.path.join(FIGS_DIR, "hist_deltas_bottom_step12.png"))
    plt.close()

    # Serie portfolio (STD)
    port_fig = (tmp_fig
                .groupby(DATECOL, as_index=False)
                .agg(y_base=("yhat_base","sum"), y_reco=("yhat_reconc","sum")))
    plt.figure(figsize=(10,5))
    plt.plot(port_fig[DATECOL], port_fig["y_base"], label="Base")
    plt.plot(port_fig[DATECOL], port_fig["y_reco"], label="Reconc P12 (STD)")
    plt.title("Paso 12 — Portfolio mensual: Base vs Reconciliado (STD)")
    plt.xlabel("Mes"); plt.ylabel("Suma mensual"); plt.legend()
    plt.tight_layout()
    plt.savefig(os.path.join(FIGS_DIR, "line_portfolio_base_vs_reco_step12.png"))
    plt.close()

    # Boxplot por FM_COST_TYPE de deltas (STD)
    box_df = tmp_fig.groupby("FM_COST_TYPE")["delta"].apply(list).to_dict()
    if len(box_df) > 0:
        plt.figure(figsize=(10,5))
        plt.boxplot(list(box_df.values()), labels=list(box_df.keys()), showfliers=False)
        plt.xticks(rotation=45, ha="right")
        plt.title("Paso 12 (STD) — Deltas por FM_COST_TYPE")
        plt.ylabel("Delta")
        plt.tight_layout()
        plt.savefig(os.path.join(FIGS_DIR, "box_deltas_por_fmcost_step12.png"))
        plt.close()

    log12(f"Figuras en {FIGS_DIR}")
except Exception as e:
    log12(f"AVISO: no fue posible generar figuras (bloque 8): {e}")

# ------------------------------
# 9) Resumen final robusto
# ------------------------------
# Asegura preds_reconc para el conteo de negativos
if 'p12_std_after' in locals():
    preds_reconc = p12_std_after.copy()
else:
    preds_reconc = pd.read_csv(OUT_P12_STD, sep=CSV_SEP, parse_dates=[DATECOL])

neg_counts = int((preds_reconc["yhat_reconc"] < 0).sum())
log12(f"Valores negativos en yhat_reconc (STD): {neg_counts} (esperado 0 por clip)")
log12("Paso 12 finalizado.")

[2025-09-27T08:25:30] Paso 12 — Inicio.
[2025-09-27T08:25:31] Nodos: bottom=2429 | top=8 | fm×pais=49 | fm×region=573
[2025-09-27T08:25:52] Matriz A: shape=(630, 2429)
[2025-09-27T08:25:53] W diagonal: MASE12² disponibles=2171 | identidad=258
[2025-09-27T08:25:53] Exportadas anclas mensuales usadas: /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3/METRICAS/target_top_month_step12.csv | /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3/METRICAS/target_fmp_month_step12.csv | /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3/METRICAS/target_fmr_month_step12.csv
[2025-09-27T08:25:53] Meses a reconciliar: 12 (2024-01-01 .. 2024-12-01)
[2025-09-27T08:26:05] [P12_STD] Guardado AFTER:  /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3/RESULTADOS/preds_reconciliadas_step12.csv (29148 filas) | negativos=0
[2025-09-27T08:26:05] [P12_STD] Guardado BEFORE: /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3/RESULTADOS/preds_reconciliadas_step12_beforeclip.csv (29148 filas)
[2025-09-27T08:26:17] [P12_ACTIVE] Guardado AFTER:  /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3/RESULTADOS/preds_reconciliadas_step12_activeonly.csv (29148 filas) | negativos=0
[2025-09-27T08:26:17] [P12_ACTIVE] Guardado BEFORE: /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3/RESULTADOS/preds_reconciliadas_step12_activeonly_beforeclip.csv (29148 filas)
[2025-09-27T08:26:18] Figuras en /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3/REPORTING/figs_step12
[2025-09-27T08:26:18] Valores negativos en yhat_reconc (STD): 0 (esperado 0 por clip)
[2025-09-27T08:26:18] Paso 12 finalizado.

# === Post-checks Paso 12 (coherencia y diagnóstico, STD vs ACTIVE) ===

ruta_base_3 = "/content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3"
CSV_SEP = ";"
RUTA_RESULTADOS = os.path.join(ruta_base_3, "RESULTADOS")
RUTA_METRICAS   = os.path.join(ruta_base_3, "METRICAS")

PAIR_COLS = ["ID_BUILDING","FM_COST_TYPE"]
DATECOL   = "FECHA"

# Entradas generadas por el paso 12 (nombres del bloque actualizado)
PATH_STD_AFTER    = os.path.join(RUTA_RESULTADOS, "preds_reconciliadas_step12.csv")
PATH_ACTIVE_AFTER = os.path.join(RUTA_RESULTADOS, "preds_reconciliadas_step12_activeonly.csv")
PATH_STD_BEFORE   = os.path.join(RUTA_RESULTADOS, "preds_reconciliadas_step12_beforeclip.csv")             # opcional
PATH_ACT_BEFORE   = os.path.join(RUTA_RESULTADOS, "preds_reconciliadas_step12_activeonly_beforeclip.csv")  # opcional

# Carga principal
std = pd.read_csv(PATH_STD_AFTER,    sep=CSV_SEP, parse_dates=[DATECOL])
dim = pd.read_csv(os.path.join(RUTA_METRICAS, "dim_buildings.csv"), sep=CSV_SEP)
ipc = pd.read_csv(os.path.join(RUTA_METRICAS, "ipc_monthly_lookup.csv"), sep=CSV_SEP, parse_dates=["FECHA"])

# Variante ACTIVE (si existe)
active_exists = os.path.exists(PATH_ACTIVE_AFTER)
if active_exists:
    act = pd.read_csv(PATH_ACTIVE_AFTER, sep=CSV_SEP, parse_dates=[DATECOL])

# 1) Coherencia FM_COST_TYPE×PAIS mensual — base vs reconc (STD y, si existe, ACTIVE)
df_std = std.merge(dim[["ID_BUILDING","PAIS"]], on="ID_BUILDING", how="left")
g_base_std = df_std.groupby(["FM_COST_TYPE","PAIS",DATECOL], as_index=False)["yhat_base"].sum()
g_reco_std = df_std.groupby(["FM_COST_TYPE","PAIS",DATECOL], as_index=False)["yhat_reconc"].sum()
cmp_fmp_std = g_base_std.merge(g_reco_std, on=["FM_COST_TYPE","PAIS",DATECOL])
cmp_fmp_std["delta"] = cmp_fmp_std["yhat_reconc"] - cmp_fmp_std["yhat_base"]

print(" Dispersión de delta (FM×PAIS mensual) — VARIANTE ESTÁNDAR:")
print(cmp_fmp_std["delta"].describe())

if active_exists:
    df_act = act.merge(dim[["ID_BUILDING","PAIS"]], on="ID_BUILDING", how="left")
    g_reco_act = df_act.groupby(["FM_COST_TYPE","PAIS",DATECOL], as_index=False)["yhat_reconc"].sum()
    cmp_fmp_act = g_base_std.merge(g_reco_act, on=["FM_COST_TYPE","PAIS",DATECOL])
    cmp_fmp_act["delta"] = cmp_fmp_act["yhat_reconc"] - cmp_fmp_act["yhat_base"]
    print("\n Dispersión de delta (FM×PAIS mensual) — VARIANTE SOLO ACTIVOS:")
    print(cmp_fmp_act["delta"].describe())

# 2) Gap contra objetivo ANUAL (preferimos ANCLA usada). Leemos anclas exportadas en el Paso 12.
path_fmp_targets = os.path.join(RUTA_METRICAS, "target_fmp_month_step12.csv")

ann_base = g_base_std.groupby(["FM_COST_TYPE","PAIS"], as_index=False)["yhat_base"] \
                     .sum().rename(columns={"yhat_base":"y_base_annual"})
ann_reco_std = g_reco_std.groupby(["FM_COST_TYPE","PAIS"], as_index=False)["yhat_reconc"] \
                         .sum().rename(columns={"yhat_reconc":"y_reco_annual_std"})
cmp_ann = ann_base.merge(ann_reco_std, on=["FM_COST_TYPE","PAIS"], how="outer")

if active_exists:
    ann_reco_act = g_reco_act.groupby(["FM_COST_TYPE","PAIS"], as_index=False)["yhat_reconc"] \
                             .sum().rename(columns={"yhat_reconc":"y_reco_annual_active"})
    cmp_ann = cmp_ann.merge(ann_reco_act, on=["FM_COST_TYPE","PAIS"], how="left")

use_ipc_factor = False
if os.path.exists(path_fmp_targets):
    t = pd.read_csv(path_fmp_targets, sep=CSV_SEP, parse_dates=[DATECOL])
    # El export del Paso 12 tiene columnas: nivel, clave, FECHA, target
    expected = {"nivel","clave",DATECOL,"target"}
    if expected.issubset(set(t.columns)):
        t = t[t["nivel"].astype(str).eq("FM_COST_TYPE|PAIS")].copy()
        # reconstruimos FM_COST_TYPE y PAIS desde 'clave'
        parts = t["clave"].astype(str).str.split("|", n=1, expand=True)
        t["FM_COST_TYPE"] = parts[0]
        t["PAIS"] = parts[1]
        ann_target = t.groupby(["FM_COST_TYPE","PAIS"], as_index=False)["target"].sum() \
                      .rename(columns={"target":"target_annual"})
        cmp_ann = cmp_ann.merge(ann_target, on=["FM_COST_TYPE","PAIS"], how="left")
        # Gaps vs ancla usada
        cmp_ann["gap_std_vs_target"] = cmp_ann["y_reco_annual_std"] - cmp_ann["target_annual"]
        cmp_ann["gap_std_rel_%"]     = 100.0 * cmp_ann["gap_std_vs_target"] / cmp_ann["target_annual"].replace(0, np.nan)
        if active_exists:
            cmp_ann["gap_active_vs_target"] = cmp_ann["y_reco_annual_active"] - cmp_ann["target_annual"]
            cmp_ann["gap_active_rel_%"]     = 100.0 * cmp_ann["gap_active_vs_target"] / cmp_ann["target_annual"].replace(0, np.nan)

        print("\n Top 10 gaps relativos FM×PAIS (anual) — vs ANCLA real (VARIANTE ESTÁNDAR):")
        print(cmp_ann.sort_values("gap_std_rel_%", ascending=False).head(10)
                 [["FM_COST_TYPE","PAIS","y_reco_annual_std","target_annual","gap_std_rel_%"]])

        if active_exists:
            print("\n Top 10 gaps relativos FM×PAIS (anual) — vs ANCLA real (VARIANTE SOLO ACTIVOS):")
            print(cmp_ann.sort_values("gap_active_rel_%", ascending=False).head(10)
                     [["FM_COST_TYPE","PAIS","y_reco_annual_active","target_annual","gap_active_rel_%"]])
    else:
        print("\n[AVISO] target_fmp_month_step12.csv no tiene el esquema esperado; haremos fallback a base×IPC.")
        use_ipc_factor = True
else:
    use_ipc_factor = True

# Fallback: aproximación base×factor (último IPC por país)
if use_ipc_factor:
    ipc_latest = (ipc.sort_values(["PAIS","FECHA"])
                    .groupby("PAIS", as_index=False).tail(1)[["PAIS","IPC_YOY_PCT"]])
    ipc_latest["factor"] = 1.0 + pd.to_numeric(ipc_latest["IPC_YOY_PCT"], errors="coerce").fillna(0.0)/100.0
    cmp_ann = cmp_ann.merge(ipc_latest[["PAIS","factor"]], on="PAIS", how="left").fillna({"factor":1.0})
    cmp_ann["target_annual_aprox"] = cmp_ann["y_base_annual"] * cmp_ann["factor"]

    cmp_ann["gap_std_vs_target"] = cmp_ann["y_reco_annual_std"] - cmp_ann["target_annual_aprox"]
    cmp_ann["gap_std_rel_%"]     = 100.0 * cmp_ann["gap_std_vs_target"] / cmp_ann["target_annual_aprox"].replace(0, np.nan)

    cols_show = ["FM_COST_TYPE","PAIS","y_reco_annual_std","target_annual_aprox","gap_std_rel_%"]
    print("\n Top 10 gaps relativos FM×PAIS (anual) — referencia con target ≈ base×IPC (STD):")
    print(cmp_ann.sort_values("gap_std_rel_%", ascending=False).head(10)[cols_show])

    if active_exists:
        cmp_ann["gap_active_vs_target"] = cmp_ann["y_reco_annual_active"] - cmp_ann["target_annual_aprox"]
        cmp_ann["gap_active_rel_%"]     = 100.0 * cmp_ann["gap_active_vs_target"] / cmp_ann["target_annual_aprox"].replace(0, np.nan)
        cols_show_act = ["FM_COST_TYPE","PAIS","y_reco_annual_active","target_annual_aprox","gap_active_rel_%"]
        print("\n Top 10 gaps relativos FM×PAIS (anual) — referencia con target ≈ base×IPC (ACTIVE):")
        print(cmp_ann.sort_values("gap_active_rel_%", ascending=False).head(10)[cols_show_act])

# 3) Coherencia portfolio: diferencia mensual total antes vs después (STD y ACTIVE)
port_std_b = std.groupby(DATECOL, as_index=False)["yhat_base"].sum().rename(columns={"yhat_base":"y_base"})
port_std_r = std.groupby(DATECOL, as_index=False)["yhat_reconc"].sum().rename(columns={"yhat_reconc":"y_std"})
port = port_std_b.merge(port_std_r, on=DATECOL)
port["delta_std"] = port["y_std"] - port["y_base"]

print("\n Delta mensual a nivel portfolio (STD — reconc - base):")
print(port[[DATECOL, "y_base", "y_std", "delta_std"]])

if active_exists:
    port_act_r = act.groupby(DATECOL, as_index=False)["yhat_reconc"].sum().rename(columns={"yhat_reconc":"y_active"})
    port = port.merge(port_act_r, on=DATECOL, how="left")
    port["delta_active"] = port["y_active"] - port["y_base"]
    print("\n Delta mensual a nivel portfolio (ACTIVE — reconc - base):")
    print(port[[DATECOL, "y_base", "y_active", "delta_active"]])

# 4) Salud numérica: ¿hay negativos tras clip? (no debería)
neg_std = int((std["yhat_reconc"] < 0).sum())
print(f"\n Valores negativos en yhat_reconc (STD): {neg_std} (esperado 0)")

if active_exists:
    neg_act = int((act["yhat_reconc"] < 0).sum())
    print(f" Valores negativos en yhat_reconc (ACTIVE): {neg_act} (esperado 0)")

# 5) (Opcional) Diagnóstico “antes del clip” si guardaste los BEFORE
if os.path.exists(PATH_STD_BEFORE):
    std_bef = pd.read_csv(PATH_STD_BEFORE, sep=CSV_SEP, parse_dates=[DATECOL]).rename(columns={"yhat_reconc":"yhat_before"})
    chk = std.merge(std_bef[[*PAIR_COLS, DATECOL, "yhat_before"]], on=[*PAIR_COLS, DATECOL], how="left")
    was_neg = int((chk["yhat_before"] < 0).sum())
    trunc_to0 = int(((chk["yhat_before"] < 0) & (chk["yhat_reconc"] == 0)).sum())
    print(f"\n STD — Negativos ANTES del clip: {was_neg} | Truncados a 0: {trunc_to0}")

if active_exists and os.path.exists(PATH_ACT_BEFORE):
    act_bef = pd.read_csv(PATH_ACT_BEFORE, sep=CSV_SEP, parse_dates=[DATECOL]).rename(columns={"yhat_reconc":"yhat_before"})
    chk2 = act.merge(act_bef[[*PAIR_COLS, DATECOL, "yhat_before"]], on=[*PAIR_COLS, DATECOL], how="left")
    was_neg2 = int((chk2["yhat_before"] < 0).sum())
    trunc_to0_2 = int(((chk2["yhat_before"] < 0) & (chk2["yhat_reconc"] == 0)).sum())
    print(f" ACTIVE — Negativos ANTES del clip: {was_neg2} | Truncados a 0: {trunc_to0_2}")

 Dispersión de delta (FM×PAIS mensual) — VARIANTE ESTÁNDAR:
count       588.000000
mean      16947.409186
std       44442.971264
min      -37718.528090
25%           0.346510
50%          49.759538
75%        3652.596253
max      213146.042284
Name: delta, dtype: float64

 Dispersión de delta (FM×PAIS mensual) — VARIANTE SOLO ACTIVOS:
count       588.000000
mean       8102.947618
std       25430.284757
min       -7013.019893
25%           0.000000
50%          16.183798
75%        1238.507493
max      159552.082655
Name: delta, dtype: float64

 Top 10 gaps relativos FM×PAIS (anual) — vs ANCLA real (VARIANTE ESTÁNDAR):
             FM_COST_TYPE                  PAIS  y_reco_annual_std  \
6               Licencias                MÉXICO       8.547584e+05
5               Licencias                ESPAÑA       7.265202e+05
37        Servicios Extra                ESPAÑA       1.282928e+06
10        Mtto. Contratos                ESPAÑA       2.391161e+06
12        Mtto. Contratos                PANAMÁ       4.588692e+05
0   Eficiencia Energética                ESPAÑA       2.221276e+05
8         Mtto. Contratos              COLOMBIA       1.167514e+04
24                  Obras                ESPAÑA       8.321506e+05
30        Servicios Ctto.                ESPAÑA       4.524549e+06
21       Mtto. Correctivo  REPÚBLICA DOMINICANA       2.960966e+05

    target_annual  gap_std_rel_%
6    1.505561e+04    5577.340300
5    8.879993e+04     718.154012
37   2.455461e+05     422.479645
10   5.724127e+05     317.733778
12   1.201991e+05     281.757487
0    7.648618e+04     190.415298
8    4.153047e+03     181.122264
24   3.049112e+05     172.915760
30   2.652928e+06      70.549248
21   2.105931e+05      40.601313

 Top 10 gaps relativos FM×PAIS (anual) — vs ANCLA real (VARIANTE SOLO ACTIVOS):
             FM_COST_TYPE                  PAIS  y_reco_annual_active  \
12        Mtto. Contratos                PANAMÁ          5.773527e+05
10        Mtto. Contratos                ESPAÑA          2.301954e+06
37        Servicios Extra                ESPAÑA          9.638202e+05
6               Licencias                MÉXICO          3.850800e+04
8         Mtto. Contratos              COLOMBIA          9.922387e+03
22                  Obras              COLOMBIA          4.203225e+04
21       Mtto. Correctivo  REPÚBLICA DOMINICANA          2.952839e+05
30        Servicios Ctto.                ESPAÑA          3.538369e+06
14        Mtto. Contratos  REPÚBLICA DOMINICANA          9.488495e+03
2   Eficiencia Energética                PANAMÁ          1.343189e+05

    target_annual  gap_active_rel_%
12   1.201991e+05        380.330097
10   5.724127e+05        302.149273
37   2.455461e+05        292.521063
6    1.505561e+04        155.771688
8    4.153047e+03        138.918219
22   1.896421e+04        121.639886
21   2.105931e+05         40.215379
30   2.652928e+06         33.375992
14   7.218467e+03         31.447505
2    1.083357e+05         23.983943

 Delta mensual a nivel portfolio (STD — reconc - base):
        FECHA        y_base         y_std      delta_std
0  2024-01-01  3.026522e+06  3.862409e+06  835887.313741
1  2024-02-01  2.960127e+06  3.770082e+06  809955.190499
2  2024-03-01  3.008443e+06  3.825553e+06  817110.184307
3  2024-04-01  3.016421e+06  3.832718e+06  816297.027888
4  2024-05-01  3.049515e+06  3.875655e+06  826140.266870
5  2024-06-01  3.092858e+06  3.921584e+06  828726.012685
6  2024-07-01  3.012608e+06  3.826037e+06  813429.020084
7  2024-08-01  2.990727e+06  3.780875e+06  790147.478770
8  2024-09-01  3.187771e+06  4.045665e+06  857893.927492
9  2024-10-01  3.314882e+06  4.188240e+06  873357.882809
10 2024-11-01  3.306467e+06  4.177553e+06  871086.541225
11 2024-12-01  3.228334e+06  4.053380e+06  825045.754716

 Delta mensual a nivel portfolio (ACTIVE — reconc - base):
        FECHA        y_base      y_active   delta_active
0  2024-01-01  3.026522e+06  3.429791e+06  403269.143230
1  2024-02-01  2.960127e+06  3.347408e+06  387280.831051
2  2024-03-01  3.008443e+06  3.398769e+06  390326.252737
3  2024-04-01  3.016421e+06  3.413974e+06  397552.871186
4  2024-05-01  3.049515e+06  3.495270e+06  445754.815138
5  2024-06-01  3.092858e+06  3.473227e+06  380368.788921
6  2024-07-01  3.012608e+06  3.392821e+06  380213.515045
7  2024-08-01  2.990727e+06  3.335106e+06  344378.276031
8  2024-09-01  3.187771e+06  3.592323e+06  404552.192622
9  2024-10-01  3.314882e+06  3.775488e+06  460605.349142
10 2024-11-01  3.306467e+06  3.709828e+06  403361.697098
11 2024-12-01  3.228334e+06  3.595203e+06  366869.467199

 Valores negativos en yhat_reconc (STD): 0 (esperado 0)
 Valores negativos en yhat_reconc (ACTIVE): 0 (esperado 0)

 STD — Negativos ANTES del clip: 2684 | Truncados a 0: 2684
 ACTIVE — Negativos ANTES del clip: 928 | Truncados a 0: 928

# Ruta del fichero base
PATH_PREDS = os.path.join(RUTA_RESULTADOS, "preds_por_serie_2024.csv")

# Cargamos
df_base = pd.read_csv(PATH_PREDS, sep=CSV_SEP, parse_dates=[DATECOL])

ZERO_THR = 1e-9  # mismo umbral que en ACTIVE

# Total de filas
total = len(df_base)

# Filas con base ≈ 0
zeros = (df_base["yhat_combo"].abs() <= ZERO_THR).sum()

# Porcentaje
pct = 100.0 * zeros / total

print(f"[Info] Total filas: {total}")
print(f"[Info] Filas con base=0 (≈0 con umbral {ZERO_THR}): {zeros} ({pct:.2f}%)")

# --- Por FM_COST_TYPE ---
g_fm = (df_base.assign(is_zero=(df_base["yhat_combo"].abs() <= ZERO_THR))
        .groupby("FM_COST_TYPE")["is_zero"]
        .agg(["sum","count"])
        .reset_index())
g_fm["pct_zero_%"] = 100.0 * g_fm["sum"] / g_fm["count"]

print("\n[Info] % de pares base=0 por FM_COST_TYPE")
print(g_fm.sort_values("pct_zero_%", ascending=False))


# --- Por PAIS ---
# Necesitamos unir con dim_buildings para conocer el país
dim = pd.read_csv(PATH_DIM, sep=CSV_SEP)
df_with_pais = df_base.merge(dim[["ID_BUILDING","PAIS"]], on="ID_BUILDING", how="left")

g_pais = (df_with_pais.assign(is_zero=(df_with_pais["yhat_combo"].abs() <= ZERO_THR))
          .groupby("PAIS")["is_zero"]
          .agg(["sum","count"])
          .reset_index())
g_pais["pct_zero_%"] = 100.0 * g_pais["sum"] / g_pais["count"]

print("\n[Info] % de pares base=0 por PAIS")
print(g_pais.sort_values("pct_zero_%", ascending=False))

[Info] Total filas: 29148
[Info] Filas con base=0 (≈0 con umbral 1e-09): 7877 (27.02%)

[Info] % de pares base=0 por FM_COST_TYPE
            FM_COST_TYPE   sum  count  pct_zero_%
1              Licencias  2552   3096   82.428941
4                  Obras   695   1080   64.351852
0  Eficiencia Energética   381    840   45.357143
6        Servicios Extra  1808   4512   40.070922
2        Mtto. Contratos  1010   4380   23.059361
3       Mtto. Correctivo   764   5364   14.243102
5        Servicios Ctto.   348   4056    8.579882
7            Suministros   319   5820    5.481100

[Info] % de pares base=0 por PAIS
                   PAIS   sum  count  pct_zero_%
2                ESPAÑA  6697  18924   35.388924
5                  PERÚ   371   1392   26.652299
3                MÉXICO   412   2304   17.881944
1            COSTA RICA    41    588    6.972789
4                PANAMÁ   181   2736    6.615497
6  REPÚBLICA DOMINICANA    18    300    6.000000
0              COLOMBIA   157   2904    5.406336

# ============================================================
# Comparativa base vs reconciliados: Paso 10, 11 y 12 (STD/ACTIVE)
# Lee:
#   RESULTADOS/preds_reconciliadas_2024.csv           (Paso 10)
#   RESULTADOS/preds_reconciliadas_2024_prod.csv      (Paso 11)
#   RESULTADOS/preds_reconciliadas_step12.csv         (Paso 12 STD)
#   RESULTADOS/preds_reconciliadas_step12_activeonly.csv (Paso 12 ACTIVE)
# Une por (ID_BUILDING, FM_COST_TYPE, FECHA)
# Calcula deltas contra base y agrega:
#   * Portfolio mensual
#   * FM_COST_TYPE × PAIS mensual
#   * Dispersión bottom
# Exporta CSVs comparativos en REPORTING/
# ============================================================


# ----- Rutas y constantes
ruta_base_3 = "/content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3"
CSV_SEP     = ";"

RUTA_RESULTADOS = os.path.join(ruta_base_3, "RESULTADOS")
RUTA_METRICAS   = os.path.join(ruta_base_3, "METRICAS")
RUTA_REPORTING  = os.path.join(ruta_base_3, "REPORTING")
os.makedirs(RUTA_REPORTING, exist_ok=True)

PATH_P10       = os.path.join(RUTA_RESULTADOS, "preds_reconciliadas_2024.csv")
PATH_P11       = os.path.join(RUTA_RESULTADOS, "preds_reconciliadas_2024_prod.csv")
PATH_P12_STD   = os.path.join(RUTA_RESULTADOS, "preds_reconciliadas_step12.csv")
PATH_P12_ACT   = os.path.join(RUTA_RESULTADOS, "preds_reconciliadas_step12_activeonly.csv")
PATH_DIM       = os.path.join(RUTA_METRICAS,   "dim_buildings.csv")

PAIR_COLS = ["ID_BUILDING", "FM_COST_TYPE"]
DATECOL   = "FECHA"

# ----- Helper de lectura
def _read_preds(path, tag):
    if not os.path.exists(path):
        print(f"[AVISO] Falta archivo de {tag}: {path}")
        return pd.DataFrame(columns=[*PAIR_COLS, DATECOL, f"yhat_base_{tag}", f"yhat_reconc_{tag}"])
    df = pd.read_csv(path, sep=CSV_SEP, parse_dates=[DATECOL])
    need = [*PAIR_COLS, DATECOL, "yhat_base", "yhat_reconc"]
    miss = [c for c in need if c not in df.columns]
    if miss:
        raise KeyError(f"[{tag}] faltan columnas {miss} en {os.path.basename(path)}")
    df = df[need].rename(columns={
        "yhat_base":   f"yhat_base_{tag}",
        "yhat_reconc": f"yhat_reconc_{tag}"
    })
    return df

# ----- Carga
p10     = _read_preds(PATH_P10, "p10")
p11     = _read_preds(PATH_P11, "p11")
p12std  = _read_preds(PATH_P12_STD, "p12std")
p12act  = _read_preds(PATH_P12_ACT, "p12act")

if all(len(d)==0 for d in [p10, p11, p12std, p12act]):
    raise FileNotFoundError("No se encontró ningún output de reconciliación (p10/p11/p12std/p12act).")

from functools import reduce
dfs = [d for d in [p10, p11, p12std, p12act] if len(d) > 0]
df_cmp = reduce(lambda a,b: pd.merge(a, b, on=PAIR_COLS+[DATECOL], how="inner"), dfs)

if df_cmp.empty:
    raise ValueError("La intersección entre p10/p11/p12std/p12act es vacía. Revisa que compartan las mismas llaves y meses.")

# ----- Verificación suave de la base
base_cols = [c for c in df_cmp.columns if c.startswith("yhat_base_")]
if base_cols:
    base_stack = df_cmp[base_cols].astype(float)
    drift    = float((base_stack.max(axis=1) - base_stack.min(axis=1)).abs().max())
    print(f"[Check] Discrepancia máxima entre 'yhat_base_*' = {drift:,.6f}")
    base_ref_col = "yhat_base_p10" if "yhat_base_p10" in base_cols else base_cols[0]
else:
    raise KeyError("No se encontró ninguna columna de base en los archivos de entrada.")

df_cmp["yhat_base_ref"] = df_cmp[base_ref_col].astype(float)

# ----- Deltas vs base
for tag in ["p10","p11","p12std","p12act"]:
    yrec = f"yhat_reconc_{tag}"
    if yrec in df_cmp.columns:
        df_cmp[f"delta_{tag}"] = df_cmp[yrec].astype(float) - df_cmp["yhat_base_ref"]

# ----- Exporte bottom comparativo
BOTTOM_OUT = os.path.join(RUTA_REPORTING, "compare_recon_steps_bottom.csv")
keep_bottom_cols = [*PAIR_COLS, DATECOL, "yhat_base_ref"] + \
                   [c for c in [
                       "yhat_reconc_p10","yhat_reconc_p11","yhat_reconc_p12std","yhat_reconc_p12act",
                       "delta_p10","delta_p11","delta_p12std","delta_p12act"
                   ] if c in df_cmp.columns]
df_cmp[keep_bottom_cols].sort_values(PAIR_COLS+[DATECOL]).to_csv(BOTTOM_OUT, sep=CSV_SEP, index=False)
print(f"[OK] Guardado bottom comparativo: {BOTTOM_OUT} ({len(df_cmp)} filas)")

# ----- Portfolio mensual
agg_parts = {
    "y_base": ("yhat_base_ref", "sum")
}
if "yhat_reconc_p10"   in df_cmp.columns: agg_parts["y_p10"]    = ("yhat_reconc_p10",   "sum")
if "yhat_reconc_p11"   in df_cmp.columns: agg_parts["y_p11"]    = ("yhat_reconc_p11",   "sum")
if "yhat_reconc_p12std" in df_cmp.columns: agg_parts["y_p12std"] = ("yhat_reconc_p12std","sum")
if "yhat_reconc_p12act" in df_cmp.columns: agg_parts["y_p12act"] = ("yhat_reconc_p12act","sum")

agg = df_cmp.groupby(DATECOL).agg(**agg_parts).reset_index()
for col in ["y_p10","y_p11","y_p12std","y_p12act"]:
    if col in agg.columns:
        agg[f"delta_{col.split('_')[-1]}"] = agg[col] - agg["y_base"]

PORT_OUT = os.path.join(RUTA_REPORTING, "compare_recon_steps_portfolio.csv")
agg.sort_values(DATECOL).to_csv(PORT_OUT, sep=CSV_SEP, index=False)
print(f"[OK] Guardado portfolio mensual: {PORT_OUT}")

# ----- FM_COST_TYPE × PAIS mensual
SEG_OUT = os.path.join(RUTA_REPORTING, "compare_recon_steps_fmcost_pais.csv")
if os.path.exists(PATH_DIM):
    dim = pd.read_csv(PATH_DIM, sep=CSV_SEP)
    # Asegura columna PAIS
    if "PAIS" not in dim.columns:
        up = {c.upper(): c for c in dim.columns}
        if "COUNTRY" in up:      dim = dim.rename(columns={up["COUNTRY"]:"PAIS"})
        if "COUNTRY_DEF" in up:  dim = dim.rename(columns={up["COUNTRY_DEF"]:"PAIS"})
    if "PAIS" in dim.columns:
        df_seg = df_cmp.merge(dim[["ID_BUILDING","PAIS"]], on="ID_BUILDING", how="left")
        grp_cols = ["FM_COST_TYPE","PAIS", DATECOL]

        seg_parts = {
            "y_base": ("yhat_base_ref", "sum")
        }
        if "yhat_reconc_p10"    in df_seg.columns: seg_parts["y_p10"]    = ("yhat_reconc_p10",    "sum")
        if "yhat_reconc_p11"    in df_seg.columns: seg_parts["y_p11"]    = ("yhat_reconc_p11",    "sum")
        if "yhat_reconc_p12std" in df_seg.columns: seg_parts["y_p12std"] = ("yhat_reconc_p12std", "sum")
        if "yhat_reconc_p12act" in df_seg.columns: seg_parts["y_p12act"] = ("yhat_reconc_p12act", "sum")

        seg = df_seg.groupby(grp_cols).agg(**seg_parts).reset_index()
        for col in ["y_p10","y_p11","y_p12std","y_p12act"]:
            if col in seg.columns:
                seg[f"delta_{col.split('_')[-1]}"] = seg[col] - seg["y_base"]

        seg.sort_values(grp_cols).to_csv(SEG_OUT, sep=CSV_SEP, index=False)
        print(f"[OK] Guardado FM_COST_TYPE×PAIS mensual: {SEG_OUT}")
    else:
        print("[AVISO] dim_buildings.csv no tiene columna PAIS. Se omite la vista FM_COST_TYPE×PAIS.")
else:
    print("[AVISO] No se encontró dim_buildings.csv. Se omite la vista FM_COST_TYPE×PAIS.")

# ----- Dispersión de deltas en bottom
stats_cols = [c for c in ["delta_p10","delta_p11","delta_p12std","delta_p12act"] if c in df_cmp.columns]
if stats_cols:
    print("\n[Resumen] Dispersión de deltas (reconc - base) en bottom:")
    for c in stats_cols:
        desc = df_cmp[c].describe(percentiles=[.01,.1,.25,.5,.75,.9,.99])
        print(f"\n{c}:\n{desc}")

# ----- Vista rápida en consola
print("\n[Muestra portfolio mensual]")
print(agg.sort_values(DATECOL).head(12))

print("\n[Muestra bottom comparativo]")
print(df_cmp[keep_bottom_cols].sort_values(PAIR_COLS+[DATECOL]).head(10))

[Check] Discrepancia máxima entre 'yhat_base_*' = 0.000000
[OK] Guardado bottom comparativo: /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3/REPORTING/compare_recon_steps_bottom.csv (29148 filas)
[OK] Guardado portfolio mensual: /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3/REPORTING/compare_recon_steps_portfolio.csv
[OK] Guardado FM_COST_TYPE×PAIS mensual: /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3/REPORTING/compare_recon_steps_fmcost_pais.csv

[Resumen] Dispersión de deltas (reconc - base) en bottom:

delta_p10:
count     29148.000000
mean        109.232172
std        2373.323362
min      -24584.321052
1%         -989.374110
10%          -4.352855
25%           0.000000
50%           0.304418
75%           5.653931
90%          50.166252
99%        2283.953159
max      171658.137550
Name: delta_p10, dtype: float64

delta_p11:
count    29148.0
mean         0.0
std          0.0
min          0.0
1%           0.0
10%          0.0
25%          0.0
50%          0.0
75%          0.0
90%          0.0
99%          0.0
max          0.0
Name: delta_p11, dtype: float64

delta_p12std:
count     29148.000000
mean        341.878571
std        5197.783342
min      -16859.044681
1%         -719.412721
10%         -38.053878
25%          -0.166144
50%           0.478442
75%          20.571566
90%         136.489526
99%        5141.801808
max      215342.684158
Name: delta_p12std, dtype: float64

delta_p12act:
count    29148.000000
mean       163.460038
std       2237.847041
min     -16859.044681
1%        -550.606555
10%         -9.741310
25%          0.000000
50%          0.000525
75%          3.324407
90%         59.445819
99%       3207.840710
max      84055.835845
Name: delta_p12act, dtype: float64

[Muestra portfolio mensual]
        FECHA        y_base         y_p10         y_p11      y_p12std  \
0  2024-01-01  3.026522e+06  3.250155e+06  3.026522e+06  3.862409e+06
1  2024-02-01  2.960127e+06  3.082621e+06  2.960127e+06  3.770082e+06
2  2024-03-01  3.008443e+06  3.049105e+06  3.008443e+06  3.825553e+06
3  2024-04-01  3.016421e+06  3.176957e+06  3.016421e+06  3.832718e+06
4  2024-05-01  3.049515e+06  2.989022e+06  3.049515e+06  3.875655e+06
5  2024-06-01  3.092858e+06  2.977756e+06  3.092858e+06  3.921584e+06
6  2024-07-01  3.012608e+06  2.931273e+06  3.012608e+06  3.826037e+06
7  2024-08-01  2.990727e+06  3.086346e+06  2.990727e+06  3.780875e+06
8  2024-09-01  3.187771e+06  4.557699e+06  3.187771e+06  4.045665e+06
9  2024-10-01  3.314882e+06  4.807961e+06  3.314882e+06  4.188240e+06
10 2024-11-01  3.306467e+06  3.322896e+06  3.306467e+06  4.177553e+06
11 2024-12-01  3.228334e+06  3.146782e+06  3.228334e+06  4.053380e+06

        y_p12act     delta_p10  delta_p11   delta_p12std   delta_p12act
0   3.429791e+06  2.236336e+05        0.0  835887.313741  403269.143230
1   3.347408e+06  1.224942e+05        0.0  809955.190499  387280.831051
2   3.398769e+06  4.066254e+04        0.0  817110.184307  390326.252737
3   3.413974e+06  1.605364e+05        0.0  816297.027888  397552.871186
4   3.495270e+06 -6.049325e+04        0.0  826140.266870  445754.815138
5   3.473227e+06 -1.151019e+05        0.0  828726.012685  380368.788921
6   3.392821e+06 -8.133424e+04        0.0  813429.020084  380213.515045
7   3.335106e+06  9.561901e+04        0.0  790147.478770  344378.276031
8   3.592323e+06  1.369928e+06        0.0  857893.927492  404552.192622
9   3.775488e+06  1.493078e+06        0.0  873357.882809  460605.349142
10  3.709828e+06  1.642908e+04        0.0  871086.541225  403361.697098
11  3.595203e+06 -8.155242e+04        0.0  825045.754716  366869.467199

[Muestra bottom comparativo]
   ID_BUILDING           FM_COST_TYPE      FECHA  yhat_base_ref  \
0            2  Eficiencia Energética 2024-01-01            0.0
1            2  Eficiencia Energética 2024-02-01            0.0
2            2  Eficiencia Energética 2024-03-01            0.0
3            2  Eficiencia Energética 2024-04-01            0.0
4            2  Eficiencia Energética 2024-05-01            0.0
5            2  Eficiencia Energética 2024-06-01            0.0
6            2  Eficiencia Energética 2024-07-01            0.0
7            2  Eficiencia Energética 2024-08-01            0.0
8            2  Eficiencia Energética 2024-09-01            0.0
9            2  Eficiencia Energética 2024-10-01            0.0

   yhat_reconc_p10  yhat_reconc_p11  yhat_reconc_p12std  yhat_reconc_p12act  \
0     0.000000e+00              0.0         7901.969069                 0.0
1     0.000000e+00              0.0         7665.118813                 0.0
2     0.000000e+00              0.0         7885.690175                 0.0
3     0.000000e+00              0.0         7820.281633                 0.0
4     1.342612e-09              0.0         7839.947659                 0.0
5     0.000000e+00              0.0         7999.217869                 0.0
6     3.243058e-10              0.0         7775.350421                 0.0
7     0.000000e+00              0.0         7798.841951                 0.0
8     0.000000e+00              0.0         8159.586637                 0.0
9     0.000000e+00              0.0         8683.304007                 0.0

      delta_p10  delta_p11  delta_p12std  delta_p12act
0  0.000000e+00        0.0   7901.969069           0.0
1  0.000000e+00        0.0   7665.118813           0.0
2  0.000000e+00        0.0   7885.690175           0.0
3  0.000000e+00        0.0   7820.281633           0.0
4  1.342612e-09        0.0   7839.947659           0.0
5  0.000000e+00        0.0   7999.217869           0.0
6  3.243058e-10        0.0   7775.350421           0.0
7  0.000000e+00        0.0   7798.841951           0.0
8  0.000000e+00        0.0   8159.586637           0.0
9  0.000000e+00        0.0   8683.304007           0.0

# ============================================================
# Gráficos comparativos Paso 10, 11 y 12 (STD y ACTIVE)
# ============================================================

FIG_DIR = os.path.join(RUTA_REPORTING, "figs_compare_recon")
os.makedirs(FIG_DIR, exist_ok=True)

# --- 1) Serie temporal portfolio ---
plt.figure(figsize=(12,6))
plt.plot(agg[DATECOL], agg["y_base"], label="Base", color="black", linewidth=2)

if "y_p10" in agg.columns:
    plt.plot(agg[DATECOL], agg["y_p10"], label="Reconc P10", linestyle="--")
if "y_p11" in agg.columns:
    plt.plot(agg[DATECOL], agg["y_p11"], label="Reconc P11", linestyle="--")

# Paso 12: soporta columna única 'y_p12' o las dos variantes 'y_p12std' y 'y_p12act'
if "y_p12std" in agg.columns or "y_p12act" in agg.columns:
    if "y_p12std" in agg.columns:
        plt.plot(agg[DATECOL], agg["y_p12std"], label="Reconc P12 STD", linestyle="--")
    if "y_p12act" in agg.columns:
        plt.plot(agg[DATECOL], agg["y_p12act"], label="Reconc P12 ACTIVE", linestyle="-.")
elif "y_p12" in agg.columns:
    plt.plot(agg[DATECOL], agg["y_p12"], label="Reconc P12", linestyle="--")

plt.title("Serie temporal portfolio — Base vs Reconciliados")
plt.xlabel("Mes")
plt.ylabel("Suma mensual")
plt.legend()
plt.tight_layout()
plt.savefig(os.path.join(FIG_DIR, "portfolio_series.png"))
plt.close()

# --- 2) Boxplots de deltas en bottom ---
plt.figure(figsize=(9,6))
deltas_data, labels = [], []

for c, lab in [
    ("delta_p10",     "Paso 10"),
    ("delta_p11",     "Paso 11"),
    ("delta_p12",     "Paso 12"),
    ("delta_p12std",  "Paso 12 STD"),
    ("delta_p12act",  "Paso 12 ACTIVE"),
]:
    if c in df_cmp.columns:
        deltas_data.append(df_cmp[c].values)
        labels.append(lab)

if deltas_data:
    plt.boxplot(deltas_data, labels=labels, showfliers=False)
    plt.axhline(0, color="grey", linestyle=":")
    plt.title("Distribución de deltas bottom (reconc - base)")
    plt.ylabel("Delta")
    plt.tight_layout()
    plt.savefig(os.path.join(FIG_DIR, "bottom_deltas_boxplot.png"))
    plt.close()
else:
    print("[AVISO] No hay columnas de delta para boxplot.")

# --- 3) Heatmaps anuales FM_COST_TYPE×PAIS (opcional) ---
#    Genera hasta dos heatmaps: P12 STD y P12 ACTIVE (si existen).
if os.path.exists(PATH_DIM):
    dim = pd.read_csv(PATH_DIM, sep=CSV_SEP) if 'dim' not in globals() else dim
if os.path.exists(PATH_DIM) and "PAIS" in dim.columns:
    # Construimos dataframe base para agregados
    needed = ["ID_BUILDING","FM_COST_TYPE", DATECOL, "yhat_base_ref",
              "yhat_reconc_p10","yhat_reconc_p11",
              "yhat_reconc_p12","yhat_reconc_p12std","yhat_reconc_p12act"]
    have = [c for c in needed if c in df_cmp.columns]
    df_seg = df_cmp[have].merge(dim[["ID_BUILDING","PAIS"]], on="ID_BUILDING", how="left")

    rename_map = {}
    if "yhat_base_ref"      in df_seg.columns: rename_map["yhat_base_ref"]      = "y_base"
    if "yhat_reconc_p10"    in df_seg.columns: rename_map["yhat_reconc_p10"]    = "y_p10"
    if "yhat_reconc_p11"    in df_seg.columns: rename_map["yhat_reconc_p11"]    = "y_p11"
    if "yhat_reconc_p12"    in df_seg.columns: rename_map["yhat_reconc_p12"]    = "y_p12"
    if "yhat_reconc_p12std" in df_seg.columns: rename_map["yhat_reconc_p12std"] = "y_p12std"
    if "yhat_reconc_p12act" in df_seg.columns: rename_map["yhat_reconc_p12act"] = "y_p12act"
    df_seg = df_seg.rename(columns=rename_map)

    df_seg["ANIO"] = pd.to_datetime(df_seg[DATECOL]).dt.year
    agg_dict = {}
    if "y_base"   in df_seg.columns: agg_dict["base"]   = ("y_base",   "sum")
    if "y_p12"    in df_seg.columns: agg_dict["p12"]    = ("y_p12",    "sum")
    if "y_p12std" in df_seg.columns: agg_dict["p12std"] = ("y_p12std", "sum")
    if "y_p12act" in df_seg.columns: agg_dict["p12act"] = ("y_p12act", "sum")

    if agg_dict:
        seg_ann = (df_seg.groupby(["FM_COST_TYPE","PAIS","ANIO"])
                          .agg(**agg_dict).reset_index())

        # Heatmap P12 STD
        if {"base","p12std"}.issubset(seg_ann.columns):
            seg_ann["gap_rel_p12std"] = 100.0 * (seg_ann["p12std"] - seg_ann["base"]) / seg_ann["base"].replace(0, np.nan)
            pivot_std = seg_ann.pivot_table(index="FM_COST_TYPE", columns="PAIS", values="gap_rel_p12std", aggfunc="mean")
            plt.figure(figsize=(14,6))
            im = plt.imshow(pivot_std.fillna(0).values, cmap="RdBu", aspect="auto", vmin=-100, vmax=100)
            plt.colorbar(im, label="% gap relativo Paso 12 STD vs base (anual)")
            plt.xticks(range(len(pivot_std.columns)), pivot_std.columns, rotation=90)
            plt.yticks(range(len(pivot_std.index)), pivot_std.index)
            plt.title("Heatmap gap relativo Paso 12 STD vs base (anual)")
            plt.tight_layout()
            plt.savefig(os.path.join(FIG_DIR, "heatmap_gap_p12std.png"))
            plt.close()

        # Heatmap P12 ACTIVE
        if {"base","p12act"}.issubset(seg_ann.columns):
            seg_ann["gap_rel_p12act"] = 100.0 * (seg_ann["p12act"] - seg_ann["base"]) / seg_ann["base"].replace(0, np.nan)
            pivot_act = seg_ann.pivot_table(index="FM_COST_TYPE", columns="PAIS", values="gap_rel_p12act", aggfunc="mean")
            plt.figure(figsize=(14,6))
            im = plt.imshow(pivot_act.fillna(0).values, cmap="RdBu", aspect="auto", vmin=-100, vmax=100)
            plt.colorbar(im, label="% gap relativo Paso 12 ACTIVE vs base (anual)")
            plt.xticks(range(len(pivot_act.columns)), pivot_act.columns, rotation=90)
            plt.yticks(range(len(pivot_act.index)), pivot_act.index)
            plt.title("Heatmap gap relativo Paso 12 ACTIVE vs base (anual)")
            plt.tight_layout()
            plt.savefig(os.path.join(FIG_DIR, "heatmap_gap_p12active.png"))
            plt.close()
    else:
        print("[AVISO] No hay columnas suficientes para heatmaps (base/p12std/p12act).")
else:
    print("[AVISO] No se encontró dim_buildings.csv o falta 'PAIS'; se omiten heatmaps.")

print(f"[OK] Gráficos guardados en {FIG_DIR}")

[OK] Gráficos guardados en /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3/REPORTING/figs_compare_recon

# ============================================================
# Visualización comparativa: Base vs Reconciliados (P10, P11, P12 STD/ACTIVE)
# Requiere CSVs:
#  - REPORTING/compare_recon_steps_portfolio.csv
#  - REPORTING/compare_recon_steps_bottom.csv
#  - REPORTING/compare_recon_steps_fmcost_pais.csv
# ============================================================
import os, numpy as np, pandas as pd
import matplotlib.pyplot as plt

# ---- Rutas base (ajusta si difiere) ----
ruta_base_3    = "/content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3"
RUTA_REPORTING = os.path.join(ruta_base_3, "REPORTING")
CSV_SEP        = ";"
DATECOL        = "FECHA"

# ---- Archivos de entrada ----
PATH_PORT   = os.path.join(RUTA_REPORTING, "compare_recon_steps_portfolio.csv")
PATH_BOTTOM = os.path.join(RUTA_REPORTING, "compare_recon_steps_bottom.csv")
PATH_FMPAIS = os.path.join(RUTA_REPORTING, "compare_recon_steps_fmcost_pais.csv")

# ---- Carpeta de figuras ----
FIG_DIR = os.path.join(RUTA_REPORTING, "figs_compare_recon")
os.makedirs(FIG_DIR, exist_ok=True)

# ------------------------------------------------------------------
# 1) Carga de datos (robusta a columnas opcionales y parseo de fechas)
# ------------------------------------------------------------------
def _read_csv_safe(path, parse_dates=None, sep=CSV_SEP):
    if not os.path.exists(path):
        raise FileNotFoundError(f"No se encontró {path}")
    df = pd.read_csv(path, sep=sep, parse_dates=parse_dates)
    df.columns = [str(c).strip() for c in df.columns]
    return df

agg    = _read_csv_safe(PATH_PORT,   parse_dates=[DATECOL])
df_cmp = _read_csv_safe(PATH_BOTTOM, parse_dates=[DATECOL])
df_seg = _read_csv_safe(PATH_FMPAIS, parse_dates=[DATECOL])

# Detectamos columnas presentes
# Portfolio: soporta y_p12 (formato antiguo) o y_p12std/y_p12act (nuevo)
has_p12std = "y_p12std" in agg.columns
has_p12act = "y_p12act" in agg.columns
has_p12    = "y_p12"    in agg.columns

# Bottom deltas
delta_cols_bottom = [c for c in ["delta_p10","delta_p11","delta_p12","delta_p12std","delta_p12act"] if c in df_cmp.columns]

# ============================================================
# 2) Serie temporal portfolio — Base vs Reconciliados
# ============================================================
plt.figure(figsize=(12,6))
if "y_base" in agg.columns:
    plt.plot(agg[DATECOL], agg["y_base"], label="Base", color="black", linewidth=2)
if "y_p10" in agg.columns:
    plt.plot(agg[DATECOL], agg["y_p10"], label="Reconc P10", linestyle="--")
if "y_p11" in agg.columns:
    plt.plot(agg[DATECOL], agg["y_p11"], label="Reconc P11", linestyle="--")

# P12 variantes
if has_p12std or has_p12act:
    if has_p12std:
        plt.plot(agg[DATECOL], agg["y_p12std"], label="Reconc P12 STD", linestyle="--")
    if has_p12act:
        plt.plot(agg[DATECOL], agg["y_p12act"], label="Reconc P12 ACTIVE", linestyle="-.")
elif has_p12:
    plt.plot(agg[DATECOL], agg["y_p12"], label="Reconc P12", linestyle="--")

plt.title("Serie temporal portfolio — Base vs Reconciliados")
plt.xlabel("Mes")
plt.ylabel("Suma mensual")
plt.legend()
plt.tight_layout()
out1 = os.path.join(FIG_DIR, "portfolio_series.png")
plt.savefig(out1, dpi=150)
plt.show()

# ============================================================
# 3) Boxplots de deltas en bottom (reconc - base)
# ============================================================
if len(delta_cols_bottom) > 0:
    plt.figure(figsize=(9,6))
    data   = [df_cmp[c].values for c in delta_cols_bottom]
    labels = []
    for c in delta_cols_bottom:
        if c == "delta_p12std": labels.append("Paso 12 STD")
        elif c == "delta_p12act": labels.append("Paso 12 ACTIVE")
        else: labels.append(c.replace("delta_","Paso "))
    plt.boxplot(data, labels=labels, showfliers=False)
    plt.axhline(0, color="grey", linestyle=":", linewidth=1)
    plt.title("Distribución de deltas bottom (reconc - base)")
    plt.ylabel("Delta")
    plt.tight_layout()
    out2 = os.path.join(FIG_DIR, "bottom_deltas_boxplot.png")
    plt.savefig(out2, dpi=150)
    plt.show()
else:
    print("[AVISO] No se encontraron columnas de delta en bottom para boxplot.")

# ============================================================
# 4) Heatmaps anuales FM_COST_TYPE×PAIS — P12 STD y P12 ACTIVE
# ============================================================
need_cols = {"FM_COST_TYPE","PAIS",DATECOL,"y_base"}
if not need_cols.issubset(df_seg.columns):
    print("[AVISO] El CSV FM_COST_TYPE×PAIS no tiene columnas mínimas para heatmaps.")
else:
    seg = df_seg.copy()
    seg["ANIO"] = pd.to_datetime(seg[DATECOL]).dt.year

    # Mapear nombres si vienen como yhat_reconc_*
    rename_map = {}
    if "yhat_reconc_p12std" in seg.columns: rename_map["yhat_reconc_p12std"] = "y_p12std"
    if "yhat_reconc_p12act" in seg.columns: rename_map["yhat_reconc_p12act"] = "y_p12act"
    if rename_map:
        seg = seg.rename(columns=rename_map)

    agg_dict = {"y_base": "sum"}
    if "y_p12std" in seg.columns: agg_dict["y_p12std"] = "sum"
    if "y_p12act" in seg.columns: agg_dict["y_p12act"] = "sum"
    if "y_p12"    in seg.columns: agg_dict["y_p12"]    = "sum"  # por compatibilidad

    seg_ann = seg.groupby(["FM_COST_TYPE","PAIS","ANIO"], as_index=False).agg(agg_dict)
    last_year = int(seg_ann["ANIO"].max())

    # Heatmap P12 STD
    if {"y_base","y_p12std"}.issubset(seg_ann.columns):
        seg_std = seg_ann[seg_ann["ANIO"]==last_year].copy()
        seg_std["gap_rel_p12std"] = 100.0 * (seg_std["y_p12std"] - seg_std["y_base"]) / seg_std["y_base"].replace(0, np.nan)
        pivot_std = seg_std.pivot_table(index="FM_COST_TYPE", columns="PAIS", values="gap_rel_p12std", aggfunc="mean")
        plt.figure(figsize=(14,6))
        im = plt.imshow(pivot_std.fillna(0).values, cmap="RdBu", aspect="auto", vmin=-100, vmax=100)
        plt.colorbar(im, label=f"% gap relativo P12 STD vs base (anual {last_year})")
        plt.xticks(range(len(pivot_std.columns)), pivot_std.columns, rotation=90)
        plt.yticks(range(len(pivot_std.index)), pivot_std.index)
        plt.title(f"Heatmap gap relativo P12 STD vs base (anual {last_year})")
        plt.tight_layout()
        plt.savefig(os.path.join(FIG_DIR, "heatmap_gap_p12std.png"), dpi=150)
        plt.show()

    # Heatmap P12 ACTIVE
    if {"y_base","y_p12act"}.issubset(seg_ann.columns):
        seg_act = seg_ann[seg_ann["ANIO"]==last_year].copy()
        seg_act["gap_rel_p12act"] = 100.0 * (seg_act["y_p12act"] - seg_act["y_base"]) / seg_act["y_base"].replace(0, np.nan)
        pivot_act = seg_act.pivot_table(index="FM_COST_TYPE", columns="PAIS", values="gap_rel_p12act", aggfunc="mean")
        plt.figure(figsize=(14,6))
        im = plt.imshow(pivot_act.fillna(0).values, cmap="RdBu", aspect="auto", vmin=-100, vmax=100)
        plt.colorbar(im, label=f"% gap relativo P12 ACTIVE vs base (anual {last_year})")
        plt.xticks(range(len(pivot_act.columns)), pivot_act.columns, rotation=90)
        plt.yticks(range(len(pivot_act.index)), pivot_act.index)
        plt.title(f"Heatmap gap relativo P12 ACTIVE vs base (anual {last_year})")
        plt.tight_layout()
        plt.savefig(os.path.join(FIG_DIR, "heatmap_gap_p12active.png"), dpi=150)
        plt.show()

# ============================================================
# 5) Extras: scatter y barras de delta (P12 STD/ACTIVE)
# ============================================================
# Scatter Base vs Reconc por mes
plt.figure(figsize=(6,6))
plotted_any = False
if "y_p10" in agg.columns:
    plt.scatter(agg["y_base"], agg["y_p10"], alpha=0.7, label="P10", marker="o"); plotted_any = True
if "y_p11" in agg.columns:
    plt.scatter(agg["y_base"], agg["y_p11"], alpha=0.7, label="P11", marker="^"); plotted_any = True
if has_p12std:
    plt.scatter(agg["y_base"], agg["y_p12std"], alpha=0.7, label="P12 STD", marker="s"); plotted_any = True
if has_p12act:
    plt.scatter(agg["y_base"], agg["y_p12act"], alpha=0.7, label="P12 ACTIVE", marker="x"); plotted_any = True
elif has_p12:
    plt.scatter(agg["y_base"], agg["y_p12"], alpha=0.7, label="P12", marker="s"); plotted_any = True

if plotted_any:
    mins = [agg["y_base"].min()]
    maxs = [agg["y_base"].max()]
    for c in agg.columns:
        if c.startswith("y_p"):
            mins.append(agg[c].min()); maxs.append(agg[c].max())
    minv, maxv = float(np.nanmin(mins)), float(np.nanmax(maxs))
    plt.plot([minv, maxv], [minv, maxv], color="grey", linestyle=":", linewidth=1)
    plt.xlabel("Portfolio mensual — Base")
    plt.ylabel("Portfolio mensual — Reconc")
    plt.title("Scatter Base vs Reconc (por mes)")
    plt.legend()
    plt.tight_layout()
    plt.savefig(os.path.join(FIG_DIR, "scatter_portfolio_base_vs_recon.png"), dpi=150)
    plt.show()
else:
    plt.close()

# Barras de delta mensuales (si existen)
if has_p12std or has_p12act or has_p12:
    xlab = agg[DATECOL].dt.strftime("%Y-%m")
    plt.figure(figsize=(12,4))
    width = 0.35
    x = np.arange(len(xlab))
    plotted_any = False

    if has_p12std:
        plt.bar(x - width/2, (agg["y_p12std"] - agg["y_base"]).values, width=width, label="P12 STD - Base"); plotted_any = True
    if has_p12act:
        plt.bar(x + width/2, (agg["y_p12act"] - agg["y_base"]).values, width=width, label="P12 ACTIVE - Base"); plotted_any = True
    if (not has_p12std and not has_p12act) and has_p12:
        plt.bar(x, (agg["y_p12"] - agg["y_base"]).values, width=0.6, label="P12 - Base"); plotted_any = True

    if plotted_any:
        plt.xticks(x, xlab, rotation=45, ha="right")
        plt.title("Delta mensual portfolio (P12 - Base)")
        plt.ylabel("Delta")
        plt.tight_layout()
        plt.savefig(os.path.join(FIG_DIR, "bar_delta_portfolio_p12.png"), dpi=150)
        plt.show()
    else:
        plt.close()

print(f"[OK] Gráficos guardados en: {FIG_DIR}")

[OK] Gráficos guardados en: /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3/REPORTING/figs_compare_recon

# ================================================
# Exploración gráfica: base==0  y  reconciliado>0 (Paso 12)
# ================================================

# --- Rutas base (ajústalas si hiciera falta) ---
ruta_base_3   = "/content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3"
CSV_SEP       = ";"
RUTA_RESULTADOS = os.path.join(ruta_base_3, "RESULTADOS")
RUTA_METRICAS   = os.path.join(ruta_base_3, "METRICAS")
RUTA_REPORTING  = os.path.join(ruta_base_3, "REPORTING")
DATECOL = "FECHA"
PAIR_COLS = ["ID_BUILDING","FM_COST_TYPE"]

# --- Cargar datos (Opción A: directamente del step12) ---
df12 = pd.read_csv(os.path.join(RUTA_RESULTADOS, "preds_reconciliadas_step12.csv"),
                   sep=CSV_SEP, parse_dates=[DATECOL])
# Nos aseguramos de que existan estas columnas
assert {"yhat_base","yhat_reconc"}.issubset(df12.columns), "Faltan columnas en preds_reconciliadas_step12.csv"

# Añadimos dimensiones para segmentar/gráficos
dim = pd.read_csv(os.path.join(RUTA_METRICAS, "dim_buildings.csv"), sep=CSV_SEP)
dim = dim[[c for c in ["ID_BUILDING","PAIS","REGION"] if c in dim.columns]].drop_duplicates()
df12 = df12.merge(dim, on="ID_BUILDING", how="left")

# --- (Opción B: usar el comparativo generado previamente) ---
# cmp_path = os.path.join(RUTA_REPORTING, "compare_recon_steps_bottom.csv")
# if os.path.exists(cmp_path):
#     df_cmp = pd.read_csv(cmp_path, sep=CSV_SEP, parse_dates=[DATECOL])
#     # Renombramos para tener las mismas columnas de interés
#     df12 = (df_cmp.rename(columns={"yhat_base_ref":"yhat_base",
#                                    "yhat_reconc_p12":"yhat_reconc"}))
#     # Unimos PAIS/REGION si hace falta
#     if "PAIS" not in df12.columns and "REGION" not in df12.columns:
#         df12 = df12.merge(dim, on="ID_BUILDING", how="left")

# --- Filtro: base ~0 y reconc > 0 ---
TOL = 1e-9
mask = (df12["yhat_base"].abs() <= TOL) & (df12["yhat_reconc"] > TOL)
df_z2p = df12.loc[mask].copy()   # zero-to-positive

print(f"[Info] Filas base≈0 & reconc>0: {len(df_z2p)} sobre {len(df12)} ({100*len(df_z2p)/max(1,len(df12)):.2f}%)")

# =======================
# 1) Resúmenes tabulares
# =======================
# Top por magnitud del cambio
df_z2p["delta"] = df_z2p["yhat_reconc"] - df_z2p["yhat_base"]
top_abs = df_z2p.assign(abs_delta=df_z2p["delta"].abs()) \
                .sort_values("abs_delta", ascending=False) \
                .head(20) \
                [[*PAIR_COLS, DATECOL, "PAIS","REGION", "yhat_base","yhat_reconc","delta"]]
print("\n[Top 20 cambios absolutos en filas base≈0 → reconc>0]")
display(top_abs)

# Conteo por FM_COST_TYPE y PAIS
by_fm = df_z2p.groupby("FM_COST_TYPE").size().rename("count").reset_index().sort_values("count", ascending=False)
by_pais = (df_z2p.dropna(subset=["PAIS"])
           .groupby("PAIS").size().rename("count").reset_index().sort_values("count", ascending=False))

print("\n[Conteo por FM_COST_TYPE]")
display(by_fm.head(20))

print("\n[Conteo por PAIS]")
display(by_pais.head(20))

# =======================
# 2) Gráficos de resumen
# =======================
FIG_DIR = os.path.join(RUTA_REPORTING, "figs_zero_to_pos_p12")
os.makedirs(FIG_DIR, exist_ok=True)

# 2.1 Histograma de yhat_reconc (solo casos base≈0)
plt.figure(figsize=(10,5))
plt.hist(df_z2p["yhat_reconc"].values, bins=40)
plt.title("Distribución de yhat_reconc para filas con base≈0 (Paso 12)")
plt.xlabel("yhat_reconc")
plt.ylabel("Frecuencia")
plt.tight_layout()
plt.savefig(os.path.join(FIG_DIR, "hist_yhat_reconc_zero2pos.png"))
plt.show()

# 2.2 Boxplot de deltas por FM_COST_TYPE (solo con base≈0)
if not df_z2p.empty:
    order_fm = (df_z2p.groupby("FM_COST_TYPE")["delta"]
                .median().sort_values(ascending=False).index.tolist())
    plt.figure(figsize=(12,6))
    data = [df_z2p.loc[df_z2p["FM_COST_TYPE"]==fm, "delta"].values for fm in order_fm]
    plt.boxplot(data, labels=order_fm, showfliers=False)
    plt.axhline(0, linestyle=":", color="grey")
    plt.title("Delta (reconc-base) por FM_COST_TYPE — casos base≈0")
    plt.ylabel("Delta")
    plt.xticks(rotation=45, ha="right")
    plt.tight_layout()
    plt.savefig(os.path.join(FIG_DIR, "box_delta_por_fm_zero2pos.png"))
    plt.show()

# 2.3 Barras horizontales: Top 20 deltas absolutos
if not df_z2p.empty:
    tops = df_z2p.assign(abs_delta=df_z2p["delta"].abs()).sort_values("abs_delta", ascending=False).head(20)
    labels = tops.apply(lambda r: f"{int(r['ID_BUILDING'])}|{r['FM_COST_TYPE']}|{str(r.get('PAIS',''))}", axis=1)
    plt.figure(figsize=(10,8))
    plt.barh(range(len(tops)), tops["delta"])
    plt.yticks(range(len(tops)), labels)
    plt.gca().invert_yaxis()
    plt.title("Top 20 ajustes (reconc - base) — base≈0")
    plt.xlabel("Delta")
    plt.tight_layout()
    plt.savefig(os.path.join(FIG_DIR, "barh_top20_zero2pos.png"))
    plt.show()

# =======================
# 3) Series temporales por ejemplo
# =======================
# Elegimos N pares con mayor delta total anual (suma de deltas en 2024)
if not df_z2p.empty:
    # Sumamos por par en el año
    annual_rank = (df_z2p.groupby(PAIR_COLS, as_index=False)["delta"]
                          .sum()
                          .sort_values("delta", ascending=False))
    N = 6  # cuántos pares mostrar
    sel_pairs = annual_rank.head(N)[PAIR_COLS].values.tolist()

    for bid, fmc in sel_pairs:
        sub = df12[(df12["ID_BUILDING"]==bid) & (df12["FM_COST_TYPE"]==fmc)].sort_values(DATECOL)
        if sub.empty:
            continue
        plt.figure(figsize=(10,4))
        plt.plot(sub[DATECOL], sub["yhat_base"], label="Base", linewidth=2, color="black")
        plt.plot(sub[DATECOL], sub["yhat_reconc"], label="Reconc P12", linestyle="--")
        pais = str(sub["PAIS"].dropna().iloc[0]) if sub["PAIS"].notna().any() else ""
        plt.title(f"Serie par {bid} | {fmc} | {pais} — base≈0→reconc>0")
        plt.xlabel("Mes"); plt.ylabel("Valor")
        plt.legend()
        plt.tight_layout()
        fname = f"ts_pair_{bid}_{fmc.replace(' ','_')}.png"
        plt.savefig(os.path.join(FIG_DIR, fname))
        plt.show()

print(f"\n[OK] Figuras guardadas en: {FIG_DIR}")

[Info] Filas base≈0 & reconc>0: 6076 sobre 29148 (20.85%)

[Top 20 cambios absolutos en filas base≈0 → reconc>0]

[Conteo por FM_COST_TYPE]

[Conteo por PAIS]

[OK] Figuras guardadas en: /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3/REPORTING/figs_zero_to_pos_p12

# ============================================================
# Paso 12 — Diagnóstico de negativos antes y después del clip
# Variantes: STD (todas las series) y ACTIVE (solo activos)
# ============================================================


ruta_base_3     = "/content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3"
CSV_SEP         = ";"
RUTA_RESULTADOS = os.path.join(ruta_base_3, "RESULTADOS")

DATECOL   = "FECHA"
PAIR_COLS = ["ID_BUILDING","FM_COST_TYPE"]

# --- Rutas de ambas variantes ---
PATHS = {
    "STD": {
        "after":  os.path.join(RUTA_RESULTADOS, "preds_reconciliadas_step12.csv"),
        "before": os.path.join(RUTA_RESULTADOS, "preds_reconciliadas_step12_beforeclip.csv"),
    },
    "ACTIVE": {
        "after":  os.path.join(RUTA_RESULTADOS, "preds_reconciliadas_step12_activeonly.csv"),
        "before": os.path.join(RUTA_RESULTADOS, "preds_reconciliadas_step12_activeonly_beforeclip.csv"),
    }
}

def _load_pair(paths_variant, variant_name):
    """Carga BEFORE/AFTER para una variante, hace merge y devuelve df listo."""
    p_after  = paths_variant["after"]
    p_before = paths_variant["before"]

    if not os.path.exists(p_after):
        print(f"[{variant_name}] AVISO: no existe AFTER: {p_after}. Se omite esta variante.")
        return None

    after = pd.read_csv(p_after, sep=CSV_SEP, parse_dates=[DATECOL]) \
             .rename(columns={"yhat_reconc":"yhat_after"})
    need_common = [*PAIR_COLS, DATECOL, "yhat_base", "yhat_after"]
    missing = [c for c in need_common if c not in after.columns]
    if missing:
        raise KeyError(f"[{variant_name}] Faltan columnas en AFTER: {missing}")

    if not os.path.exists(p_before):
        print(f"[{variant_name}] AVISO: no existe BEFORE: {p_before}. "
              f"Podrás ver negativos post-clip, pero no los truncados.")
        # Creamos un df mínimo sin BEFORE para poder contar negativos post-clip
        df_min = after.copy()
        df_min["yhat_before"]   = np.nan
        df_min["was_negative"]  = False
        df_min["truncated_to0"] = False
        return df_min

    before = pd.read_csv(p_before, sep=CSV_SEP, parse_dates=[DATECOL]) \
              .rename(columns={"yhat_reconc":"yhat_before"})
    need_b = [*PAIR_COLS, DATECOL, "yhat_base", "yhat_before"]
    miss_b = [c for c in need_b if c not in before.columns]
    if miss_b:
        raise KeyError(f"[{variant_name}] Faltan columnas en BEFORE: {miss_b}")

    df = before.merge(after, on=[*PAIR_COLS, DATECOL, "yhat_base"], how="inner")
    df["was_negative"]  = df["yhat_before"] < 0
    df["truncated_to0"] = df["was_negative"] & (df["yhat_after"] == 0)
    return df

def _report_variant(df, variant_name):
    """Imprime resumen, dibuja histogramas y muestra una muestra de truncados."""
    if df is None:
        return

    total = len(df)
    neg_after = int((df["yhat_after"] < 0).sum()) if "yhat_after" in df else 0
    has_before = df["yhat_before"].notna().any()

    print(f"\n===== {variant_name} =====")
    print(f"[Info] Total filas: {total}")
    if has_before:
        print(f"[Info] Negativos ANTES del clip: {int(df['was_negative'].sum())}")
        print(f"[Info] Truncados a 0: {int(df['truncated_to0'].sum())}")
    else:
        print("[Info] BEFORE no disponible: no se puede contar truncados.")

    print(f"[Info] Negativos DESPUÉS del clip: {neg_after} (esperado 0)")

    # Histogramas
    if has_before:
        plt.figure(figsize=(10,5))
        plt.hist(df["yhat_before"], bins=50, alpha=0.5, label="Antes del clip")
        plt.hist(df["yhat_after"],  bins=50, alpha=0.5, label="Después del clip")
        plt.axvline(0, color="black", linestyle="--")
        plt.title(f"Paso 12 — {variant_name}: distribución antes vs. después del clip")
        plt.xlabel("Valor predicho"); plt.ylabel("Frecuencia")
        plt.legend(); plt.tight_layout(); plt.show()
    else:
        plt.figure(figsize=(10,5))
        plt.hist(df["yhat_after"],  bins=50, alpha=0.8, label="Después del clip")
        plt.axvline(0, color="black", linestyle="--")
        plt.title(f"Paso 12 — {variant_name}: distribución (solo post-clip)")
        plt.xlabel("Valor predicho"); plt.ylabel("Frecuencia")
        plt.legend(); plt.tight_layout(); plt.show()

    # Muestra de truncados
    if has_before and df["truncated_to0"].any():
        sample = df[df["truncated_to0"]].sample(min(10, df["truncated_to0"].sum()), random_state=1)
        print("\n[Muestra de pares truncados a 0]")
        display(sample[[*PAIR_COLS, DATECOL, "yhat_base", "yhat_before", "yhat_after"]])

# --- Ejecutamos para ambas variantes ---
df_std    = _load_pair(PATHS["STD"],    "STD")
df_active = _load_pair(PATHS["ACTIVE"], "ACTIVE")

_report_variant(df_std, "STD")
_report_variant(df_active, "ACTIVE")

===== STD =====
[Info] Total filas: 29148
[Info] Negativos ANTES del clip: 2684
[Info] Truncados a 0: 2684
[Info] Negativos DESPUÉS del clip: 0 (esperado 0)

[Muestra de pares truncados a 0]

===== ACTIVE =====
[Info] Total filas: 29148
[Info] Negativos ANTES del clip: 928
[Info] Truncados a 0: 928
[Info] Negativos DESPUÉS del clip: 0 (esperado 0)

[Muestra de pares truncados a 0]

# ============================================================
# Comparación contra datos reales 2024
# - Requiere:
#   RESULTADOS/test_full_2024.csv (real)
#   REPORTING/compare_recon_steps_bottom.csv (predicciones P10, P11, P12 STD/ACTIVE)
# ============================================================

# --- Rutas
PATH_REAL = os.path.join(RUTA_RESULTADOS, "test_full_2024.csv")   # verdad terreno 2024
PATH_BOTTOM = os.path.join(RUTA_REPORTING, "compare_recon_steps_bottom.csv")
OUT_METRICS = os.path.join(RUTA_REPORTING, "compare_recon_vs_real_metrics.csv")
OUT_BOTTOM  = os.path.join(RUTA_REPORTING, "compare_recon_vs_real_bottom.csv")

PAIR_COLS = ["ID_BUILDING","FM_COST_TYPE"]

# --- Lectura real
real_df = pd.read_csv(PATH_REAL, sep=CSV_SEP, parse_dates=[DATECOL])
if "cost_float_mod" not in real_df.columns:
    raise KeyError("Esperábamos columna 'cost_float_mod' en test_full_2024.csv (valor real).")

# --- Lectura predicciones reconciliadas
pred_df = pd.read_csv(PATH_BOTTOM, sep=CSV_SEP, parse_dates=[DATECOL])

# --- Merge
df = pred_df.merge(
    real_df[PAIR_COLS+[DATECOL,"cost_float_mod"]],
    on=PAIR_COLS+[DATECOL], how="left"
).rename(columns={"cost_float_mod":"y_real"})

# --- Calcular errores por cada paso
def _calc_errors(df, col_pred, col_real="y_real"):
    err  = df[col_pred] - df[col_real]
    abs_ = err.abs()
    return {
        "MAE": abs_.mean(),
        "RMSE": np.sqrt((err**2).mean()),
        "MAPE%": (abs_ / df[col_real].replace(0,np.nan)).mean()*100,
        "Bias": err.mean()
    }

metrics = {}
for col in ["yhat_reconc_p10","yhat_reconc_p11","yhat_reconc_p12std","yhat_reconc_p12act"]:
    if col in df.columns:
        metrics[col] = _calc_errors(df, col)

metrics_df = pd.DataFrame(metrics).T
metrics_df.index.name = "Modelo"
metrics_df.to_csv(OUT_METRICS, sep=CSV_SEP)
print(f"[OK] Guardadas métricas comparativas en {OUT_METRICS}")
print(metrics_df)

# --- Export bottom con errores fila a fila (opcional)
keep_cols = [*PAIR_COLS, DATECOL, "y_real"]
for col in ["yhat_reconc_p10","yhat_reconc_p11","yhat_reconc_p12std","yhat_reconc_p12act"]:
    if col in df.columns:
        keep_cols.append(col)
        df[f"error_abs_{col}"] = (df[col] - df["y_real"]).abs()
        keep_cols.append(f"error_abs_{col}")
df[keep_cols].to_csv(OUT_BOTTOM, sep=CSV_SEP, index=False)
print(f"[OK] Guardado bottom con errores: {OUT_BOTTOM}")

[OK] Guardadas métricas comparativas en /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3/REPORTING/compare_recon_vs_real_metrics.csv
                           MAE         RMSE       MAPE%        Bias
Modelo
yhat_reconc_p10     500.917395  2094.018508   91.919262   -3.830971
yhat_reconc_p11     539.231234  2428.586784   89.970338 -122.757539
yhat_reconc_p12std  814.768026  3830.653335  236.143547  112.612548
yhat_reconc_p12act  738.094872  3298.421851  202.837720   49.644224
[OK] Guardado bottom con errores: /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3/REPORTING/compare_recon_vs_real_bottom.csv

# ============================================================
# Métricas vs Real 2024: MAE/RMSE/Bias/MAPE + sMAPE/WAPE + MASE/WASE
#   - Global, por PAIS, por FM_COST_TYPE, y PAIS×FM_COST_TYPE
# Requiere:
#   RESULTADOS/test_full_2024.csv
#   REPORTING/compare_recon_steps_bottom.csv
#   RESULTADOS/train_full_2021_2023.csv  (para MASE)
#   METRICAS/dim_buildings.csv           (para PAIS)
# ============================================================

# --- Rutas base ---
ruta_base_3   = "/content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3"
CSV_SEP       = ";"
DATECOL       = "FECHA"
PAIR_COLS     = ["ID_BUILDING","FM_COST_TYPE"]

RUTA_RESULTADOS = os.path.join(ruta_base_3, "RESULTADOS")
RUTA_METRICAS   = os.path.join(ruta_base_3, "METRICAS")
RUTA_REPORTING  = os.path.join(ruta_base_3, "REPORTING")
os.makedirs(RUTA_REPORTING, exist_ok=True)

PATH_REAL   = os.path.join(RUTA_RESULTADOS, "test_full_2024.csv")
PATH_BOTTOM = os.path.join(RUTA_REPORTING, "compare_recon_steps_bottom.csv")
PATH_TRAIN  = os.path.join(RUTA_RESULTADOS, "train_full_2021_2023.csv")
PATH_DIM    = os.path.join(RUTA_METRICAS,   "dim_buildings.csv")

OUT_OVERALL = os.path.join(RUTA_REPORTING, "compare_recon_vs_real_metrics_overall.csv")
OUT_PAISES  = os.path.join(RUTA_REPORTING, "compare_recon_vs_real_metrics_by_pais.csv")
OUT_FMCOST  = os.path.join(RUTA_REPORTING, "compare_recon_vs_real_metrics_by_fmcost.csv")
OUT_GRID    = os.path.join(RUTA_REPORTING, "compare_recon_vs_real_metrics_by_pais_fmcost.csv")

# --- Lecturas robustas ---
def _read_csv(path, parse_dates=None, required=True):
    if not os.path.exists(path):
        if required:
            raise FileNotFoundError(f"Falta {path}")
        return pd.DataFrame()
    return pd.read_csv(path, sep=CSV_SEP, parse_dates=parse_dates)

real_df  = _read_csv(PATH_REAL,  parse_dates=[DATECOL], required=True)
pred_df  = _read_csv(PATH_BOTTOM, parse_dates=[DATECOL], required=True)
train_df = _read_csv(PATH_TRAIN, parse_dates=[DATECOL], required=True)
dim_df   = _read_csv(PATH_DIM,   required=False)

# --- Chequeo columnas clave ---
if "cost_float_mod" not in real_df.columns:
    raise KeyError("Esperábamos 'cost_float_mod' en test_full_2024.csv")
need_preds = set(PAIR_COLS+[DATECOL])
if not need_preds.issubset(pred_df.columns):
    raise KeyError(f"compare_recon_steps_bottom.csv debe contener {need_preds}")
pred_cols = [c for c in ["yhat_reconc_p10","yhat_reconc_p11","yhat_reconc_p12std","yhat_reconc_p12act"] if c in pred_df.columns]
if not pred_cols:
    raise KeyError("No se encontraron columnas de predicción (p10/p11/p12std/p12act) en compare_recon_steps_bottom.csv")

# --- Merge predicciones con verdad 2024 ---
df = pred_df.merge(
    real_df[PAIR_COLS+[DATECOL,"cost_float_mod"]],
    on=PAIR_COLS+[DATECOL], how="left"
).rename(columns={"cost_float_mod":"y_real"})

# --- Denominador MASE por pareja (naive estacional 12) usando 2021-2023 ---
train = train_df.copy()
train[DATECOL] = pd.to_datetime(train[DATECOL]).dt.to_period("M").dt.to_timestamp()
if "cost_float_mod" not in train.columns:
    raise KeyError("Esperábamos 'cost_float_mod' en train_full_2021_2023.csv")

train = train.sort_values(PAIR_COLS+[DATECOL])
train["y_lag12"]   = train.groupby(PAIR_COLS)["cost_float_mod"].shift(12)
train["abs_diff12"] = (train["cost_float_mod"] - train["y_lag12"]).abs()
scale = (train.groupby(PAIR_COLS, as_index=False)["abs_diff12"]
              .mean()
              .rename(columns={"abs_diff12":"MASE_scale"}))

# Fallback si el denominador falta o es ~0
scale["MASE_scale"] = pd.to_numeric(scale["MASE_scale"], errors="coerce").fillna(0.0)
min_eps = 1e-8
scale.loc[scale["MASE_scale"] < min_eps, "MASE_scale"] = min_eps

# Unimos el denominador a df 2024
df = df.merge(scale, on=PAIR_COLS, how="left")
df["MASE_scale"] = df["MASE_scale"].fillna(1.0)  # último fallback

# --- Funciones de métricas ---
def _safe_div(a, b):
    return a / np.where(b==0, np.nan, b)

def _metrics_block(sub, y_col_pred, y_col_real="y_real", scale_col="MASE_scale"):
    y_real = sub[y_col_real].astype(float)
    y_pred = sub[y_col_pred].astype(float)
    err    = y_pred - y_real
    abs_e  = err.abs()

    # clásicos
    mse    = (err**2).mean()
    rmse   = np.sqrt(mse) if np.isfinite(mse) else np.nan
    mape   = (100.0 * _safe_div(abs_e, y_real.abs())).mean()
    bias   = err.mean()
    mae    = abs_e.mean()

    # sMAPE y WAPE
    smape = (100.0 * _safe_div(abs_e, (y_real.abs() + y_pred.abs())/2.0)).mean()
    wape  = 100.0 * abs_e.sum() / (y_real.abs().sum() if y_real.abs().sum()!=0 else np.nan)

    # MASE / WASE con escala por pareja
    scale_vals = sub[scale_col].astype(float).replace(0, np.nan)
    mase = (abs_e / scale_vals).mean()
    wase = abs_e.sum() / scale_vals.sum()

    return pd.Series({
        "n": len(sub),
        "MAE": mae,
        "RMSE": rmse,
        "MAPE%": mape,
        "sMAPE%": smape,
        "WAPE%": wape,
        "Bias": bias,
        "MASE": mase,
        "WASE": wase,
    })

def _compute_by(df_in, group_cols):
    """
    Calcula métricas por grupo. Si group_cols es vacío/None, calcula global sin groupby.
    """
    rows = []

    def _one_group(g, key_vals):
        for col in pred_cols:
            met = _metrics_block(g, col)
            # construimos fila
            if group_cols:
                base = {k: v for k, v in zip(group_cols, key_vals)}
            else:
                base = {}
            row = {**base, "Modelo": col, **met.to_dict()}
            rows.append(row)

    if not group_cols:
        _one_group(df_in, ())
    else:
        for key_vals, g in df_in.groupby(group_cols, dropna=False):
            if not isinstance(key_vals, tuple):
                key_vals = (key_vals,)
            _one_group(g, key_vals)

    return pd.DataFrame(rows)


# --- Añadimos PAIS desde dim_buildings (si existe) ---
if not dim_df.empty and "PAIS" in dim_df.columns:
    dim_keep = dim_df[["ID_BUILDING","PAIS"]].drop_duplicates()
    df = df.merge(dim_keep, on="ID_BUILDING", how="left")
else:
    df["PAIS"] = np.nan

# --- Cálculos de métricas ---
overall = _compute_by(df, [])  # GLOBAL sin groupby
by_pais = _compute_by(df, ["PAIS"])
by_fmc  = _compute_by(df, ["FM_COST_TYPE"])
by_grid = _compute_by(df, ["PAIS","FM_COST_TYPE"])

# --- Orden y export ---
def _nice_order(d):
    metric_cols = ["n","MAE","RMSE","MAPE%","sMAPE%","WAPE%","Bias","MASE","WASE"]
    lead_cols = [c for c in ["PAIS","FM_COST_TYPE"] if c in d.columns]
    return d[lead_cols + ["Modelo"] + metric_cols]

overall = _nice_order(overall.sort_values(["Modelo"]))
by_pais = _nice_order(by_pais.sort_values(["PAIS","Modelo"]))
by_fmc  = _nice_order(by_fmc.sort_values(["FM_COST_TYPE","Modelo"]))
by_grid = _nice_order(by_grid.sort_values(["PAIS","FM_COST_TYPE","Modelo"]))

overall.to_csv(OUT_OVERALL, sep=CSV_SEP, index=False)
by_pais.to_csv(OUT_PAISES, sep=CSV_SEP, index=False)
by_fmc.to_csv(OUT_FMCOST, sep=CSV_SEP, index=False)
by_grid.to_csv(OUT_GRID, sep=CSV_SEP, index=False)

print("[OK] Métricas globales ->", OUT_OVERALL)
print("[OK] Métricas por país ->", OUT_PAISES)
print("[OK] Métricas por FM_COST_TYPE ->", OUT_FMCOST)
print("[OK] Métricas por país×FM_COST_TYPE ->", OUT_GRID)

print("\n[Resumen Global]\n", overall)

[OK] Métricas globales -> /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3/REPORTING/compare_recon_vs_real_metrics_overall.csv
[OK] Métricas por país -> /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3/REPORTING/compare_recon_vs_real_metrics_by_pais.csv
[OK] Métricas por FM_COST_TYPE -> /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3/REPORTING/compare_recon_vs_real_metrics_by_fmcost.csv
[OK] Métricas por país×FM_COST_TYPE -> /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3/REPORTING/compare_recon_vs_real_metrics_by_pais_fmcost.csv

[Resumen Global]
                Modelo        n         MAE         RMSE       MAPE%  \
0     yhat_reconc_p10  29148.0  500.917395  2094.018508   98.351446
1     yhat_reconc_p11  29148.0  539.231234  2428.586784   96.567198
3  yhat_reconc_p12act  29148.0  738.094872  3298.421851  209.405188
2  yhat_reconc_p12std  29148.0  814.768026  3830.653335  242.672793

       sMAPE%      WAPE%        Bias          MASE      WASE
0  105.824619  33.330729   -3.830971  7.874199e+07  0.810279
1   95.392406  35.880108 -122.757539  1.405050e+08  0.872255
3   97.440512  49.112370   49.644224  1.419558e+08  1.193934
2  110.956662  54.214153  112.612548  3.082074e+08  1.317960

# --- Lecturas rápidas de los CSV de salida para verlos en consola ---
over_v  = pd.read_csv(OUT_OVERALL, sep=CSV_SEP)
pais_v  = pd.read_csv(OUT_PAISES,  sep=CSV_SEP)
fmc_v   = pd.read_csv(OUT_FMCOST,  sep=CSV_SEP)
grid_v  = pd.read_csv(OUT_GRID,    sep=CSV_SEP)

print("\n[Resumen Global]")
print(over_v.sort_values(["Modelo"]).to_string(index=False, max_cols=None))

print("\n[Top 10 por país (ordenado por MAE, asc)]")
print(pais_v.sort_values(["PAIS","MAE"], ascending=[True, True]).groupby("PAIS").head(10).to_string(index=False))

print("\n[Top 10 por FM_COST_TYPE (ordenado por MAE, asc)]")
print(fmc_v.sort_values(["FM_COST_TYPE","MAE"], ascending=[True, True]).groupby("FM_COST_TYPE").head(10).to_string(index=False))

print("\n[Cuadrícula PAIS × FM_COST_TYPE (Top 5 por grupo por MAE, asc)]")
tmp = grid_v.sort_values(["PAIS","FM_COST_TYPE","MAE"], ascending=[True, True, True]) \
            .groupby(["PAIS","FM_COST_TYPE"]).head(5)
print(tmp.to_string(index=False))

[Resumen Global]
            Modelo       n        MAE        RMSE      MAPE%     sMAPE%     WAPE%        Bias         MASE     WASE
   yhat_reconc_p10 29148.0 500.917395 2094.018508  98.351446 105.824619 33.330729   -3.830971 7.874199e+07 0.810279
   yhat_reconc_p11 29148.0 539.231234 2428.586784  96.567198  95.392406 35.880108 -122.757539 1.405050e+08 0.872255
yhat_reconc_p12act 29148.0 738.094872 3298.421851 209.405188  97.440512 49.112370   49.644224 1.419558e+08 1.193934
yhat_reconc_p12std 29148.0 814.768026 3830.653335 242.672793 110.956662 54.214153  112.612548 3.082074e+08 1.317960

[Top 10 por país (ordenado por MAE, asc)]
                PAIS             Modelo       n         MAE         RMSE      MAPE%     sMAPE%      WAPE%         Bias         MASE     WASE
            COLOMBIA    yhat_reconc_p10  2904.0 1110.301675  3327.725852 135.601404  80.382058  36.876977  -582.826995 1.728318e+00 1.865126
            COLOMBIA    yhat_reconc_p11  2904.0 1564.163061  5906.984486 134.915358  80.641754  51.951291 -1133.107565 2.068377e+00 2.627540
            COLOMBIA yhat_reconc_p12act  2904.0 1571.493727  5916.075444 136.047401  81.452745  52.194768 -1142.838336 2.149696e+00 2.639854
            COLOMBIA yhat_reconc_p12std  2904.0 1657.316940  6015.896934 141.093122  86.428769  55.045255 -1271.484523 2.428517e+00 2.784023
          COSTA RICA    yhat_reconc_p11   588.0  330.501355  1244.124553  66.729904  62.366822  75.116375   160.340370 7.008589e-01 0.291624
          COSTA RICA yhat_reconc_p12std   588.0  333.771506  1243.616478  69.850223  62.534155  75.859615   168.437277 7.175600e-01 0.294509
          COSTA RICA yhat_reconc_p12act   588.0  334.109946  1244.943597  69.686357  62.691095  75.936536   167.513568 7.197277e-01 0.294808
          COSTA RICA    yhat_reconc_p10   588.0  349.531707  1399.918073  67.184554  61.385723  79.441595   195.201343 7.050406e-01 0.308415
              ESPAÑA    yhat_reconc_p10 18924.0  258.619590   992.707699  99.764514 126.213381  46.307968    54.279866 9.409426e+07 0.546679
              ESPAÑA    yhat_reconc_p11 18924.0  286.861525  1324.770131  99.709430 115.210209  51.364919    16.829146 1.919598e+08 0.606378
              ESPAÑA yhat_reconc_p12act 18924.0  562.318281  3024.229004 235.705090 118.494114 100.687721   250.584846 1.919638e+08 1.188648
              ESPAÑA yhat_reconc_p12std 18924.0  623.166062  3421.857688 280.517500 131.270424 111.583017   324.621084 3.523202e+08 1.317270
              MÉXICO    yhat_reconc_p11  2304.0 1047.056997  2385.452758  67.461204  57.194380  22.821076  -205.825754 6.106580e+07 0.873972
              MÉXICO yhat_reconc_p12act  2304.0 1077.686595  2410.677660  91.398867  57.645617  23.488662  -167.814686 6.106581e+07 0.899539
              MÉXICO    yhat_reconc_p10  2304.0 1103.906774  2540.256547  73.423184  65.386543  24.060142   -45.551501 6.242629e+07 0.921424
              MÉXICO yhat_reconc_p12std  2304.0 1435.936127  5592.044460 220.090347  74.211843  31.296871   187.202941 1.134584e+08 1.198567
              PANAMÁ    yhat_reconc_p11  2736.0  671.403211  1652.268449  80.616057  50.322384  19.069255  -145.348814 1.341920e+00 0.717190
              PANAMÁ    yhat_reconc_p10  2736.0  771.228764  2126.552452  86.947612  54.026495  21.904509    17.229607 1.446692e+00 0.823824
              PANAMÁ yhat_reconc_p12std  2736.0  853.673028  2111.931703 314.025898  56.633059  24.246099    44.919143 4.703995e+01 0.911890
              PANAMÁ yhat_reconc_p12act  2736.0  863.051542  2240.553438 346.300820  52.017629  24.512468    57.354144 6.109035e+01 0.921909
                PERÚ    yhat_reconc_p11  1392.0  286.970929   688.551255  80.158217  84.072905  23.830577   -38.976733 2.178934e+08 0.609317
                PERÚ yhat_reconc_p12act  1392.0  288.011392   686.696352  81.156963  83.640964  23.916979   -33.543776 2.178934e+08 0.611526
                PERÚ    yhat_reconc_p10  1392.0  290.780812   712.416878  82.601795 102.532918  24.146956   -20.062427 2.343858e+08 0.617406
                PERÚ yhat_reconc_p12std  1392.0  300.605929   683.122214  82.992855 103.784851  24.962851   -16.052080 1.438022e+09 0.638268
REPÚBLICA DOMINICANA yhat_reconc_p12std   300.0 2339.610106  6495.210832  83.084722  86.723633  51.139081  1172.930698 2.886056e+08 1.159960
REPÚBLICA DOMINICANA    yhat_reconc_p11   300.0 2342.340228  6505.162690  76.324739  76.989970  51.198755  1177.299541 1.839465e+08 1.161313
REPÚBLICA DOMINICANA yhat_reconc_p12act   300.0 2349.269232  6504.275674  85.284051  81.769871  51.350209  1184.908055 3.244241e+08 1.164749
REPÚBLICA DOMINICANA    yhat_reconc_p10   300.0 3398.151843 12421.437106  80.110567  83.144414  74.276633  1969.175522 1.843750e+08 1.684776

[Top 10 por FM_COST_TYPE (ordenado por MAE, asc)]
         FM_COST_TYPE             Modelo      n         MAE        RMSE       MAPE%     sMAPE%       WAPE%        Bias         MASE     WASE
Eficiencia Energética    yhat_reconc_p10  840.0  138.898582  593.175398   32.095701  66.453216   27.413572   14.517002 1.146632e+09 0.416257
Eficiencia Energética    yhat_reconc_p11  840.0  207.510397  784.715757   34.039350  41.379164   40.955070  -38.297901 3.781648e+09 0.621876
Eficiencia Energética yhat_reconc_p12act  840.0  229.986726  782.057432   56.347817  48.233333   45.391087    1.691593 3.781648e+09 0.689234
Eficiencia Energética yhat_reconc_p12std  840.0  330.154482  813.217197   66.958533  86.580553   65.160591  118.079773 1.029452e+10 0.989421
            Licencias    yhat_reconc_p11 3096.0   56.584079  239.244704  100.388030 193.992326  175.573061   -2.311593 3.429715e+08 0.573533
            Licencias    yhat_reconc_p10 3096.0   57.390285  215.891907   93.480921 195.463468  178.074613    3.388258 3.527779e+08 0.581705
            Licencias yhat_reconc_p12act 3096.0   78.238391  371.983199  120.561188 193.341006  242.763582   20.965428 3.429715e+08 0.793020
            Licencias yhat_reconc_p12std 3096.0  585.136595 5343.602401 1537.138234 196.424203 1815.602991  532.092650 4.577426e+08 5.930913
      Mtto. Contratos    yhat_reconc_p10 4380.0  122.031691  300.622975   56.254698 125.071811   31.493601   -1.944875 1.177952e+07 1.188278
      Mtto. Contratos    yhat_reconc_p11 4380.0  127.218634  312.705883   58.458333 114.418880   32.832232  -36.486314 1.175214e+07 1.238786
      Mtto. Contratos yhat_reconc_p12std 4380.0  584.979000 2444.441331  418.849803 128.626188  150.969759  420.762737 1.843884e+07 5.696206
      Mtto. Contratos yhat_reconc_p12act 4380.0  671.215517 3246.446273  555.666321 119.947555  173.225441  501.265303 2.072728e+07 6.535931
     Mtto. Correctivo    yhat_reconc_p10 5364.0  544.749446 1096.758884  176.876199 110.881061   71.932376   12.158834 3.380292e+07 0.952662
     Mtto. Correctivo    yhat_reconc_p11 5364.0  550.574494 1105.905509  176.724690 107.939244   72.701555  -13.891866 3.391718e+07 0.962848
     Mtto. Correctivo yhat_reconc_p12act 5364.0  621.713616 1522.150384  189.925199 109.885251   82.095243   36.046911 3.391718e+07 1.087257
     Mtto. Correctivo yhat_reconc_p12std 5364.0  623.831565 1338.521291  214.187997 110.271154   82.374911   76.161893 3.928932e+07 1.090961
                Obras    yhat_reconc_p10 1080.0  282.138998  738.096242   76.988315 145.838566   32.942737  -43.070322 9.228150e-01 0.386801
                Obras    yhat_reconc_p11 1080.0  709.693403 3124.609779  226.693472 139.317273   82.864275 -249.691655 1.206986e+00 0.972960
                Obras yhat_reconc_p12act 1080.0  973.446855 3300.389459  391.347413 140.066773  113.660304   33.929918 7.940854e+00 1.334555
                Obras yhat_reconc_p12std 1080.0 2212.621623 8886.225726 1163.678566 142.280306  258.347175 1219.591144 8.035125e+07 3.033412
      Servicios Ctto.    yhat_reconc_p11 4056.0  332.245934  907.199089   56.084223  88.347319   15.961314 -138.322589 2.592835e+06 0.726376
      Servicios Ctto.    yhat_reconc_p10 4056.0  356.153235  892.424560   63.957686  91.877839   17.109836   -4.491260 2.648529e+06 0.778644
      Servicios Ctto. yhat_reconc_p12act 4056.0  659.857585 2862.296816  253.098549  92.470325   31.699994   73.043492 2.609691e+06 1.442620
      Servicios Ctto. yhat_reconc_p12std 4056.0  894.962060 4790.594207  425.310706  94.887415   42.994568  320.505763 2.623268e+06 1.956620
      Servicios Extra    yhat_reconc_p11 4512.0  265.447316 2089.690435  142.358710 140.988843   94.663755  -91.962747 8.340545e+07 1.429719
      Servicios Extra    yhat_reconc_p10 4512.0  286.939249 1611.465887  166.897301 153.699864  102.328203    5.119555 8.365123e+07 1.545476
      Servicios Extra yhat_reconc_p12std 4512.0  407.687960 3138.208678  215.886134 156.548326  145.389578   62.198957 8.489535e+07 2.195838
      Servicios Extra yhat_reconc_p12act 4512.0  427.433973 3692.681173  224.447248 141.016068  152.431396   76.870081 8.340557e+07 2.302191
          Suministros    yhat_reconc_p10 5820.0 1250.727017 4126.457572   58.803759  43.668852   28.888309  -28.838059 1.252266e+00 0.720285
          Suministros    yhat_reconc_p11 5820.0 1420.067483 4704.964679   62.097654  44.730558   32.799601 -351.362720 1.413498e+00 0.817808
          Suministros yhat_reconc_p12std 5820.0 1475.481092 4783.799018   61.784262  48.248719   34.079501 -441.806131 1.501874e+00 0.849720
          Suministros yhat_reconc_p12act 5820.0 1494.474387 4926.126118   66.710955  45.674310   34.518194 -305.132378 1.791503e+00 0.860658

[Cuadrícula PAIS × FM_COST_TYPE (Top 5 por grupo por MAE, asc)]
                PAIS          FM_COST_TYPE             Modelo      n          MAE         RMSE       MAPE%     sMAPE%       WAPE%          Bias         MASE         WASE
            COLOMBIA             Licencias    yhat_reconc_p10   12.0          NaN          NaN         NaN        NaN         NaN           NaN          NaN 0.000000e+00
            COLOMBIA             Licencias    yhat_reconc_p11   12.0          NaN          NaN         NaN        NaN         NaN           NaN          NaN 0.000000e+00
            COLOMBIA             Licencias yhat_reconc_p12act   12.0          NaN          NaN         NaN        NaN         NaN           NaN          NaN 0.000000e+00
            COLOMBIA             Licencias yhat_reconc_p12std   12.0          NaN          NaN         NaN        NaN         NaN           NaN          NaN 0.000000e+00
            COLOMBIA       Mtto. Contratos    yhat_reconc_p11   48.0 8.602605e+01 1.564903e+02   34.382645  93.474635   77.188585 -3.227475e+01 2.467171e+00 2.145381e+00
            COLOMBIA       Mtto. Contratos    yhat_reconc_p10   48.0 8.950868e+01 1.501993e+02   40.387405 105.428891   80.313448 -1.893927e+01 2.567026e+00 2.232234e+00
            COLOMBIA       Mtto. Contratos yhat_reconc_p12act   48.0 1.524254e+02 2.348966e+02   93.606815  93.873738  136.766749  9.526720e+01 3.809845e+00 3.801298e+00
            COLOMBIA       Mtto. Contratos yhat_reconc_p12std   48.0 1.885530e+02 2.489393e+02   70.678136 111.779068  169.182918  1.317829e+02 6.911367e+00 4.702274e+00
            COLOMBIA      Mtto. Correctivo yhat_reconc_p12act  888.0 4.418766e+02 6.635099e+02  235.640478  81.244682   68.352682  1.139509e+02 8.392020e-01 7.351206e-01
            COLOMBIA      Mtto. Correctivo    yhat_reconc_p10  888.0 4.425735e+02 6.603662e+02  240.035774  82.078806   68.460488  1.256100e+02 8.416215e-01 7.362800e-01
            COLOMBIA      Mtto. Correctivo    yhat_reconc_p11  888.0 4.426605e+02 6.636764e+02  235.460520  81.253604   68.473953  1.124105e+02 8.423218e-01 7.364248e-01
            COLOMBIA      Mtto. Correctivo yhat_reconc_p12std  888.0 4.486512e+02 6.672469e+02  244.324060  82.815262   69.400635  1.293727e+02 9.489757e-01 7.463911e-01
            COLOMBIA                 Obras    yhat_reconc_p10  168.0          NaN          NaN         NaN        NaN         NaN           NaN          NaN 0.000000e+00
            COLOMBIA                 Obras    yhat_reconc_p11  168.0          NaN          NaN         NaN        NaN         NaN           NaN          NaN 0.000000e+00
            COLOMBIA                 Obras yhat_reconc_p12act  168.0          NaN          NaN         NaN        NaN         NaN           NaN          NaN 0.000000e+00
            COLOMBIA                 Obras yhat_reconc_p12std  168.0          NaN          NaN         NaN        NaN         NaN           NaN          NaN 0.000000e+00
            COLOMBIA       Servicios Ctto.    yhat_reconc_p11   24.0 9.661344e+02 1.232601e+03   36.497305  36.647099   36.101647 -7.581767e+02 4.191610e+00 1.324417e+00
            COLOMBIA       Servicios Ctto.    yhat_reconc_p10   24.0 9.665327e+02 1.230126e+03   36.848030  36.886379   36.116531 -7.533263e+02 4.273413e+00 1.324963e+00
            COLOMBIA       Servicios Ctto. yhat_reconc_p12act   24.0 1.160137e+03 1.291332e+03   68.103600 122.117872   43.350966 -1.160137e+03 1.081178e+01 1.590364e+00
            COLOMBIA       Servicios Ctto. yhat_reconc_p12std   24.0 1.161847e+03 1.335397e+03   68.096060  73.199806   43.414867 -8.031317e+02 1.087115e+01 1.592708e+00
            COLOMBIA       Servicios Extra yhat_reconc_p12act  876.0 9.002190e+01 1.610567e+02  147.972778 117.309055   96.121866  2.956006e+00 8.691726e-01 8.489199e-01
            COLOMBIA       Servicios Extra    yhat_reconc_p11  876.0 9.033841e+01 1.613576e+02  147.871422 117.464561   96.459824  3.199186e+00 8.876542e-01 8.519047e-01
            COLOMBIA       Servicios Extra    yhat_reconc_p10  876.0 9.045936e+01 1.613609e+02  148.303465 118.538624   96.588976  3.734279e+00 8.901743e-01 8.530453e-01
            COLOMBIA       Servicios Extra yhat_reconc_p12std  876.0 9.250879e+01 1.674328e+02  147.949851 119.118394   98.777277  5.428950e+00 1.008171e+00 8.723717e-01
            COLOMBIA           Suministros    yhat_reconc_p10  888.0 2.843154e+03 5.784948e+03   30.676294  40.731907   33.771866 -1.895771e+03 3.327711e+00 2.613579e+00
            COLOMBIA           Suministros    yhat_reconc_p11  888.0 4.235636e+03 1.032117e+04   33.751415  43.944223   50.312196 -3.569215e+03 4.380259e+00 3.893623e+00
            COLOMBIA           Suministros yhat_reconc_p12act  888.0 4.250387e+03 1.033687e+04   33.849609  44.212382   50.487414 -3.596395e+03 4.399560e+00 3.907183e+00
            COLOMBIA           Suministros yhat_reconc_p12std  888.0 4.502428e+03 1.051161e+04   40.019443  55.786519   53.481241 -4.020510e+03 4.838715e+00 4.138873e+00
          COSTA RICA       Mtto. Contratos    yhat_reconc_p10  108.0 2.090395e+02 3.412823e+02   94.437230  70.127574   64.857570  1.464631e+02 8.916542e-01 9.303899e-01
          COSTA RICA       Mtto. Contratos    yhat_reconc_p11  108.0 2.106987e+02 3.396254e+02   93.331942  75.443614   65.372352  8.655281e+01 8.874159e-01 9.377745e-01
          COSTA RICA       Mtto. Contratos yhat_reconc_p12std  108.0 2.212114e+02 3.489684e+02  101.513187  74.801343   68.634092  1.077226e+02 9.381215e-01 9.845645e-01
          COSTA RICA       Mtto. Contratos yhat_reconc_p12act  108.0 2.260652e+02 3.537979e+02  105.011372  77.110730   70.140049  1.134218e+02 9.653356e-01 1.006168e+00
          COSTA RICA      Mtto. Correctivo    yhat_reconc_p11  120.0 2.650117e+02 3.930480e+02  124.198586  85.617393   59.727139 -8.515727e-01 7.601019e-01 7.393726e-01
          COSTA RICA      Mtto. Correctivo yhat_reconc_p12act  120.0 2.652603e+02 3.919777e+02  124.907970  85.409398   59.783171  2.872679e+00 7.635030e-01 7.400662e-01
          COSTA RICA      Mtto. Correctivo    yhat_reconc_p10  120.0 2.666940e+02 3.914438e+02  124.989253  85.667053   60.106297  4.591948e+00 7.632649e-01 7.440663e-01
          COSTA RICA      Mtto. Correctivo yhat_reconc_p12std  120.0 2.689895e+02 3.927227e+02  129.010945  85.735395   60.623653  1.227309e+01 7.795721e-01 7.504707e-01
          COSTA RICA                 Obras    yhat_reconc_p10   12.0          NaN          NaN         NaN        NaN         NaN           NaN          NaN 0.000000e+00
          COSTA RICA                 Obras    yhat_reconc_p11   12.0          NaN          NaN         NaN        NaN         NaN           NaN          NaN 0.000000e+00
          COSTA RICA                 Obras yhat_reconc_p12act   12.0          NaN          NaN         NaN        NaN         NaN           NaN          NaN 0.000000e+00
          COSTA RICA                 Obras yhat_reconc_p12std   12.0          NaN          NaN         NaN        NaN         NaN           NaN          NaN 0.000000e+00
          COSTA RICA       Servicios Ctto.    yhat_reconc_p11  108.0 5.561821e+01 9.700808e+01   13.670388  25.816044   12.457179 -2.438456e-01 5.796498e-01 5.185397e-01
          COSTA RICA       Servicios Ctto. yhat_reconc_p12act  108.0 5.562710e+01 9.704542e+01   13.669048  25.811259   12.459171 -2.095185e-01 5.797147e-01 5.186226e-01
          COSTA RICA       Servicios Ctto. yhat_reconc_p12std  108.0 5.563211e+01 9.707430e+01   13.667388  25.806742   12.460293 -1.797925e-01 5.797412e-01 5.186693e-01
          COSTA RICA       Servicios Ctto.    yhat_reconc_p10  108.0 5.568180e+01 9.776440e+01   13.622305  25.594305   12.471422  1.522387e+00 5.820670e-01 5.191325e-01
          COSTA RICA       Servicios Extra    yhat_reconc_p11  120.0 1.370177e+02 2.060673e+02   39.416173  48.659177   27.250967  5.633716e+01 5.943168e-01 4.802084e-01
          COSTA RICA       Servicios Extra yhat_reconc_p12std  120.0 1.370249e+02 2.060720e+02   39.409447  49.891631   27.252391  5.638540e+01 5.943273e-01 4.802335e-01
          COSTA RICA       Servicios Extra yhat_reconc_p12act  120.0 1.370270e+02 2.060730e+02   39.408611  48.627366   27.252811  5.639767e+01 5.943199e-01 4.802409e-01
          COSTA RICA       Servicios Extra    yhat_reconc_p10  120.0 1.373846e+02 2.067580e+02   39.458854  49.913652   27.323933  5.699699e+01 5.948116e-01 4.814942e-01
          COSTA RICA           Suministros yhat_reconc_p12std  120.0 6.465353e+03 7.561012e+03  100.000000 200.000000  833.825244  4.914583e+03 5.850810e-01 1.860343e-01
          COSTA RICA           Suministros yhat_reconc_p12act  120.0 6.472184e+03 7.567896e+03  100.000000 200.000000  834.706304  4.921415e+03 5.856992e-01 1.862309e-01
          COSTA RICA           Suministros    yhat_reconc_p11  120.0 6.472407e+03 7.567938e+03  100.000000 200.000000  834.735038  4.921638e+03 5.857194e-01 1.862373e-01
          COSTA RICA           Suministros    yhat_reconc_p10  120.0 7.208459e+03 8.563702e+03   99.522594 199.809945  929.662357  5.665093e+03 6.523283e-01 2.074165e-01
              ESPAÑA Eficiencia Energética    yhat_reconc_p10  276.0 1.996341e+02 6.681947e+02   75.288505 168.007316   36.542168 -1.211499e+02 3.333023e+09 6.415581e-01
              ESPAÑA Eficiencia Energética yhat_reconc_p12act  276.0 5.274124e+02 1.303192e+03  116.144669 171.210057   96.540605 -2.305626e+02 1.276658e+10 1.694930e+00
              ESPAÑA Eficiencia Energética    yhat_reconc_p11  276.0 5.395867e+02 1.322858e+03  116.212032 171.791678   98.769050 -2.278881e+02 1.276658e+10 1.734054e+00
              ESPAÑA Eficiencia Energética yhat_reconc_p12std  276.0 7.606144e+02 1.349497e+03  154.009210 183.948998  139.227229  5.263372e+01 2.712789e+10 2.444364e+00
              ESPAÑA             Licencias    yhat_reconc_p10 2712.0 4.820633e+01 1.508958e+02   94.807736 195.760911  179.215517  3.510925e+00 3.428023e+08 4.643468e-01
              ESPAÑA             Licencias    yhat_reconc_p11 2712.0 5.052590e+01 1.711367e+02  102.712488 194.752832  187.838898  1.268480e+00 3.328306e+08 4.866900e-01
              ESPAÑA             Licencias yhat_reconc_p12act 2712.0 6.354883e+01 2.166021e+02   97.970115 194.127343  236.253932  1.520108e+01 3.328306e+08 6.121332e-01
              ESPAÑA             Licencias yhat_reconc_p12std 2712.0 2.250443e+02 1.845398e+03  257.492276 196.864184  836.641831  1.799589e+02 4.064774e+08 2.167736e+00
              ESPAÑA       Mtto. Contratos    yhat_reconc_p10 3108.0 1.123645e+02 2.668509e+02   77.608144 150.975188   61.405148  1.600283e+01 3.271766e+00 1.760724e+00
              ESPAÑA       Mtto. Contratos    yhat_reconc_p11 3108.0 1.200335e+02 2.993945e+02   82.956805 144.738313   65.596143 -1.496156e+01 3.311149e+00 1.880896e+00
              ESPAÑA       Mtto. Contratos yhat_reconc_p12std 3108.0 6.399644e+02 2.698899e+03  351.693902 154.923618  349.728950  4.955256e+02 1.707786e+02 1.002809e+01
              ESPAÑA       Mtto. Contratos yhat_reconc_p12act 3108.0 7.198411e+02 3.581909e+03  463.038823 152.955502  393.380134  5.662103e+02 2.093438e+02 1.127974e+01
              ESPAÑA      Mtto. Correctivo    yhat_reconc_p10 3168.0 3.910281e+02 8.215922e+02  138.284816 134.550139   85.300348 -8.600146e+01 1.766795e+00 1.284041e+00
              ESPAÑA      Mtto. Correctivo    yhat_reconc_p11 3168.0 4.158931e+02 8.983615e+02  146.470494 131.579565   90.724478 -9.069739e+01 1.770968e+00 1.365692e+00
              ESPAÑA      Mtto. Correctivo yhat_reconc_p12std 3168.0 4.997329e+02 1.161020e+03  164.324931 132.554205  109.013611  5.520675e+00 5.795190e+00 1.641001e+00
              ESPAÑA      Mtto. Correctivo yhat_reconc_p12act 3168.0 5.252093e+02 1.611736e+03  160.555164 135.609618  114.571128 -2.306559e+01 2.064205e+00 1.724660e+00
              ESPAÑA                 Obras    yhat_reconc_p10  816.0 3.003415e+02 7.615337e+02   76.988315 143.376683   32.942737 -4.584905e+01 9.820125e-01 3.992246e-01
              ESPAÑA                 Obras    yhat_reconc_p11  816.0 7.554801e+02 3.223828e+03  226.693472 139.317273   82.864275 -2.658008e+02 1.284856e+00 1.004211e+00
              ESPAÑA                 Obras yhat_reconc_p12act  816.0 1.036250e+03 3.405190e+03  391.347413 140.066773  113.660304  3.611894e+01 8.453167e+00 1.377420e+00
              ESPAÑA                 Obras yhat_reconc_p12std  816.0 2.355086e+03 9.168397e+03 1163.678566 141.649490  258.315902  1.297989e+03 5.702347e+07 3.130464e+00
              ESPAÑA       Servicios Ctto.    yhat_reconc_p10 2796.0 1.517697e+02 5.145702e+02   57.092712 117.107335   14.821149 -1.171388e+00 3.851354e+06 7.371908e-01
              ESPAÑA       Servicios Ctto.    yhat_reconc_p11 2796.0 1.927070e+02 7.187720e+02   51.877846 114.219389   18.818912 -4.715020e+01 3.770367e+06 9.360359e-01
              ESPAÑA       Servicios Ctto. yhat_reconc_p12act 2796.0 6.673740e+02 3.351419e+03  389.190483 119.841246   65.172769  2.636087e+02 3.794877e+06 3.241636e+00
              ESPAÑA       Servicios Ctto. yhat_reconc_p12std 2796.0 1.009206e+03 5.717512e+03  684.446336 121.852081   98.554575  6.202412e+02 3.814621e+06 4.902017e+00
              ESPAÑA       Servicios Extra    yhat_reconc_p11 2352.0 2.353164e+02 2.798129e+03  211.091721 178.106379  121.544532 -9.282737e+01 1.623507e+08 1.823178e+00
              ESPAÑA       Servicios Extra    yhat_reconc_p10 2352.0 2.783898e+02 2.099504e+03  291.026203 184.968571  143.792616  8.444804e+01 1.628291e+08 2.156902e+00
              ESPAÑA       Servicios Extra yhat_reconc_p12std 2352.0 5.114136e+02 4.301113e+03  452.064995 188.750601  264.153004  2.059871e+02 1.652508e+08 3.962318e+00
              ESPAÑA       Servicios Extra yhat_reconc_p12act 2352.0 5.508455e+02 5.086561e+03  480.078512 181.208216  284.520234  2.354741e+02 1.623510e+08 4.267828e+00
              ESPAÑA           Suministros    yhat_reconc_p10 3696.0 4.470256e+02 1.006780e+03   75.562949  50.076791   41.579036  2.755214e+02 8.407124e-01 2.875603e-01
              ESPAÑA           Suministros    yhat_reconc_p11 3696.0 4.841622e+02 1.069207e+03   80.128235  51.055702   45.033209  3.023840e+02 8.502245e-01 3.114494e-01
              ESPAÑA           Suministros yhat_reconc_p12std 3696.0 5.090017e+02 1.177697e+03   78.536631  53.718040   47.343598  2.871335e+02 8.769141e-01 3.274280e-01
              ESPAÑA           Suministros yhat_reconc_p12act 3696.0 5.969939e+02 2.095606e+03   87.241136  52.466698   55.527988  3.820775e+02 1.435527e+00 3.840312e-01
              MÉXICO Eficiencia Energética    yhat_reconc_p11   24.0 3.458000e+02 1.694067e+03    4.760073   3.029561    6.399805  3.458000e+02 8.692598e-02 1.202973e-01
              MÉXICO Eficiencia Energética yhat_reconc_p12act   24.0 3.498822e+02 1.696260e+03    4.816266   3.082074    6.475356  3.498822e+02 8.795215e-02 1.217174e-01
              MÉXICO Eficiencia Energética yhat_reconc_p12std   24.0 3.509407e+02 1.696161e+03    4.830837   3.096801    6.494945  3.509407e+02 8.821823e-02 1.220857e-01
              MÉXICO Eficiencia Energética    yhat_reconc_p10   24.0 1.033264e+03 2.378422e+03   14.223279  10.870121   19.122861  8.930316e+02 2.597381e-01 3.594529e-01
              MÉXICO             Licencias    yhat_reconc_p11  360.0 9.012045e+01 4.603941e+02   90.033625 190.605922  145.986827 -2.212986e+01 3.991086e+08 1.331120e+00
              MÉXICO             Licencias    yhat_reconc_p10  360.0 1.082300e+02 4.225867e+02   87.570567 193.906442  175.322657  2.709212e+00 4.080004e+08 1.598607e+00
              MÉXICO             Licencias yhat_reconc_p12act  360.0 1.595556e+02 8.028950e+02  221.194146 189.928600  258.465407  5.287523e+01 3.991087e+08 2.356710e+00
              MÉXICO             Licencias yhat_reconc_p12std  360.0 2.578505e+03 1.295259e+04 7237.379322 194.384054 4176.939912  2.481404e+03 7.415318e+08 3.808570e+01
              MÉXICO       Mtto. Contratos    yhat_reconc_p11  360.0 2.388590e+02 4.744705e+02   11.120825  18.081346    9.316478 -2.252095e+02 1.118095e+00 5.913739e-01
              MÉXICO       Mtto. Contratos    yhat_reconc_p10  360.0 2.438610e+02 5.864598e+02   10.856864  17.365888    9.511579 -1.581693e+02 1.131542e+00 6.037582e-01
              MÉXICO       Mtto. Contratos yhat_reconc_p12std  360.0 3.315708e+02 6.556694e+02   91.375547  21.213760   12.932618 -1.022876e+02 1.940925e+01 8.209125e-01
              MÉXICO       Mtto. Contratos yhat_reconc_p12act  360.0 3.659308e+02 7.773484e+02  119.598913  21.249635   14.272798 -6.022180e+01 2.600152e+01 9.059819e-01
              MÉXICO      Mtto. Correctivo    yhat_reconc_p11  384.0 1.293459e+03 2.112047e+03  180.772482  70.704551   57.929066  1.146384e+01 9.305130e-01 7.995337e-01
              MÉXICO      Mtto. Correctivo yhat_reconc_p12act  384.0 1.294224e+03 2.111691e+03  181.781608  70.609185   57.963341  1.716024e+01 9.323254e-01 8.000068e-01
              MÉXICO      Mtto. Correctivo yhat_reconc_p12std  384.0 1.297287e+03 2.111066e+03  184.239828  70.480368   58.100519  3.199768e+01 9.369424e-01 8.019001e-01
              MÉXICO      Mtto. Correctivo    yhat_reconc_p10  384.0 1.377875e+03 2.157477e+03  205.430086  72.781081   61.709759  1.844418e+02 9.788445e-01 8.517146e-01
              MÉXICO                 Obras    yhat_reconc_p10   48.0          NaN          NaN         NaN        NaN         NaN           NaN          NaN 0.000000e+00
              MÉXICO                 Obras    yhat_reconc_p11   48.0          NaN          NaN         NaN        NaN         NaN           NaN          NaN 0.000000e+00
              MÉXICO                 Obras yhat_reconc_p12act   48.0          NaN          NaN         NaN        NaN         NaN           NaN          NaN 0.000000e+00
              MÉXICO                 Obras yhat_reconc_p12std   48.0          NaN          NaN         NaN        NaN         NaN           NaN          NaN 0.000000e+00
              MÉXICO       Servicios Ctto. yhat_reconc_p12std  372.0 1.205576e+03 1.583236e+03   17.712200  17.659197   15.606150 -1.153989e+03 8.134833e-01 7.209804e-01
              MÉXICO       Servicios Ctto. yhat_reconc_p12act  372.0 1.205609e+03 1.583265e+03   17.712880  17.660159   15.606577 -1.154024e+03 8.135409e-01 7.210001e-01
              MÉXICO       Servicios Ctto.    yhat_reconc_p11  372.0 1.205646e+03 1.583297e+03   17.713571  17.661225   15.607055 -1.154064e+03 8.136060e-01 7.210222e-01
              MÉXICO       Servicios Ctto.    yhat_reconc_p10  372.0 1.256766e+03 1.659155e+03   19.792533  17.041092   16.268811 -9.196107e+02 8.448010e-01 7.515943e-01
              MÉXICO       Servicios Extra    yhat_reconc_p10  384.0 7.783359e+02 1.449973e+03   85.333999 122.428547   67.510553 -3.127893e+02 1.754318e+00 1.408148e+00
              MÉXICO       Servicios Extra yhat_reconc_p12act  384.0 7.850293e+02 1.467116e+03   84.414310 117.194154   68.091124 -3.446144e+02 1.765017e+00 1.420258e+00
              MÉXICO       Servicios Extra yhat_reconc_p12std  384.0 7.850439e+02 1.467166e+03   84.440099 124.361213   68.092390 -3.447651e+02 1.764662e+00 1.420284e+00
              MÉXICO       Servicios Extra    yhat_reconc_p11  384.0 7.850917e+02 1.467387e+03   84.584961 117.556604   68.096532 -3.454633e+02 1.764436e+00 1.420371e+00
              MÉXICO           Suministros yhat_reconc_p12std  372.0 2.523998e+03 4.826934e+03   43.728503  28.266964   18.908743  3.653785e+02 1.095180e+00 9.681665e-01
              MÉXICO           Suministros yhat_reconc_p12act  372.0 2.565097e+03 4.877628e+03   47.265516  28.407935   19.216637  4.390553e+02 1.117792e+00 9.839313e-01
              MÉXICO           Suministros    yhat_reconc_p11  372.0 2.566828e+03 4.880635e+03   47.527178  28.405516   19.229606  4.437887e+02 1.118909e+00 9.845954e-01
              MÉXICO           Suministros    yhat_reconc_p10  372.0 2.707181e+03 5.260576e+03   48.702784  28.682353   20.281070  8.313456e+02 1.154554e+00 1.038432e+00
              PANAMÁ Eficiencia Energética    yhat_reconc_p11  420.0 6.635516e+01 1.312892e+02   28.421826  24.684063   27.350951  2.456932e+01 3.757856e-01 2.921270e-01
              PANAMÁ Eficiencia Energética    yhat_reconc_p10  420.0 6.852578e+01 1.302864e+02   29.214597  25.154686   28.245660  2.716541e+01 3.831320e-01 3.016831e-01
              PANAMÁ Eficiencia Energética yhat_reconc_p12act  420.0 1.090331e+02 1.846150e+02   54.160348  33.046208   44.942390  9.180045e+01 9.168524e-01 4.800156e-01
              PANAMÁ Eficiencia Energética yhat_reconc_p12std  420.0 1.250056e+02 2.145591e+02   63.139800  35.547533   51.526096  1.092329e+02 1.120453e+00 5.503342e-01
              PANAMÁ       Mtto. Contratos    yhat_reconc_p10  456.0 8.032613e+01 2.076829e+02   22.628851  31.161390   24.894311 -4.356891e+01 2.038193e+00 7.672442e-01
              PANAMÁ       Mtto. Contratos    yhat_reconc_p11  456.0 8.440131e+01 2.715856e+02   20.648333  31.974975   26.157272 -6.956163e+01 2.046855e+00 8.061687e-01
              PANAMÁ       Mtto. Contratos yhat_reconc_p12std  456.0 8.453532e+02 2.705351e+03 1039.718026  34.354192  261.988056  7.101036e+02 2.644371e+02 8.074488e+00
              PANAMÁ       Mtto. Contratos yhat_reconc_p12act  456.0 1.109494e+03 3.635373e+03 1401.081320  34.638977  343.849452  9.767355e+02 3.553386e+02 1.059746e+01
              PANAMÁ      Mtto. Correctivo    yhat_reconc_p10  480.0 8.345387e+02 1.156740e+03  205.066659  74.600907   59.636311  9.717648e+01 1.373729e+00 6.800536e-01
              PANAMÁ      Mtto. Correctivo    yhat_reconc_p11  480.0 8.360584e+02 1.155348e+03  202.009318  75.083126   59.744913  5.415855e+01 1.396133e+00 6.812920e-01
              PANAMÁ      Mtto. Correctivo yhat_reconc_p12act  480.0 9.161377e+02 1.356590e+03  255.329328  75.209723   65.467392  1.557578e+02 4.357639e+00 7.465474e-01
              PANAMÁ      Mtto. Correctivo yhat_reconc_p12std  480.0 1.089764e+03 2.037241e+03  428.940679  75.276363   77.874802  3.588873e+02 1.080204e+01 8.880334e-01
              PANAMÁ                 Obras    yhat_reconc_p10   12.0          NaN          NaN         NaN        NaN         NaN           NaN          NaN 0.000000e+00
              PANAMÁ                 Obras    yhat_reconc_p11   12.0          NaN          NaN         NaN        NaN         NaN           NaN          NaN 0.000000e+00
              PANAMÁ                 Obras yhat_reconc_p12act   12.0          NaN          NaN         NaN        NaN         NaN           NaN          NaN 0.000000e+00
              PANAMÁ                 Obras yhat_reconc_p12std   12.0          NaN          NaN         NaN        NaN         NaN           NaN          NaN 0.000000e+00
              PANAMÁ       Servicios Ctto. yhat_reconc_p12std  444.0 5.999912e+02 1.366406e+03  122.053990  17.330603   11.696189  1.203363e+02 7.214208e-01 4.738209e-01
              PANAMÁ       Servicios Ctto. yhat_reconc_p12act  444.0 6.000564e+02 1.366429e+03  122.052743  17.332683   11.697459  1.202428e+02 7.221703e-01 4.738724e-01
              PANAMÁ       Servicios Ctto.    yhat_reconc_p11  444.0 6.001130e+02 1.366420e+03  122.044181  17.334554   11.698563  1.201381e+02 7.228624e-01 4.739171e-01
              PANAMÁ       Servicios Ctto.    yhat_reconc_p10  444.0 1.016816e+03 1.726027e+03  150.303417  21.607832   19.821752  8.172884e+02 1.336917e+00 8.029932e-01
              PANAMÁ       Servicios Extra    yhat_reconc_p10  480.0 4.602213e+02 1.158791e+03   92.744079 145.318572  109.103449 -1.395756e+02 2.421084e+00 1.690418e+00
              PANAMÁ       Servicios Extra yhat_reconc_p12act  480.0 4.610241e+02 1.188274e+03   93.674997 141.469584  109.293758 -1.665621e+02 2.425195e+00 1.693367e+00
              PANAMÁ       Servicios Extra yhat_reconc_p12std  480.0 4.611320e+02 1.188580e+03   93.609225 150.963355  109.319341 -1.666156e+02 2.426508e+00 1.693763e+00
              PANAMÁ       Servicios Extra    yhat_reconc_p11  480.0 4.614420e+02 1.190097e+03   93.726149 141.680637  109.392817 -1.679184e+02 2.427938e+00 1.694902e+00
              PANAMÁ           Suministros yhat_reconc_p12act  444.0 1.899531e+03 3.382916e+03   14.145671  17.269090   14.124891 -8.224103e+02 8.878280e-01 7.773480e-01
              PANAMÁ           Suministros    yhat_reconc_p11  444.0 1.900735e+03 3.383862e+03   14.161453  17.279268   14.133845 -8.200811e+02 8.897528e-01 7.778408e-01
              PANAMÁ           Suministros yhat_reconc_p12std  444.0 1.910780e+03 3.398735e+03   14.182591  17.370231   14.208537 -8.592755e+02 8.971305e-01 7.819514e-01
              PANAMÁ           Suministros    yhat_reconc_p10  444.0 2.085373e+03 4.598738e+03   14.803782  17.600825   15.506813 -6.496110e+02 9.346436e-01 8.534006e-01
                PERÚ Eficiencia Energética    yhat_reconc_p11  120.0 2.232143e+01 1.546474e+02  100.000000 200.000000  100.000000 -2.232143e+01 2.232143e+09 8.928571e+08
                PERÚ Eficiencia Energética yhat_reconc_p12act  120.0 2.232143e+01 1.546474e+02  100.000000 200.000000  100.000000 -2.232143e+01 2.232143e+09 8.928571e+08
                PERÚ Eficiencia Energética    yhat_reconc_p10  120.0 2.694043e+01 1.534520e+02   98.822603 199.870700  120.693127 -1.717680e+01 2.694043e+09 1.077617e+09
                PERÚ Eficiencia Energética yhat_reconc_p12std  120.0 3.467278e+02 3.556328e+02   64.693662 197.825531 1553.340390  3.178467e+02 3.467278e+10 1.386911e+10
                PERÚ             Licencias    yhat_reconc_p10   12.0          NaN          NaN         NaN        NaN         NaN           NaN          NaN 0.000000e+00
                PERÚ             Licencias    yhat_reconc_p11   12.0          NaN          NaN         NaN        NaN         NaN           NaN          NaN 0.000000e+00
                PERÚ             Licencias yhat_reconc_p12act   12.0          NaN          NaN         NaN        NaN         NaN           NaN          NaN 0.000000e+00
                PERÚ             Licencias yhat_reconc_p12std   12.0          NaN          NaN         NaN        NaN         NaN           NaN          NaN 0.000000e+00
                PERÚ       Mtto. Contratos    yhat_reconc_p11  240.0 1.342286e+02 3.048659e+02   94.041481 164.135807  108.552129 -4.726695e+01 8.262865e-01 9.558674e-01
                PERÚ       Mtto. Contratos    yhat_reconc_p10  240.0 1.393512e+02 2.822793e+02  105.181445 177.115518  112.694842  3.200905e+00 8.899263e-01 9.923465e-01
                PERÚ       Mtto. Contratos yhat_reconc_p12act  240.0 1.399238e+02 3.013646e+02   96.556927 161.118637  113.157858 -2.705981e+01 8.674427e-01 9.964236e-01
                PERÚ       Mtto. Contratos yhat_reconc_p12std  240.0 1.455293e+02 2.971595e+02   93.408127 178.724346  117.691088 -1.981663e+01 9.309128e-01 1.036341e+00
                PERÚ      Mtto. Correctivo    yhat_reconc_p10  264.0 5.673446e+02 9.001501e+02  173.831241  84.490067   61.199668 -1.228779e+02 6.714489e+08 9.958465e-01
                PERÚ      Mtto. Correctivo    yhat_reconc_p11  264.0 5.729770e+02 9.202647e+02  168.351561  80.305813   61.807244 -1.497527e+02 6.737184e+08 1.005733e+00
                PERÚ      Mtto. Correctivo yhat_reconc_p12act  264.0 5.731183e+02 9.161022e+02  171.487454  80.006859   61.822482 -1.409987e+02 6.737184e+08 1.005981e+00
                PERÚ      Mtto. Correctivo yhat_reconc_p12std  264.0 5.731815e+02 9.060736e+02  179.241892  86.358858   61.829299 -1.170435e+02 7.804287e+08 1.006092e+00
                PERÚ                 Obras    yhat_reconc_p11   24.0 0.000000e+00 0.000000e+00         NaN        NaN         NaN  0.000000e+00 0.000000e+00 0.000000e+00
                PERÚ                 Obras yhat_reconc_p12act   24.0 0.000000e+00 0.000000e+00         NaN        NaN         NaN  0.000000e+00 0.000000e+00 0.000000e+00
                PERÚ                 Obras    yhat_reconc_p10   24.0 5.254146e-11 9.816350e-11         NaN 200.000000         NaN  5.254146e-11 5.254146e-03 5.254146e-03
                PERÚ                 Obras yhat_reconc_p12std   24.0 4.419319e+00 1.530897e+01         NaN 200.000000         NaN  4.419319e+00 4.419319e+08 4.419319e+08
                PERÚ       Servicios Ctto. yhat_reconc_p12std  252.0 1.801628e+02 5.115126e+02   48.512657  95.652572   65.568485 -1.428097e+02 1.720110e+00 2.707069e+00
                PERÚ       Servicios Ctto. yhat_reconc_p12act  252.0 1.802323e+02 5.118008e+02   48.540786  90.511150   65.593751 -1.429818e+02 1.720164e+00 2.708112e+00
                PERÚ       Servicios Ctto.    yhat_reconc_p11  252.0 1.803013e+02 5.121327e+02   48.554952  90.564413   65.618869 -1.431665e+02 1.720313e+00 2.709149e+00
                PERÚ       Servicios Ctto.    yhat_reconc_p10  252.0 1.803374e+02 5.109301e+02   49.342854  94.144298   65.632021 -1.382955e+02 1.721000e+00 2.709692e+00
                PERÚ       Servicios Extra    yhat_reconc_p11  240.0 7.825841e+01 1.661053e+02   92.468819 117.805026  109.410023 -9.556473e-01 3.642955e-01 3.735150e-01
                PERÚ       Servicios Extra yhat_reconc_p12act  240.0 7.826297e+01 1.661090e+02   92.476617 117.798348  109.416405 -9.406693e-01 3.643230e-01 3.735368e-01
                PERÚ       Servicios Extra yhat_reconc_p12std  240.0 7.826441e+01 1.661048e+02   92.472918 129.440318  109.418423 -9.382836e-01 3.643385e-01 3.735437e-01
                PERÚ       Servicios Extra    yhat_reconc_p10  240.0 7.846058e+01 1.660491e+02   92.863194 128.097864  109.692679 -3.473198e-01 3.651818e-01 3.744800e-01
                PERÚ           Suministros yhat_reconc_p12std  240.0 5.250505e+02 1.106467e+03    8.176102  11.287446   10.386624  1.479579e+02 3.703562e-01 3.453627e-01
                PERÚ           Suministros yhat_reconc_p12act  240.0 5.273371e+02 1.117389e+03    8.190363  11.292858   10.431857  1.548806e+02 3.716885e-01 3.468668e-01
                PERÚ           Suministros    yhat_reconc_p11  240.0 5.274492e+02 1.118737e+03    8.189844  11.291893   10.434074  1.553164e+02 3.717653e-01 3.469405e-01
                PERÚ           Suministros    yhat_reconc_p10  240.0 5.481222e+02 1.220131e+03    8.306354  11.396462   10.843031  1.716174e+02 3.847903e-01 3.605386e-01
REPÚBLICA DOMINICANA       Mtto. Contratos    yhat_reconc_p11   60.0 1.908080e+01 2.427800e+01   11.282162  45.233373   17.716734  1.908080e+01 8.461538e+08 1.719753e+00
REPÚBLICA DOMINICANA       Mtto. Contratos    yhat_reconc_p10   60.0 1.913037e+01 2.431592e+01   11.329943  45.270453   17.762759  1.913037e+01 8.481249e+08 1.724220e+00
REPÚBLICA DOMINICANA       Mtto. Contratos yhat_reconc_p12std   60.0 4.244583e+01 4.509879e+01   37.881797  61.580149   39.411422  4.244583e+01 1.327586e+09 3.825643e+00
REPÚBLICA DOMINICANA       Mtto. Contratos yhat_reconc_p12act   60.0 5.044228e+01 5.273298e+01   47.072963  66.194538   46.836215  5.044228e+01 1.492351e+09 4.546363e+00
REPÚBLICA DOMINICANA      Mtto. Correctivo    yhat_reconc_p11   60.0 3.030747e+03 4.480070e+03  227.590802  71.267040   81.401150  2.413365e+03 9.180218e-01 1.036302e+00
REPÚBLICA DOMINICANA      Mtto. Correctivo yhat_reconc_p12act   60.0 3.032668e+03 4.484630e+03  227.935220  71.276482   81.452756  2.418622e+03 9.185722e-01 1.036959e+00
REPÚBLICA DOMINICANA      Mtto. Correctivo yhat_reconc_p12std   60.0 3.042176e+03 4.495613e+03  228.823392  71.397801   81.708134  2.431892e+03 9.216547e-01 1.040210e+00
REPÚBLICA DOMINICANA      Mtto. Correctivo    yhat_reconc_p10   60.0 3.353584e+03 5.002303e+03  238.655315  73.090788   90.072070  2.750004e+03 9.874069e-01 1.146690e+00
REPÚBLICA DOMINICANA       Servicios Ctto.    yhat_reconc_p11   60.0 3.830549e+02 7.529380e+02    8.398192  44.652899    8.717037 -1.124331e+02 4.089023e+00 4.422237e-01
REPÚBLICA DOMINICANA       Servicios Ctto. yhat_reconc_p12act   60.0 3.855593e+02 7.545538e+02    8.308262  44.575566    8.774028 -1.092769e+02 4.204374e+00 4.451149e-01
REPÚBLICA DOMINICANA       Servicios Ctto. yhat_reconc_p12std   60.0 3.878831e+02 7.560343e+02    8.252380  44.529030    8.826911 -1.065438e+02 4.311360e+00 4.477977e-01
REPÚBLICA DOMINICANA       Servicios Ctto.    yhat_reconc_p10   60.0 4.594580e+02 7.956928e+02   10.988926  45.238182   10.455713  1.036926e+02 4.248579e+00 5.304284e-01
REPÚBLICA DOMINICANA       Servicios Extra    yhat_reconc_p11   60.0 1.886345e+02 3.647939e+02  157.927299 175.538037  199.168880  9.116547e+01 1.128140e+00 8.307934e-01
REPÚBLICA DOMINICANA       Servicios Extra yhat_reconc_p12act   60.0 1.887183e+02 3.648411e+02  157.949094 175.526483  199.257384  9.126794e+01 1.130918e+00 8.311626e-01
REPÚBLICA DOMINICANA       Servicios Extra yhat_reconc_p12std   60.0 1.887233e+02 3.648282e+02  157.904547 182.626676  199.262639  9.129053e+01 1.130825e+00 8.311845e-01
REPÚBLICA DOMINICANA       Servicios Extra    yhat_reconc_p10   60.0 1.908413e+02 3.671794e+02  159.116394 181.707282  201.498950  9.400661e+01 1.136945e+00 8.405128e-01
REPÚBLICA DOMINICANA           Suministros yhat_reconc_p12std   60.0 7.747158e+03 1.331222e+04   31.506105  89.599952   57.434841  3.441033e+03 1.320262e+00 1.317660e+00
REPÚBLICA DOMINICANA           Suministros yhat_reconc_p12act   60.0 7.793528e+03 1.333558e+04   32.076883  89.805474   57.778613  3.501500e+03 1.335461e+00 1.325546e+00
REPÚBLICA DOMINICANA           Suministros    yhat_reconc_p11   60.0 7.797124e+03 1.333897e+04   32.106415  89.818076   57.805277  3.505306e+03 1.336526e+00 1.326158e+00
REPÚBLICA DOMINICANA           Suministros    yhat_reconc_p10   60.0 1.231737e+04 2.624850e+04   33.930902  91.402644   91.316865  6.660176e+03 1.774957e+00 2.094975e+00

# Ruta del notebook .ipynb que vamos a convertir

RUTA_NOTEBOOK = "/content/drive/MyDrive/PROYECTO_NEITH/DOCUMENTOS_PROYECTOS/Neith.ipynb"

# Carpeta de salida y nombre del HTML
OUTPUT_DIR = "/content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO"
OUTPUT_HTML = "Neith_notebook_v5_estrategia_3_paso_de_1_a_12_con_metrica.html"

# Aseguramos carpeta de salida (ya en cabecera)
# import os
os.makedirs(OUTPUT_DIR, exist_ok=True)

# Instalamos nbconvert (ya en cabecera)
# !pip install -q "nbconvert>=7.0.0"

# Convertimos notebook a HTML y lo guardamos con el nombre indicado en la carpeta de salida
!jupyter nbconvert --to html "$RUTA_NOTEBOOK" --output "$OUTPUT_HTML" --output-dir "$OUTPUT_DIR"

# Verificamos
salida = os.path.join(OUTPUT_DIR, OUTPUT_HTML)
print("Archivo HTML generado en:", salida, "\nExiste:", os.path.exists(salida))

[NbConvertApp] Converting notebook /content/drive/MyDrive/PROYECTO_NEITH/DOCUMENTOS_PROYECTOS/Neith.ipynb to html
[NbConvertApp] WARNING | Alternative text is missing on 60 image(s).
[NbConvertApp] Writing 9085411 bytes to /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/Neith_notebook_v5_estrategia_3_paso_de_1_a_12_con_metrica.html
Archivo HTML generado en: /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/Neith_notebook_v5_estrategia_3_paso_de_1_a_12_con_metrica.html
Existe: True

# ============================================
# PASO 13 — Paquete de entrega + Hoja de ruta
# ============================================
# En este script:
#  - Preparamos el entregable "PRIMER_ENTREGABLE"
#  - Leemos suavemente artefactos de 2024/2025 (sin romper si faltan)
#  - Calculamos métricas/figuras/informes si hay datos
#  - Generamos bundle de inferencia para la web (2024 y 2025)
#  - Creamos placeholders cuando algo falte para mantener contrato estable
#  - Integramos fuentes reales: dim_buildings.csv, ipc_monthly_lookup.csv e ipc_anual_diccionario.csv


# ==================
# 0) PARÁMETROS
# ==================

np.random.seed(7)

# Ruta de ENTRADA (artefactos generados por pasos previos)
RUTA_ENTRADA = "/content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3"

# Ruta del ENTREGABLE (salida de este paso 13)
ruta_base_3 = "/content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3/PRIMER_ENTREGABLE"

# CSV y umbrales
CSV_SEP      = ";"
ZERO_THR     = 1e-9
N_TOP        = 50
UMBRAL_MASE12= 1.0
UMBRAL_SMAPE = 30.0

# Variante final por defecto; degradamos si no existe
VAR_FINAL_DEF = "P12_STD"  # fallback: P11 -> P10 -> BASE

# Candidatas de fecha que solemos ver
DATE_CANDS = ["FECHA","Date","DS","PERIOD","MONTH","Mes","MES"]

# ==================
# 1) CARPETAS SALIDA
# ==================
RUTA_ENTREGABLE = ruta_base_3
CARP_RES   = os.path.join(RUTA_ENTREGABLE, "RESULTADOS")
CARP_MET   = os.path.join(RUTA_ENTREGABLE, "METRICAS")
CARP_FIG   = os.path.join(RUTA_ENTREGABLE, "FIGURAS")
CARP_INF   = os.path.join(RUTA_ENTREGABLE, "INFORMES")
CARP_LOG   = os.path.join(RUTA_ENTREGABLE, "REGISTROS")
CARP_BUNDLE= os.path.join(RUTA_ENTREGABLE, "PAQUETE_INFERENCIA")

for d in [RUTA_ENTREGABLE, CARP_RES, CARP_MET, CARP_FIG, CARP_INF, CARP_LOG, CARP_BUNDLE]:
    os.makedirs(d, exist_ok=True)

# ============
# 2) LOGGING
# ============
LOG_PATH = os.path.join(CARP_LOG, "paso13.log")

def log13(msg: str):
    ts = datetime.now().isoformat(timespec="seconds")
    line = f"[{ts}] {msg}"
    print(line)
    with open(LOG_PATH, "a", encoding="utf-8") as f:
        f.write(line + "\n")

log13("Iniciamos Paso 13 — Paquete de entrega + Hoja de ruta.")

# =====================
# 3) HELPERS GENERALES
# =====================
def _ensure_ms_inplace(df: pd.DataFrame, candidate_cols=DATE_CANDS):
    # Normalizamos la columna de fecha si existe alguna candidata
    if df is None or df.empty: return
    for c in candidate_cols:
        if c in df.columns:
            try:
                df[c] = pd.to_datetime(df[c], errors="coerce").dt.to_period("M").dt.to_timestamp()
            except Exception:
                pass

def _read_csv_soft(path: str, candidate_date_cols=None, sep=CSV_SEP) -> pd.DataFrame:
    # Leemos un CSV sin romper; solo parseamos fechas que EXISTAN
    if not os.path.exists(path):
        rel = path.replace(RUTA_ENTRADA, "").lstrip("/").lstrip("\\")
        log13(f"FALTA {rel}. Seguimos sin romper.")
        return pd.DataFrame()
    try:
        header = pd.read_csv(path, sep=sep, nrows=0)
        header_cols = [str(c).strip().lstrip("\ufeff") for c in header.columns]
        parse_dates = None
        if candidate_date_cols:
            parse_dates = [c for c in candidate_date_cols if c in header_cols] or None
        df = pd.read_csv(path, sep=sep, parse_dates=parse_dates)
        df.columns = [str(c).strip().lstrip("\ufeff") for c in df.columns]
        return df
    except Exception as e:
        log13(f"ERROR leyendo {path}: {e}. Devolvemos DF vacío.")
        return pd.DataFrame()

def _norm_cols(df: pd.DataFrame, variant_tag: str=None):
    # Normalizamos nombres clave y columnas de predicción para cada variante
    if df is None or df.empty:
        # Devolvemos un df vacío con contrato mínimo y pred_col=None
        return pd.DataFrame(columns=["ID_BUILDING","FM_COST_TYPE","FECHA","yhat"]), None

    out = df.copy()
    # Unificamos posibles nombres de columnas clave
    colmap = {}
    for c in out.columns:
        cu = str(c).strip().upper()
        if cu in ("ID_BUILDING","IDBUILDING","BUILDING_ID","ID_EDIFICIO"):
            colmap[c] = "ID_BUILDING"
        elif cu in ("FM_COST_TYPE","FM_COST","FMCOST","COST_TYPE"):
            colmap[c] = "FM_COST_TYPE"
        elif cu in ("FECHA","DATE","DS","PERIOD","MONTH","MES"):
            colmap[c] = "FECHA"
    if colmap:
        out = out.rename(columns=colmap)

    # Deducimos columna de predicción
    pred_col = None
    for candidate in ["yhat_reconc","yhat_combo","yhat","YHAT","pred","PRED","Y","valor","VALOR","forecast"]:
        if candidate in out.columns:
            pred_col = candidate
            break
    if variant_tag and not pred_col:
        tagged = f"yhat_reconc_{variant_tag}"
        if tagged in out.columns:
            pred_col = tagged

    return out, pred_col

def _norm_unpack(df, variant_tag=None):
    # Helper de desempaquetado a prueba de versiones antiguas
    ret = _norm_cols(df, variant_tag)
    if isinstance(ret, tuple) and len(ret) == 2:
        return ret[0], ret[1]
    return ret, None

def _ensure_contract(df: pd.DataFrame, pred_col: str, year: int) -> pd.DataFrame:
    # Aseguramos contrato mínimo: ID_BUILDING, FM_COST_TYPE, FECHA, yhat
    cols = ["ID_BUILDING","FM_COST_TYPE","FECHA", pred_col] if pred_col else ["ID_BUILDING","FM_COST_TYPE","FECHA"]
    df2 = df.copy()[[c for c in cols if c in df.columns]] if not df.empty else pd.DataFrame(columns=["ID_BUILDING","FM_COST_TYPE","FECHA", "yhat"])
    if not df2.empty and pred_col and pred_col != "yhat":
        df2 = df2.rename(columns={pred_col: "yhat"})
    if "FECHA" in df2.columns:
        _ensure_ms_inplace(df2, ["FECHA"])
        # filtramos al año deseado si procede
        try:
            df2 = df2[df2["FECHA"].dt.year.eq(year)]
        except Exception:
            pass
    # Tipos suaves
    if "ID_BUILDING" in df2.columns:
        df2["ID_BUILDING"] = pd.to_numeric(df2["ID_BUILDING"], errors="coerce").astype("Int64")
    if "FM_COST_TYPE" in df2.columns:
        df2["FM_COST_TYPE"] = df2["FM_COST_TYPE"].astype(str).str.strip()
    return df2

def _save_csv(df: pd.DataFrame, path: str):
    os.makedirs(os.path.dirname(path), exist_ok=True)
    df.to_csv(path, sep=CSV_SEP, index=False)
    log13(f"Guardado CSV: {path} ({len(df)} filas).")

def _metrics_block(df: pd.DataFrame, col_pred: str, col_real: str, scale_mase=None):
    if df.empty or col_pred not in df.columns or col_real not in df.columns:
        return {"MAE":np.nan,"RMSE":np.nan,"MAPE%":np.nan,"sMAPE%":np.nan,"WAPE%":np.nan,"Bias":np.nan,"MASE":np.nan,"WASE":np.nan}
    y = pd.to_numeric(df[col_real], errors="coerce")
    yhat = pd.to_numeric(df[col_pred], errors="coerce")
    err = yhat - y
    abs_e = err.abs()
    mae  = abs_e.mean()
    rmse = float(np.sqrt((err**2).mean())) if len(err)>0 else np.nan
    mape = float((100.0 * (abs_e / y.replace(0,np.nan).abs())).mean())
    smape= float((100.0 * (abs_e / ((y.abs() + yhat.abs())/2.0).replace(0,np.nan))).mean())
    wape = float(100.0 * abs_e.sum() / y.abs().sum()) if y.abs().sum()!=0 else np.nan
    bias = err.mean()
    # MASE/WASE
    if scale_mase is not None and "MASE_scale" in scale_mase.columns:
        df_sc = df.merge(scale_mase, on=["ID_BUILDING","FM_COST_TYPE"], how="left")
        scale = pd.to_numeric(df_sc["MASE_scale"], errors="coerce").replace(0,np.nan)
        mase = float((abs_e / scale).mean())
        wase = float(abs_e.sum() / scale.sum()) if scale.sum() not in [0,np.nan] else np.nan
    else:
        mase, wase = np.nan, np.nan
    return {"MAE":mae,"RMSE":rmse,"MAPE%":mape,"sMAPE%":smape,"WAPE%":wape,"Bias":bias,"MASE":mase,"WASE":wase}

def _build_mase_scale(train_df: pd.DataFrame):
    # Construimos denominador MASE estacional-12 con 2021–2023 si existe
    if train_df.empty or "FECHA" not in train_df.columns or "cost_float_mod" not in train_df.columns:
        return pd.DataFrame(columns=["ID_BUILDING","FM_COST_TYPE","MASE_scale"])
    t = train_df.copy()
    _ensure_ms_inplace(t, ["FECHA"])
    t = t.sort_values(["ID_BUILDING","FM_COST_TYPE","FECHA"])
    t["y_lag12"] = t.groupby(["ID_BUILDING","FM_COST_TYPE"])["cost_float_mod"].shift(12)
    t["abs_diff12"] = (t["cost_float_mod"] - t["y_lag12"]).abs()
    sc = (t.groupby(["ID_BUILDING","FM_COST_TYPE"], as_index=False)["abs_diff12"]
            .mean().rename(columns={"abs_diff12":"MASE_scale"}))
    sc["MASE_scale"] = pd.to_numeric(sc["MASE_scale"], errors="coerce").fillna(0.0)
    sc.loc[sc["MASE_scale"]<=0, "MASE_scale"] = 1e-8
    return sc

def _select_variant_availability(base_2024, p10_2024, p11_2024, p12s_2024):
    # Decidimos variante final según disponibilidad de 2024
    availability = {
        "P12_STD": not p12s_2024.empty,
        "P11":     not p11_2024.empty,
        "P10":     not p10_2024.empty,
        "BASE":    not base_2024.empty
    }
    order = ["P12_STD","P11","P10","BASE"]
    chosen = None
    for v in order:
        if availability[v]:
            chosen = v
            break
    if chosen is None:
        log13("AVISO: no hay ninguna variante disponible (ni BASE). El paso continuará con métricas vacías.")
    else:
        log13(f"Variante final seleccionada: {chosen}")
    return chosen, availability

def _schema_md_report(paths_dict: dict, frames_dict: dict, out_md: str):
    # Generamos un informe rápido de esquemas de entrada
    lines = ["# Esquema de entradas — Paso 13\n"]
    for k, p in paths_dict.items():
        lines.append(f"## {k}\n")
        lines.append(f"- Ruta: `{p}`")
        df = frames_dict.get(k, pd.DataFrame())
        if df is None or df.empty:
            lines.append("- Estado: **vacío o ausente**\n")
        else:
            cols = ", ".join(df.columns.tolist())
            n = len(df)
            lines.append(f"- Filas: {n}")
            lines.append(f"- Columnas: {cols}\n")
    with open(out_md, "w", encoding="utf-8") as f:
        f.write("\n".join(lines))
    log13(f"Guardado informe de esquema: {out_md}")

# ========================
# 4) RUTAS DE ENTRADA
# ========================
# 2024
P_BASE = os.path.join(RUTA_ENTRADA, "RESULTADOS", "preds_por_serie_2024.csv")
P_P10  = os.path.join(RUTA_ENTRADA, "RESULTADOS", "preds_reconciliadas_2024.csv")
P_P11  = os.path.join(RUTA_ENTRADA, "RESULTADOS", "preds_reconciliadas_2024_prod.csv")
P_P12S = os.path.join(RUTA_ENTRADA, "RESULTADOS", "preds_reconciliadas_step12.csv")
P_P12A = os.path.join(RUTA_ENTRADA, "RESULTADOS", "preds_reconciliadas_step12_activeonly.csv")
P_COMP = os.path.join(RUTA_ENTRADA, "REPORTING", "compare_recon_steps_bottom.csv")

P_REAL = os.path.join(RUTA_ENTRADA, "RESULTADOS", "test_full_2024.csv")
P_TRN  = os.path.join(RUTA_ENTRADA, "RESULTADOS", "train_full_2021_2023.csv")

# Dimensiones e IPC (fuentes reales de tu entorno)
P_DIM      = os.path.join(RUTA_ENTRADA, "METRICAS", "dim_buildings.csv")        # legacy real
P_DIM_ES   = os.path.join(RUTA_ENTRADA, "METRICAS", "dim_edificios.csv")        # si existiera
P_P8       = os.path.join(RUTA_ENTRADA, "METRICAS", "paso8_metrics_2024_por_pareja.csv")
P_IPC      = os.path.join(RUTA_ENTRADA, "METRICAS", "ipc_monthly_lookup.csv")   # real
P_IPC_ES   = os.path.join(RUTA_ENTRADA, "METRICAS", "ipc_mensual.csv")          # si existiera
P_IPC25    = os.path.join(RUTA_ENTRADA, "METRICAS", "ipc_mensual_2025.csv")     # si ya existiera
P_IPC_ANO  = os.path.join(RUTA_ENTRADA, "METRICAS", "ipc_anual_diccionario.csv")# real anual (fallback 2025)

# 2025 (si ya existen)
P_BASE_2025 = os.path.join(RUTA_ENTRADA, "RESULTADOS", "preds_base_2025.csv")
P_P11_2025  = os.path.join(RUTA_ENTRADA, "RESULTADOS", "preds_reconciliadas_2025_P11.csv")
P_P12_2025  = os.path.join(RUTA_ENTRADA, "RESULTADOS", "preds_reconciliadas_2025_P12STD.csv")

# ==========================
# 5) LECTURA SUAVE ENTRADAS
# ==========================
df_base  = _read_csv_soft(P_BASE,  candidate_date_cols=DATE_CANDS)
df_p10   = _read_csv_soft(P_P10,   candidate_date_cols=DATE_CANDS)
df_p11   = _read_csv_soft(P_P11,   candidate_date_cols=DATE_CANDS)
df_p12s  = _read_csv_soft(P_P12S,  candidate_date_cols=DATE_CANDS)
df_p12a  = _read_csv_soft(P_P12A,  candidate_date_cols=DATE_CANDS)

df_comp  = _read_csv_soft(P_COMP,  candidate_date_cols=DATE_CANDS)
df_real  = _read_csv_soft(P_REAL,  candidate_date_cols=DATE_CANDS)
df_train = _read_csv_soft(P_TRN,   candidate_date_cols=DATE_CANDS)

# Dimensiones: preferimos españolizado si existe
df_dim = _read_csv_soft(P_DIM_ES)
if df_dim.empty:
    df_dim = _read_csv_soft(P_DIM)

# IPC: preferimos españolizado si existe; si no, monthly real
df_ipc  = _read_csv_soft(P_IPC_ES, candidate_date_cols=DATE_CANDS)
if df_ipc.empty:
    df_ipc = _read_csv_soft(P_IPC, candidate_date_cols=DATE_CANDS)

# IPC 2025: si no existe mensual 2025, fabricamos desde el diccionario anual
df_ipc25 = _read_csv_soft(P_IPC25, candidate_date_cols=DATE_CANDS)
if df_ipc25.empty and os.path.exists(P_IPC_ANO):
    anual = _read_csv_soft(P_IPC_ANO)
    if not anual.empty:
        cols_up = {c.upper(): c for c in anual.columns}
        df_ipc25 = pd.DataFrame()
        # Caso largo: PAIS, ANIO, valor
        if {"PAIS","ANIO"}.issubset(set(cols_up.keys())):
            c_pais = cols_up["PAIS"]; c_year = cols_up["ANIO"]
            c_val  = None
            for k in ("IPC_YOY_PCT","IPC","IPC_ANUAL_%","YOY","VAR_INTERANUAL"):
                if k in cols_up:
                    c_val = cols_up[k]; break
            if c_val:
                tmp = anual[[c_pais, c_year, c_val]].copy()
                tmp.columns = ["PAIS","ANIO","IPC_YOY_PCT"]
                pref_year = 2025 if (tmp["ANIO"]==2025).any() else 2024
                tmp = tmp[tmp["ANIO"]==pref_year].copy()
                if not tmp.empty:
                    tmp["FECHA"] = pd.Timestamp("2025-01-01")
                    df_ipc25 = tmp[["PAIS","FECHA","IPC_YOY_PCT"]].copy()
        else:
            # Caso ancho: columnas de años
            if "PAIS" in cols_up:
                c_pais = cols_up["PAIS"]
                year_cols = []
                for c in anual.columns:
                    s = str(c).strip()
                    if s.isdigit() and len(s)==4:
                        try:
                            y = int(s)
                            if 2000 <= y <= 2100:
                                year_cols.append(c)
                        except:
                            pass
                if year_cols:
                    pick = "2025" if "2025" in [str(c) for c in year_cols] else ("2024" if "2024" in [str(c) for c in year_cols] else None)
                    if pick is not None and pick in anual.columns:
                        tmp = anual[[c_pais, pick]].copy()
                        tmp.columns = ["PAIS","IPC_YOY_PCT"]
                        tmp["FECHA"] = pd.Timestamp("2025-01-01")
                        df_ipc25 = tmp[["PAIS","FECHA","IPC_YOY_PCT"]].copy()
    if df_ipc25.empty:
        log13("AVISO: no fabricamos ipc_mensual_2025.csv (no pudimos inferir del diccionario anual).")

# Informe de esquemas
_schema_md_report(
    {
        "BASE_2024":P_BASE, "P10_2024":P_P10, "P11_2024":P_P11, "P12_STD_2024":P_P12S, "P12_ACTIVE_2024":P_P12A,
        "COMPARATIVA_BOTTOM":P_COMP, "REAL_2024":P_REAL, "TRAIN_2021_2023":P_TRN,
        "DIM_EDIFICIOS": (P_DIM_ES if os.path.exists(P_DIM_ES) else P_DIM),
        "PASO8_METRICAS":P_P8, "IPC_MENSUAL": (P_IPC_ES if os.path.exists(P_IPC_ES) else P_IPC),
        "IPC_MENSUAL_2025":(P_IPC25 if os.path.exists(P_IPC25) else P_IPC_ANO),
        "BASE_2025":P_BASE_2025, "P11_2025":P_P11_2025, "P12_2025":P_P12_2025
    },
    {
        "BASE_2024":df_base, "P10_2024":df_p10, "P11_2024":df_p11, "P12_STD_2024":df_p12s, "P12_ACTIVE_2024":df_p12a,
        "COMPARATIVA_BOTTOM":df_comp, "REAL_2024":df_real, "TRAIN_2021_2023":df_train,
        "DIM_EDIFICIOS":df_dim, "PASO8_METRICAS":_read_csv_soft(P_P8),
        "IPC_MENSUAL":df_ipc, "IPC_MENSUAL_2025":df_ipc25 if not df_ipc25.empty else _read_csv_soft(P_IPC_ANO),
        "BASE_2025":_read_csv_soft(P_BASE_2025, DATE_CANDS), "P11_2025":_read_csv_soft(P_P11_2025, DATE_CANDS),
        "P12_2025":_read_csv_soft(P_P12_2025, DATE_CANDS)
    },
    os.path.join(CARP_INF, "esquema_entradas.md")
)

# Normalizamos esquemas + extraemos columnas de predicción (para 2024)
df_base, base_pred_col = _norm_cols(df_base, None)
df_p10,  p10_pred_col  = _norm_cols(df_p10,  "p10")
df_p11,  p11_pred_col  = _norm_cols(df_p11,  "p11")
df_p12s, p12s_pred_col = _norm_cols(df_p12s, "p12std")
df_p12a, p12a_pred_col = _norm_cols(df_p12a, "p12act")

# Ajustamos contrato al año 2024 por variante
base_2024 = _ensure_contract(df_base, base_pred_col, 2024)
p10_2024  = _ensure_contract(df_p10,  p10_pred_col  or "yhat_reconc", 2024)
p11_2024  = _ensure_contract(df_p11,  p11_pred_col  or "yhat_reconc", 2024)
p12s_2024 = _ensure_contract(df_p12s, p12s_pred_col or "yhat_reconc", 2024)
p12a_2024 = _ensure_contract(df_p12a, p12a_pred_col or "yhat_reconc", 2024)

# Construimos escala MASE si tenemos TRAIN
mase_scale = pd.DataFrame()
if not df_train.empty and "cost_float_mod" in df_train.columns:
    _ensure_ms_inplace(df_train, ["FECHA"])
    mase_scale = _build_mase_scale(df_train)

# ===============================
# 6) MÉTRICAS CLAVE (si hay real)
# ===============================
# Real 2024
real_2024 = pd.DataFrame()
if not df_real.empty and "cost_float_mod" in df_real.columns:
    real_2024 = df_real.copy()
    _ensure_ms_inplace(real_2024, ["FECHA"])
    real_2024 = real_2024.rename(columns={"cost_float_mod":"y_real"})

# Merge helper
def _merge_real(pred_df):
    if pred_df.empty or real_2024.empty:
        return pd.DataFrame()
    cols = ["ID_BUILDING","FM_COST_TYPE","FECHA","yhat"]
    m = pred_df[cols].merge(real_2024[["ID_BUILDING","FM_COST_TYPE","FECHA","y_real"]],
                            on=["ID_BUILDING","FM_COST_TYPE","FECHA"], how="inner")
    return m

# Elegimos variante final
chosen_var, availability = _select_variant_availability(base_2024, p10_2024, p11_2024, p12s_2024)

# Construimos tabla de métricas globales por variante disponible
metric_rows = []
var_map = {
    "BASE": base_2024, "P10": p10_2024, "P11": p11_2024, "P12_STD": p12s_2024, "P12_ACTIVE": p12a_2024
}
for vname, vdf in var_map.items():
    if vdf.empty:
        continue
    m = _merge_real(vdf) if not real_2024.empty else pd.DataFrame()
    if not m.empty:
        met = _metrics_block(m, "yhat", "y_real", scale_mase=mase_scale)
    else:
        met = {"MAE":np.nan,"RMSE":np.nan,"MAPE%":np.nan,"sMAPE%":np.nan,"WAPE%":np.nan,"Bias":np.nan,"MASE":np.nan,"WASE":np.nan}
    # Cobertura
    if not vdf.empty and not real_2024.empty:
        merged = vdf.merge(real_2024[["ID_BUILDING","FM_COST_TYPE","FECHA"]], on=["ID_BUILDING","FM_COST_TYPE","FECHA"], how="left")
        cobertura = 100.0 * merged["FECHA_y"].notna().mean() if "FECHA_y" in merged.columns else np.nan
    else:
        cobertura = np.nan
    row = {"Variante": vname, **met, "Cobertura%": cobertura}
    metric_rows.append(row)

df_metrics_global = pd.DataFrame(metric_rows)
_save_csv(df_metrics_global, os.path.join(CARP_MET, "metricas_globales_2024.csv"))

# Métricas por FM_COST_TYPE × PAIS (si tenemos real y dim)
df_metrics_seg = pd.DataFrame()
if not real_2024.empty and not df_dim.empty:
    dim = df_dim.copy()
    # Normalizamos PAIS/REGION/ID_BUILDING
    colmap_dim = {}
    for c in dim.columns:
        cu = str(c).upper()
        if cu in ("ID_BUILDING","IDBUILDING","BUILDING_ID","ID_EDIFICIO"):
            colmap_dim[c] = "ID_BUILDING"
        elif cu in ("PAIS","COUNTRY","COUNTRY_DEF"):
            colmap_dim[c] = "PAIS"
        elif cu in ("REGION","REGIÓN"):
            colmap_dim[c] = "REGION"
        elif cu in ("STATUS","ESTADO"):
            colmap_dim[c] = "STATUS"
    if colmap_dim:
        dim = dim.rename(columns=colmap_dim)

    for vname, vdf in var_map.items():
        if vdf.empty:
            continue
        merged = vdf.merge(dim[["ID_BUILDING","PAIS"]] if "PAIS" in dim.columns else vdf[["ID_BUILDING"]].assign(PAIS="SIN_PAIS"),
                           on="ID_BUILDING", how="left")
        merged = merged.merge(real_2024[["ID_BUILDING","FM_COST_TYPE","FECHA","y_real"]],
                              on=["ID_BUILDING","FM_COST_TYPE","FECHA"], how="inner")
        if merged.empty:
            continue
        rows = []
        for (fmc, pais), g in merged.groupby(["FM_COST_TYPE","PAIS"] if "PAIS" in merged.columns else ["FM_COST_TYPE"]):
            met = _metrics_block(g, "yhat","y_real", scale_mase=mase_scale)
            base = {"Variante": vname, "FM_COST_TYPE":fmc}
            if "PAIS" in merged.columns:
                base["PAIS"]=pais
            rows.append({**base, **met})
        if rows:
            segi = pd.DataFrame(rows)
            df_metrics_seg = pd.concat([df_metrics_seg, segi], ignore_index=True)
else:
    log13("AVISO: sin reales 2024 no generamos métricas segmentadas.")

_save_csv(df_metrics_seg, os.path.join(CARP_MET, "metricas_por_fmcost_pais_2024.csv"))

# Subconjunto estable 2024
df_subset = pd.DataFrame()
if chosen_var is not None and not real_2024.empty:
    chosen_df = var_map.get(chosen_var, pd.DataFrame())
    merged = _merge_real(chosen_df)
    if not merged.empty:
        rows = []
        ms = mase_scale if not mase_scale.empty else None
        for (bid, fmc), g in merged.groupby(["ID_BUILDING","FM_COST_TYPE"]):
            met = _metrics_block(g, "yhat","y_real", scale_mase=ms)
            rows.append({"ID_BUILDING":bid,"FM_COST_TYPE":fmc, **met})
        subm = pd.DataFrame(rows)
        mask = (subm["MASE"]<=UMBRAL_MASE12) & (subm["sMAPE%"]<=UMBRAL_SMAPE)
        df_subset = subm.loc[mask].copy()
else:
    log13("AVISO: no tenemos variante elegida y real suficiente para subset estable.")

_save_csv(df_subset, os.path.join(CARP_MET, "subconjunto_estable_2024.csv"))

# Top-N (si tenemos métricas por pareja)
if not df_subset.empty or (chosen_var is not None and not real_2024.empty):
    if df_subset.empty and chosen_var is not None and not real_2024.empty:
        chosen_df = var_map.get(chosen_var, pd.DataFrame())
        merged = _merge_real(chosen_df)
        rows = []
        for (bid, fmc), g in merged.groupby(["ID_BUILDING","FM_COST_TYPE"]):
            met = _metrics_block(g, "yhat","y_real", scale_mase=mase_scale if not mase_scale.empty else None)
            rows.append({"ID_BUILDING":bid,"FM_COST_TYPE":fmc, **met})
        df_pairs = pd.DataFrame(rows)
    else:
        df_pairs = df_subset.copy()
    if "WAPE%" in df_pairs.columns:
        top_mej = df_pairs.sort_values("WAPE%", ascending=True).head(N_TOP)
        top_peor= df_pairs.sort_values("WAPE%", ascending=False).head(N_TOP)
        _save_csv(top_mej, os.path.join(CARP_MET, "top_mejores_wape.csv"))
        _save_csv(top_peor, os.path.join(CARP_MET, "top_peores_wape.csv"))
    else:
        _save_csv(pd.DataFrame(), os.path.join(CARP_MET, "top_mejores_wape.csv"))
        _save_csv(pd.DataFrame(), os.path.join(CARP_MET, "top_peores_wape.csv"))
else:
    _save_csv(pd.DataFrame(), os.path.join(CARP_MET, "top_mejores_wape.csv"))
    _save_csv(pd.DataFrame(), os.path.join(CARP_MET, "top_peores_wape.csv"))

# Top outliers por delta P12 (si tenemos comparativa bottom)
top_out = pd.DataFrame()
if not df_comp.empty:
    d = df_comp.copy()
    for cand in ["delta_p12std","delta_p12","delta_p11","delta_p10"]:
        if cand in d.columns:
            d["abs_delta"] = pd.to_numeric(d[cand], errors="coerce").abs()
            top_out = d.sort_values("abs_delta", ascending=False).head(N_TOP)
            break
_save_csv(top_out, os.path.join(CARP_MET, "top_outliers_delta_p12.csv"))

# =================
# 7) FIGURAS
# =================
# 7.1 WAPE portfolio mensual (variante elegida) si tenemos real
if chosen_var is not None and not real_2024.empty:
    chosen_df = var_map.get(chosen_var, pd.DataFrame())
    m = _merge_real(chosen_df)
    if not m.empty:
        by_m = (m.groupby("FECHA", as_index=False)
                  .apply(lambda g: pd.Series({"WAPE%": (100.0 * (g["yhat"]-g["y_real"]).abs().sum() /
                                                       g["y_real"].abs().sum()) if g["y_real"].abs().sum()!=0 else np.nan})))
        plt.figure(figsize=(10,5))
        plt.plot(by_m["FECHA"], by_m["WAPE%"])
        plt.title(f"WAPE mensual portfolio — {chosen_var} (2024)")
        plt.xlabel("Mes"); plt.ylabel("WAPE %")
        plt.tight_layout()
        plt.savefig(os.path.join(CARP_FIG, "wape_portfolio_mensual.png"), dpi=150)
        plt.close()
    else:
        log13("AVISO: sin merge real no generamos figura wape_portfolio_mensual.")
else:
    log13("AVISO: sin reales 2024 no generamos figura wape_portfolio_mensual.")

# 7.2 Barras WAPE por FM_COST_TYPE (si tenemos real y dim)
if chosen_var is not None and not real_2024.empty:
    chosen_df = var_map.get(chosen_var, pd.DataFrame())
    merged = chosen_df.merge(real_2024, on=["ID_BUILDING","FM_COST_TYPE","FECHA"], how="inner")
    if not merged.empty:
        w_rows = []
        for fmc, g in merged.groupby("FM_COST_TYPE"):
            den = g["y_real"].abs().sum()
            wape = 100.0 * (g["yhat"]-g["y_real"]).abs().sum() / den if den!=0 else np.nan
            w_rows.append({"FM_COST_TYPE":fmc,"WAPE%":wape})
        wdf = pd.DataFrame(w_rows).sort_values("WAPE%", ascending=False)
        plt.figure(figsize=(10,5))
        plt.bar(wdf["FM_COST_TYPE"], wdf["WAPE%"])
        plt.xticks(rotation=45, ha="right")
        plt.title(f"WAPE por FM_COST_TYPE — {chosen_var} (2024)")
        plt.ylabel("WAPE %")
        plt.tight_layout()
        plt.savefig(os.path.join(CARP_FIG, "wape_por_fmcost.png"), dpi=150)
        plt.close()
    else:
        log13("AVISO: sin merge real no generamos figura wape_por_fmcost.")
else:
    log13("AVISO: sin reales 2024 no generamos figura wape_por_fmcost.")

# 7.3 Boxplot de deltas de reconciliación (si hay comparativa)
if not df_comp.empty:
    d = df_comp.copy()
    delta_cols = [c for c in ["delta_p10","delta_p11","delta_p12","delta_p12std","delta_p12act"] if c in d.columns]
    if delta_cols:
        plt.figure(figsize=(9,6))
        plt.boxplot([pd.to_numeric(d[c], errors="coerce").dropna().values for c in delta_cols], labels=delta_cols, showfliers=False)
        plt.axhline(0, color="grey", linestyle=":")
        plt.title("Distribución de deltas bottom (reconc - base)")
        plt.ylabel("Delta")
        plt.tight_layout()
        plt.savefig(os.path.join(CARP_FIG, "boxplot_deltas_reconciliacion.png"), dpi=150)
        plt.close()
    else:
        log13("AVISO: no fue posible generar boxplot de deltas (faltan columnas delta_*).")
else:
    log13("AVISO: comparativa bottom ausente; no generamos boxplot de deltas.")

# =================
# 8) INFORMES
# =================
# 8.1 Resumen ejecutivo (1 página)
resumen_lines = [
"# Resumen ejecutivo — Paso 13",
"",
"Este documento recoge las cifras clave de desempeño 2024, el subconjunto estable, oportunidades rápidas (quick-wins) y las referencias a los principales artefactos del entregable.",
"",
"## 1) Cinco cifras clave (2024)",
]
if not df_metrics_global.empty:
    row = None
    if chosen_var is not None:
        rows = df_metrics_global[df_metrics_global["Variante"].eq(chosen_var)]
        if not rows.empty: row = rows.iloc[0]
    if row is None:
        row = df_metrics_global.iloc[0]
    resumen_lines += [
        f"- **Variante**: {row['Variante']}",
        f"- **WAPE**: {row['WAPE%']:.2f}%",
        f"- **sMAPE**: {row['sMAPE%']:.2f}%",
        f"- **MASE**: {row['MASE']:.4f}",
        f"- **Cobertura**: {row['Cobertura%']:.2f}%"
    ]
else:
    resumen_lines += ["- No disponemos de métricas pobladas (faltan reales o variantes)."]

resumen_lines += [
"",
"## 2) Subconjunto estable",
f"- Fichero: `METRICAS/subconjunto_estable_2024.csv` (filas: {len(df_subset)})",
"",
"## 3) Quick-wins propuestos",
"- Intermitentes/TSB: priorizar pares con alta esporadicidad donde el TSB mejora el error.",
"- Combos/Ensemble: consolidar rutas con mejor MASE/WAPE por segmento.",
"- Outliers (mod_3): revisar top outliers en `METRICAS/top_outliers_delta_p12.csv` y considerar winsorización.",
"",
"## 4) Enlaces a outputs del entregable",
"- `METRICAS/metricas_globales_2024.csv`",
"- `METRICAS/metricas_por_fmcost_pais_2024.csv`",
"- `FIGURAS/wape_portfolio_mensual.png`",
"- `FIGURAS/wape_por_fmcost.png`",
"- `FIGURAS/boxplot_deltas_reconciliacion.png`",
"",
"## 5) Checklist de verificación",
"- Predicciones disponibles para todas las series o fallback documentado.",
"- Ficheros METRICAS completos generados.",
"- Nombres y rutas coherentes con el estándar del proyecto.",
]
with open(os.path.join(CARP_INF, "resumen_ejecutivo.md"), "w", encoding="utf-8") as f:
    f.write("\n".join(resumen_lines))

# 8.2 Checklist (casillas de verificación)
check_lines = [
"# Checklist de verificación — Paso 13",
"",
"- [ ] Predicciones 2024 presentes (BASE/P10/P11/P12_STD) o placeholders creados",
"- [ ] Predicciones 2025 presentes (BASE/P11/P12_STD) o placeholders creados",
"- [ ] Métricas globales 2024 generadas",
"- [ ] Métricas por FM_COST_TYPE × PAIS generadas",
"- [ ] Subconjunto estable generado",
"- [ ] Top-N (mejores/peores WAPE) generados",
"- [ ] Outliers delta P12 disponibles",
"- [ ] Figuras generadas (si hay datos)",
"- [ ] Bundle de inferencia listo (manifest.json, predict.py, matrices si aplica)",
]
with open(os.path.join(CARP_INF, "checklist_verificacion.md"), "w", encoding="utf-8") as f:
    f.write("\n".join(check_lines))

# =========================================
# 9) EXPORT DE INFERENCIA PARA LA WEB
#    (añadimos 2024 y 2025; creamos placeholders si faltan)
# =========================================
# 9.1 Construimos resultados 2024 en ENTREGABLE/RESULTADOS
if base_2024.empty:
    base_2024_export = pd.DataFrame(columns=["ID_BUILDING","FM_COST_TYPE","FECHA","yhat"])
else:
    base_2024_export = base_2024.copy()

_save_csv(base_2024_export, os.path.join(CARP_RES, "preds_base_2024.csv"))
_save_csv(p11_2024 if not p11_2024.empty else pd.DataFrame(columns=["ID_BUILDING","FM_COST_TYPE","FECHA","yhat"]),
          os.path.join(CARP_RES, "preds_reconciliadas_2024_P11.csv"))
_save_csv(p12s_2024 if not p12s_2024.empty else pd.DataFrame(columns=["ID_BUILDING","FM_COST_TYPE","FECHA","yhat"]),
          os.path.join(CARP_RES, "preds_reconciliadas_2024_P12STD.csv"))

# 9.2 Resultados 2025 (intentamos leer de entrada; si no, placeholders)
base_2025_raw = _read_csv_soft(P_BASE_2025, DATE_CANDS); base_2025_df, base25_pred = _norm_unpack(base_2025_raw, None)
base_2025 = _ensure_contract(base_2025_df, base25_pred, 2025)

p11_2025_raw = _read_csv_soft(P_P11_2025, DATE_CANDS); p11_2025_df, p11_25_pred = _norm_unpack(p11_2025_raw, "p11")
p11_2025 = _ensure_contract(p11_2025_df, p11_25_pred or "yhat_reconc", 2025)

p12_2025_raw = _read_csv_soft(P_P12_2025, DATE_CANDS); p12_2025_df, p12_25_pred = _norm_unpack(p12_2025_raw, "p12std")
p12_2025 = _ensure_contract(p12_2025_df, p12_25_pred or "yhat_reconc", 2025)

_save_csv(base_2025 if not base_2025.empty else pd.DataFrame(columns=["ID_BUILDING","FM_COST_TYPE","FECHA","yhat"]),
          os.path.join(CARP_RES, "preds_base_2025.csv"))
_save_csv(p11_2025 if not p11_2025.empty else pd.DataFrame(columns=["ID_BUILDING","FM_COST_TYPE","FECHA","yhat"]),
          os.path.join(CARP_RES, "preds_reconciliadas_2025_P11.csv"))
_save_csv(p12_2025 if not p12_2025.empty else pd.DataFrame(columns=["ID_BUILDING","FM_COST_TYPE","FECHA","yhat"]),
          os.path.join(CARP_RES, "preds_reconciliadas_2025_P12STD.csv"))

# 9.3 Ruteo por par (modelo final, ensemble, fallback, métricas básicas si disponemos)
def _availability_table(df: pd.DataFrame, year: int, variant: str):
    if df.empty:
        return pd.DataFrame(columns=["ID_BUILDING","FM_COST_TYPE","year","variant","available"])
    out = (df.groupby(["ID_BUILDING","FM_COST_TYPE"], as_index=False).size()
             .rename(columns={"size":"n"}))
    out["year"] = year
    out["variant"] = variant
    out["available"] = 1
    return out[["ID_BUILDING","FM_COST_TYPE","year","variant","available"]]

ruteo_parts = []
ruteo_parts.append(_availability_table(base_2024_export, 2024, "BASE"))
ruteo_parts.append(_availability_table(p11_2024,        2024, "P11"))
ruteo_parts.append(_availability_table(p12s_2024,       2024, "P12_STD"))
ruteo_parts.append(_availability_table(base_2025,        2025, "BASE"))
ruteo_parts.append(_availability_table(p11_2025,         2025, "P11"))
ruteo_parts.append(_availability_table(p12_2025,         2025, "P12_STD"))

ruteo = pd.concat([p for p in ruteo_parts if p is not None and not p.empty], ignore_index=True) if any([not p.empty for p in ruteo_parts if p is not None]) else pd.DataFrame(columns=["ID_BUILDING","FM_COST_TYPE","year","variant","available"])
_save_csv(ruteo, os.path.join(CARP_MET, "ruteo_por_par.csv"))

# 9.4 Manifest del bundle
manifest = {
    "generated_at": datetime.now().isoformat(timespec="seconds"),
    "base_path": RUTA_ENTREGABLE,
    "results": {
        "2024": {
            "BASE":   "RESULTADOS/preds_base_2024.csv",
            "P11":    "RESULTADOS/preds_reconciliadas_2024_P11.csv",
            "P12_STD":"RESULTADOS/preds_reconciliadas_2024_P12STD.csv"
        },
        "2025": {
            "BASE":   "RESULTADOS/preds_base_2025.csv",
            "P11":    "RESULTADOS/preds_reconciliadas_2025_P11.csv",
            "P12_STD":"RESULTADOS/preds_reconciliadas_2025_P12STD.csv"
        }
    },
    "metrics": {
        "global_2024": "METRICAS/metricas_globales_2024.csv",
        "seg_2024":    "METRICAS/metricas_por_fmcost_pais_2024.csv",
        "subset_2024": "METRICAS/subconjunto_estable_2024.csv",
        "ruteo":       "METRICAS/ruteo_por_par.csv"
    },
    "figures": {
        "wape_mensual": "FIGURAS/wape_portfolio_mensual.png",
        "wape_fmcost":  "FIGURAS/wape_por_fmcost.png",
        "box_deltas":   "FIGURAS/boxplot_deltas_reconciliacion.png"
    }
}
with open(os.path.join(CARP_BUNDLE, "manifest.json"), "w", encoding="utf-8") as f:
    json.dump(manifest, f, ensure_ascii=False, indent=2)

# 9.5 Copias útiles al entregable (dimensiones e IPC) con nombres españolizados
# Normalizamos dim a español (si procede) y guardamos como dim_edificios.csv
if not df_dim.empty:
    dim_out = df_dim.copy()
    colmap_dim = {}
    for c in dim_out.columns:
        cu = str(c).upper()
        if cu in ("ID_BUILDING","IDBUILDING","BUILDING_ID","ID_EDIFICIO"):
            colmap_dim[c] = "ID_BUILDING"
        elif cu in ("PAIS","COUNTRY","COUNTRY_DEF"):
            colmap_dim[c] = "PAIS"
        elif cu in ("REGION","REGIÓN"):
            colmap_dim[c] = "REGION"
        elif cu in ("STATUS","ESTADO"):
            colmap_dim[c] = "STATUS"
    if colmap_dim:
        dim_out = dim_out.rename(columns=colmap_dim)
    _save_csv(dim_out, os.path.join(CARP_MET, "dim_edificios.csv"))
else:
    _save_csv(pd.DataFrame(), os.path.join(CARP_MET, "dim_edificios.csv"))

# IPC mensual (lookup real) -> ipc_mensual.csv
if not df_ipc.empty:
    # Aseguramos nombres básicos
    ipcout = df_ipc.copy()
    colmap_ipc = {}
    up = {c.upper(): c for c in ipcout.columns}
    if "PAIS" not in up:
        for k in ("COUNTRY","COUNTRY_DEF"):
            if k in up: colmap_ipc[up[k]] = "PAIS"; break
    if "FECHA" not in up:
        for k in ("DATE","DS","PERIOD","MES","MONTH"):
            if k in up: colmap_ipc[up[k]] = "FECHA"; break
    if "IPC_YOY_PCT" not in up:
        for k in ("IPC","IPC_YOY","YOY","VAR_INTERANUAL","IPC_ANUAL_%"):
            if k in up: colmap_ipc[up[k]] = "IPC_YOY_PCT"; break
    if colmap_ipc:
        ipcout = ipcout.rename(columns=colmap_ipc)
    if "FECHA" in ipcout.columns:
        ipcout["FECHA"] = pd.to_datetime(ipcout["FECHA"], errors="coerce").dt.to_period("M").dt.to_timestamp()
    _save_csv(ipcout, os.path.join(CARP_MET, "ipc_mensual.csv"))
else:
    _save_csv(pd.DataFrame(), os.path.join(CARP_MET, "ipc_mensual.csv"))

# IPC 2025 fabricado/real -> ipc_mensual_2025.csv si lo logramos
if not df_ipc25.empty:
    # Aseguramos FECHA mensual (si viniera como 2025-01-01 la dejamos tal cual)
    ip25 = df_ipc25.copy()
    if "FECHA" in ip25.columns:
        ip25["FECHA"] = pd.to_datetime(ip25["FECHA"], errors="coerce")
    _save_csv(ip25, os.path.join(CARP_MET, "ipc_mensual_2025.csv"))
else:
    log13("AVISO: no guardamos ipc_mensual_2025.csv porque no lo pudimos construir ni leer.")

# 9.6 predict.py (para servir desde la web)
predict_py = f'''# predict.py — lector de predicciones 2024/2025 para la web
# Buscamos CSVs en el entregable y devolvemos un DataFrame con (ID_BUILDING, FM_COST_TYPE, FECHA, yhat)
import os, pandas as pd

CSV_SEP = "{CSV_SEP}"

def _auto_base():
    # Detectamos base del paquete partiendo de este archivo
    here = os.path.abspath(os.path.dirname(__file__))
    # El manifest.json está en el mismo directorio
    return os.path.abspath(os.path.join(here, ".."))

def _read(path):
    if not os.path.exists(path):
        return pd.DataFrame(columns=["ID_BUILDING","FM_COST_TYPE","FECHA","yhat"])
    hdr = pd.read_csv(path, sep=CSV_SEP, nrows=0)
    parse_dates = ["FECHA"] if "FECHA" in hdr.columns else None
    df = pd.read_csv(path, sep=CSV_SEP, parse_dates=parse_dates)
    df.columns = [str(c).strip() for c in df.columns]
    # Aseguramos contrato mínimo
    need = ["ID_BUILDING","FM_COST_TYPE","FECHA","yhat"]
    for n in need:
        if n not in df.columns:
            df[n] = pd.Series(dtype="float64")
    return df[need]

def predict(ids, year=2025, variant="P12_STD", ruta_base="__AUTO__"):
    """
    ids: lista de ID_BUILDING
    year: 2024 o 2025
    variant: 'P12_STD' | 'P11' | 'BASE'
    ruta_base: si '__AUTO__', resolvemos base a partir de este paquete
    """
    if ruta_base == "__AUTO__":
        ruta_base = _auto_base()
    # Rutas relativas respecto a la base del entregable
    if year == 2024:
        files = {{
            "BASE":    os.path.join(ruta_base, "RESULTADOS", "preds_base_2024.csv"),
            "P11":     os.path.join(ruta_base, "RESULTADOS", "preds_reconciliadas_2024_P11.csv"),
            "P12_STD": os.path.join(ruta_base, "RESULTADOS", "preds_reconciliadas_2024_P12STD.csv")
        }}
    else:
        files = {{
            "BASE":    os.path.join(ruta_base, "RESULTADOS", "preds_base_2025.csv"),
            "P11":     os.path.join(ruta_base, "RESULTADOS", "preds_reconciliadas_2025_P11.csv"),
            "P12_STD": os.path.join(ruta_base, "RESULTADOS", "preds_reconciliadas_2025_P12STD.csv")
        }}
    path = files.get(variant.upper(), files["BASE"])
    df = _read(path)
    # Filtramos por los IDs solicitados
    if ids is not None and len(ids)>0 and "ID_BUILDING" in df.columns:
        sids = pd.Series(ids).astype("Int64") if hasattr(pd.Series(ids), "astype") else pd.Series(ids)
        df = df[df["ID_BUILDING"].astype("Int64").isin(sids)]
    return df
'''
with open(os.path.join(CARP_BUNDLE, "predict.py"), "w", encoding="utf-8") as f:
    f.write(predict_py)

# (Opcional) Matrices A / W si existieran en entrada (las traemos al bundle)
A_CAND = os.path.join(RUTA_ENTRADA, "METRICAS", "A_matriz.npz")
W_CAND = os.path.join(RUTA_ENTRADA, "METRICAS", "W_diag.npy")
if os.path.exists(A_CAND):
    import shutil
    shutil.copy(A_CAND, os.path.join(CARP_BUNDLE, "A_matriz.npz"))
if os.path.exists(W_CAND):
    import shutil
    shutil.copy(W_CAND, os.path.join(CARP_BUNDLE, "W_diag.npy"))

log13("Paso 13 finalizado.")

[2025-09-27T17:19:04] Iniciamos Paso 13 — Paquete de entrega + Hoja de ruta.
[2025-09-27T17:19:05] FALTA METRICAS/dim_edificios.csv. Seguimos sin romper.
[2025-09-27T17:19:05] FALTA METRICAS/ipc_mensual.csv. Seguimos sin romper.
[2025-09-27T17:19:05] FALTA METRICAS/ipc_mensual_2025.csv. Seguimos sin romper.
[2025-09-27T17:19:05] AVISO: no fabricamos ipc_mensual_2025.csv (no pudimos inferir del diccionario anual).
[2025-09-27T17:19:05] Guardado informe de esquema: /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3/PRIMER_ENTREGABLE/INFORMES/esquema_entradas.md
[2025-09-27T17:19:06] Variante final seleccionada: P12_STD
[2025-09-27T17:19:06] Guardado CSV: /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3/PRIMER_ENTREGABLE/METRICAS/metricas_globales_2024.csv (5 filas).
[2025-09-27T17:19:09] Guardado CSV: /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3/PRIMER_ENTREGABLE/METRICAS/metricas_por_fmcost_pais_2024.csv (215 filas).
[2025-09-27T17:19:36] Guardado CSV: /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3/PRIMER_ENTREGABLE/METRICAS/subconjunto_estable_2024.csv (0 filas).
[2025-09-27T17:19:59] Guardado CSV: /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3/PRIMER_ENTREGABLE/METRICAS/top_mejores_wape.csv (50 filas).
[2025-09-27T17:19:59] Guardado CSV: /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3/PRIMER_ENTREGABLE/METRICAS/top_peores_wape.csv (50 filas).
[2025-09-27T17:19:59] Guardado CSV: /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3/PRIMER_ENTREGABLE/METRICAS/top_outliers_delta_p12.csv (50 filas).
[2025-09-27T17:20:01] Guardado CSV: /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3/PRIMER_ENTREGABLE/RESULTADOS/preds_base_2024.csv (29148 filas).
[2025-09-27T17:20:01] Guardado CSV: /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3/PRIMER_ENTREGABLE/RESULTADOS/preds_reconciliadas_2024_P11.csv (29148 filas).
[2025-09-27T17:20:02] Guardado CSV: /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3/PRIMER_ENTREGABLE/RESULTADOS/preds_reconciliadas_2024_P12STD.csv (29148 filas).
[2025-09-27T17:20:02] Guardado CSV: /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3/PRIMER_ENTREGABLE/RESULTADOS/preds_base_2025.csv (0 filas).
[2025-09-27T17:20:02] Guardado CSV: /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3/PRIMER_ENTREGABLE/RESULTADOS/preds_reconciliadas_2025_P11.csv (0 filas).
[2025-09-27T17:20:02] Guardado CSV: /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3/PRIMER_ENTREGABLE/RESULTADOS/preds_reconciliadas_2025_P12STD.csv (0 filas).
[2025-09-27T17:20:02] Guardado CSV: /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3/PRIMER_ENTREGABLE/METRICAS/ruteo_por_par.csv (7287 filas).
[2025-09-27T17:20:02] Guardado CSV: /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3/PRIMER_ENTREGABLE/METRICAS/dim_edificios.csv (1650 filas).
[2025-09-27T17:20:02] Guardado CSV: /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3/PRIMER_ENTREGABLE/METRICAS/ipc_mensual.csv (12 filas).
[2025-09-27T17:20:02] AVISO: no guardamos ipc_mensual_2025.csv porque no lo pudimos construir ni leer.
[2025-09-27T17:20:02] Paso 13 finalizado.

# ---------- RUTAS ----------
ruta_origen = Path("/content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3/METRICAS/metrics_subset_estable.csv")
ruta_portfolio = ruta_origen.with_name("metrics_subset_estable_portfolio.csv")
ruta_segmento  = ruta_origen.with_name("metrics_subset_estable_segmento.csv")

# ---------- PARSER ROBUSTO ----------
def _smart_split(line: str):
    """Divide una línea usando ';' o '\t' (elige el que más columnas dé)."""
    line = line.rstrip("\n").rstrip("\r")
    if not line:
        return []
    parts_sc = line.split(";")
    parts_tb = line.split("\t")
    # elegimos el separador que produzca MÁS columnas
    return parts_tb if len(parts_tb) > len(parts_sc) else parts_sc

def cargar_metrics_subset_estable(ruta: Path):
    """
    Lee el CSV mixto (portfolio + segmento), aunque haya separadores distintos
    y columnas extra en SEGMENTO.
    Devuelve dos DataFrames: df_portfolio, df_segmento
    """
    # Esquema canonizado
    cols_port = ["FECHA","MAE","SMAPE","WAPE","MASE1","MASE12","n_series_mes","peso_gasto_mes","scope"]
    cols_segm = ["DIM_NAME","DIM_VALUE","FECHA","MAE","SMAPE","WAPE","MASE1","MASE12","n_series_mes","peso_gasto_mes","flag","scope"]

    rows_port, rows_seg = [], []

    with ruta.open("r", encoding="utf-8") as f:
        # Intentamos detectar si hay cabecera en la primera línea
        first = f.readline()
        parts = _smart_split(first)
        has_header = any(str(p).strip().upper() in {"FECHA","MAE","SMAPE","WAPE"} for p in parts)
        if not has_header and parts:
            # Procesamos esa primera línea como datos
            if len(parts) >= 12:  # SEGMENTO
                rows_seg.append(parts[:11] + [";".join(parts[11:]) if len(parts) > 12 else parts[11]])
            elif len(parts) >= 9: # PORTFOLIO
                rows_port.append(parts[:8] + [";".join(parts[8:]) if len(parts) > 9 else parts[8]])

        # Resto de líneas
        for line in f:
            parts = _smart_split(line)
            if not parts:
                continue
            if len(parts) >= 12:
                # SEGMENTO: tomamos 12 campos (si hay más, lo que sobra lo unimos dentro de 'scope')
                seg = parts[:11]
                scope = ";".join(parts[11:])  # por si el scope tuviera ';' o '\t' embebidos
                rows_seg.append(seg + [scope])
            elif len(parts) >= 9:
                # PORTFOLIO
                port = parts[:8]
                scope = ";".join(parts[8:])
                rows_port.append(port + [scope])
            else:
                # línea demasiado corta -> la ignoramos
                continue

    # Construcción de DataFrames
    df_port = pd.DataFrame(rows_port, columns=cols_port) if rows_port else pd.DataFrame(columns=cols_port)
    df_seg  = pd.DataFrame(rows_seg,  columns=cols_segm) if rows_seg  else pd.DataFrame(columns=cols_segm)

    # Normalización de tipos
    def _to_num(df, cols):
        for c in cols:
            if c in df.columns:
                df[c] = pd.to_numeric(df[c], errors="coerce")

    # PORTFOLIO
    if not df_port.empty:
        df_port["FECHA"] = pd.to_datetime(df_port["FECHA"], errors="coerce")
        _to_num(df_port, ["MAE","SMAPE","WAPE","MASE1","MASE12","n_series_mes","peso_gasto_mes"])
        df_port["scope"] = df_port["scope"].astype(str).str.strip().str.upper()

    # SEGMENTO
    if not df_seg.empty:
        df_seg["FECHA"] = pd.to_datetime(df_seg["FECHA"], errors="coerce")
        _to_num(df_seg, ["MAE","SMAPE","WAPE","MASE1","MASE12","n_series_mes","peso_gasto_mes","flag"])
        df_seg["DIM_NAME"]  = df_seg["DIM_NAME"].astype(str).str.strip()
        df_seg["DIM_VALUE"] = df_seg["DIM_VALUE"].astype(str).str.strip()
        df_seg["scope"]     = df_seg["scope"].astype(str).str.strip().str.upper()

    # Filtrado final por scope (por si viniera mezclado)
    if "scope" in df_port.columns:
        df_port = df_port[df_port["scope"].eq("PORTFOLIO")].copy()
    if "scope" in df_seg.columns:
        df_seg = df_seg[df_seg["scope"].eq("SEGMENTO")].copy()

    return df_port, df_seg

# ---------- EJECUCIÓN ----------
df_portfolio, df_segmento = cargar_metrics_subset_estable(ruta_origen)

# Guardamos
df_portfolio.to_csv(ruta_portfolio, sep=";", index=False)
df_segmento.to_csv(ruta_segmento,  sep=";", index=False)

print("PORTFOLIO -> filas:", df_portfolio.shape[0], "columnas:", df_portfolio.shape[1])
print("SEGMENTO  -> filas:", df_segmento.shape[0],  "columnas:", df_segmento.shape[1])
print("Guardados en:\n -", ruta_portfolio, "\n -", ruta_segmento)

# Vistazo rápido de los segmentos detectados
if not df_segmento.empty:
    print("\nDimensiones de segmento encontradas:")
    print(df_segmento["DIM_NAME"].value_counts())
    print("\nEjemplos:")
    print(df_segmento.head(5))

PORTFOLIO → filas: 12 columnas: 9
SEGMENTO  → filas: 1560 columnas: 12
Guardados en:
 - /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3/METRICAS/metrics_subset_estable_portfolio.csv
 - /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3/METRICAS/metrics_subset_estable_segmento.csv

Dimensiones de segmento encontradas:
DIM_NAME
REGION          1212
TIPO_USO         168
FM_COST_TYPE      96
PAIS              84
Name: count, dtype: int64

Ejemplos:
       DIM_NAME              DIM_VALUE      FECHA         MAE      SMAPE  \
0  FM_COST_TYPE  Eficiencia Energética 2024-01-01  113.729380  29.980846
1  FM_COST_TYPE  Eficiencia Energética 2024-02-01  279.804372  24.098283
2  FM_COST_TYPE  Eficiencia Energética 2024-03-01   73.678876  11.336812
3  FM_COST_TYPE  Eficiencia Energética 2024-04-01   97.060095  26.780269
4  FM_COST_TYPE  Eficiencia Energética 2024-05-01  412.795984  22.038239

       WAPE     MASE1    MASE12  n_series_mes  peso_gasto_mes  flag     scope
0  0.317148  2.552630  0.422272          53.0    19005.824455   1.0  SEGMENTO
1  0.394905  2.682955  0.571354          53.0    37552.424455   1.0  SEGMENTO
2  0.205462  0.059328  0.065404          53.0    19005.824455   1.0  SEGMENTO
3  0.282817  2.617944  0.375707          53.0    18189.088091   1.0  SEGMENTO
4  0.595557  3.062982  0.573264          53.0    36735.688091   1.0  SEGMENTO

# Rutas de los nuevos ficheros
ruta_portfolio = "/content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3/METRICAS/metrics_subset_estable_portfolio.csv"
ruta_segmento  = "/content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3/METRICAS/metrics_subset_estable_segmento.csv"

# Cargar los dos CSV
df_portfolio = pd.read_csv(ruta_portfolio, sep=";")
df_segmento  = pd.read_csv(ruta_segmento, sep=";")

# Información básica
print("PORTFOLIO -> filas:", df_portfolio.shape[0], "columnas:", df_portfolio.shape[1])
print("SEGMENTO --> filas:", df_segmento.shape[0], "columnas:", df_segmento.shape[1])

# Mostrar primeras filas de cada dataframe
print("\nPrimeras filas - PORTFOLIO:")
print(df_portfolio.head())

print("\nPrimeras filas - SEGMENTO:")
print(df_segmento.head())

PORTFOLIO → filas: 12 columnas: 9
SEGMENTO → filas: 1560 columnas: 12

Primeras filas - PORTFOLIO:
        FECHA         MAE      SMAPE      WAPE     MASE1    MASE12  \
0  2024-01-01  415.751344  76.869628  0.286340  2.501419  4.006826
1  2024-02-01  494.764016  77.727396  0.358687  2.149515  2.814806
2  2024-03-01  386.877205  77.150171  0.283530  2.103683  5.993599
3  2024-04-01  416.530706  77.473258  0.291653  2.455745  2.634847
4  2024-05-01  408.472574  82.163432  0.306586  2.372222  4.252092

   n_series_mes  peso_gasto_mes      scope
0        2231.0    3.239299e+06  PORTFOLIO
1        2231.0    3.077384e+06  PORTFOLIO
2        2231.0    3.044203e+06  PORTFOLIO
3        2231.0    3.186256e+06  PORTFOLIO
4        2231.0    2.972418e+06  PORTFOLIO

Primeras filas - SEGMENTO:
       DIM_NAME              DIM_VALUE       FECHA         MAE      SMAPE  \
0  FM_COST_TYPE  Eficiencia Energética  2024-01-01  113.729380  29.980846
1  FM_COST_TYPE  Eficiencia Energética  2024-02-01  279.804372  24.098283
2  FM_COST_TYPE  Eficiencia Energética  2024-03-01   73.678876  11.336812
3  FM_COST_TYPE  Eficiencia Energética  2024-04-01   97.060095  26.780269
4  FM_COST_TYPE  Eficiencia Energética  2024-05-01  412.795984  22.038239

       WAPE     MASE1    MASE12  n_series_mes  peso_gasto_mes  flag     scope
0  0.317148  2.552630  0.422272          53.0    19005.824455   1.0  SEGMENTO
1  0.394905  2.682955  0.571354          53.0    37552.424455   1.0  SEGMENTO
2  0.205462  0.059328  0.065404          53.0    19005.824455   1.0  SEGMENTO
3  0.282817  2.617944  0.375707          53.0    18189.088091   1.0  SEGMENTO
4  0.595557  3.062982  0.573264          53.0    36735.688091   1.0  SEGMENTO

# ============================================
# PASO 13 — Paquete de entrega + Hoja de ruta
# ============================================
# En este script:
#  - Preparamos el entregable "PRIMER_ENTREGABLE"
#  - Leemos artefactos de 2024/2025 (sin romper si faltan)
#  - Calculamos métricas/figuras/informes si hay datos
#  - Generamos paquete para las inferencia en la web (2024 y 2025)
#  - Creamos placeholders para 2025 y otros cuando algo falte para mantener el flujo estable
#  - Integramos fuentes reales: dim_buildings.csv, ipc_monthly_lookup.csv e ipc_anual_diccionario.csv
#  - Alineamos a la política de reconciliación del modelo con el diagnóstico: P10 global por defecto; reglas por país/tipo
#  - Realizamos la partición y normalizamos metrics_subset_estable.csv -> portfolio / segmento

# ==================
# SEMILLA
# ==================

np.random.seed(7)

# ==================
# 0) PARÁMETROS
# ==================
# Ruta de ENTRADA (artefactos generados por pasos previos)
RUTA_ENTRADA = "/content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3"

# Ruta del ENTREGABLE (salida de este paso 13)
ruta_base_3 = "/content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3/PRIMER_ENTREGABLE"

# CSV y umbrales
CSV_SEP       = ";"
ZERO_THR      = 1e-9
N_TOP         = 50
UMBRAL_MASE12 = 1.0
UMBRAL_SMAPE  = 30.0

# Variante global por defecto según diagnóstico (no según disponibilidad)
VAR_FINAL_DEF = "P10"

# Candidatas de fecha
DATE_CANDS = ["FECHA","Date","DS","PERIOD","MONTH","Mes","MES"]

# ==================
# 1) CARPETAS SALIDA
# ==================
RUTA_ENTREGABLE = ruta_base_3
CARP_RES    = os.path.join(RUTA_ENTREGABLE, "RESULTADOS")
CARP_MET    = os.path.join(RUTA_ENTREGABLE, "METRICAS")
CARP_FIG    = os.path.join(RUTA_ENTREGABLE, "FIGURAS")
CARP_INF    = os.path.join(RUTA_ENTREGABLE, "INFORMES")
CARP_LOG    = os.path.join(RUTA_ENTREGABLE, "REGISTROS")
CARP_BUNDLE = os.path.join(RUTA_ENTREGABLE, "PAQUETE_INFERENCIA")

for d in [RUTA_ENTREGABLE, CARP_RES, CARP_MET, CARP_FIG, CARP_INF, CARP_LOG, CARP_BUNDLE]:
    os.makedirs(d, exist_ok=True)

# ============
# 2) LOGGING
# ============
LOG_PATH = os.path.join(CARP_LOG, "paso13.log")

def log13(msg: str):
    ts = datetime.now().isoformat(timespec="seconds")
    line = f"[{ts}] {msg}"
    print(line)
    with open(LOG_PATH, "a", encoding="utf-8") as f:
        f.write(line + "\n")

log13("Iniciamos Paso 13 — Paquete de entrega + Hoja de ruta (alineado con diagnóstico).")

# =====================
# 3) HELPERS GENERALES
# =====================
def _ensure_ms_inplace(df: pd.DataFrame, candidate_cols=DATE_CANDS):
    if df is None or df.empty: return
    for c in candidate_cols:
        if c in df.columns:
            try:
                df[c] = pd.to_datetime(df[c], errors="coerce").dt.to_period("M").dt.to_timestamp()
            except Exception:
                pass

def _read_csv_soft(path: str, candidate_date_cols=None, sep=CSV_SEP) -> pd.DataFrame:
    if not os.path.exists(path):
        rel = path.replace(RUTA_ENTRADA, "").lstrip("/").lstrip("\\")
        log13(f"FALTA {rel}. Seguimos sin romper.")
        return pd.DataFrame()
    try:
        header = pd.read_csv(path, sep=sep, nrows=0)
        header_cols = [str(c).strip().lstrip("\ufeff") for c in header.columns]
        parse_dates = None
        if candidate_date_cols:
            parse_dates = [c for c in candidate_date_cols if c in header_cols] or None
        df = pd.read_csv(path, sep=sep, parse_dates=parse_dates)
        df.columns = [str(c).strip().lstrip("\ufeff") for c in df.columns]
        return df
    except Exception as e:
        log13(f"ERROR leyendo {path}: {e}. Devolvemos DF vacío.")
        return pd.DataFrame()

def _norm_cols(df: pd.DataFrame, variant_tag: str=None):
    if df is None or df.empty:
        return pd.DataFrame(columns=["ID_BUILDING","FM_COST_TYPE","FECHA","yhat"]), None
    out = df.copy()
    colmap = {}
    for c in out.columns:
        cu = str(c).strip().upper()
        if cu in ("ID_BUILDING","IDBUILDING","BUILDING_ID","ID_EDIFICIO"):
            colmap[c] = "ID_BUILDING"
        elif cu in ("FM_COST_TYPE","FM_COST","FMCOST","COST_TYPE"):
            colmap[c] = "FM_COST_TYPE"
        elif cu in ("FECHA","DATE","DS","PERIOD","MONTH","MES"):
            colmap[c] = "FECHA"
    if colmap:
        out = out.rename(columns=colmap)
    pred_col = None
    for candidate in ["yhat_reconc","yhat_combo","yhat","YHAT","pred","PRED","Y","valor","VALOR","forecast"]:
        if candidate in out.columns:
            pred_col = candidate
            break
    if variant_tag and not pred_col:
        tagged = f"yhat_reconc_{variant_tag}"
        if tagged in out.columns:
            pred_col = tagged
    return out, pred_col

def _norm_unpack(df, variant_tag=None):
    ret = _norm_cols(df, variant_tag)
    if isinstance(ret, tuple) and len(ret) == 2:
        return ret[0], ret[1]
    return ret, None

def _ensure_contract(df: pd.DataFrame, pred_col: str, year: int) -> pd.DataFrame:
    cols = ["ID_BUILDING","FM_COST_TYPE","FECHA", pred_col] if pred_col else ["ID_BUILDING","FM_COST_TYPE","FECHA"]
    df2 = df.copy()[[c for c in cols if c in df.columns]] if not df.empty else pd.DataFrame(columns=["ID_BUILDING","FM_COST_TYPE","FECHA","yhat"])
    if not df2.empty and pred_col and pred_col != "yhat":
        df2 = df2.rename(columns={pred_col: "yhat"})
    if "FECHA" in df2.columns:
        _ensure_ms_inplace(df2, ["FECHA"])
        try:
            df2 = df2[df2["FECHA"].dt.year.eq(year)]
        except Exception:
            pass
    if "ID_BUILDING" in df2.columns:
        df2["ID_BUILDING"] = pd.to_numeric(df2["ID_BUILDING"], errors="coerce").astype("Int64")
    if "FM_COST_TYPE" in df2.columns:
        df2["FM_COST_TYPE"] = df2["FM_COST_TYPE"].astype(str).str.strip()
    return df2

def _save_csv(df: pd.DataFrame, path: str):
    os.makedirs(os.path.dirname(path), exist_ok=True)
    df.to_csv(path, sep=CSV_SEP, index=False)
    log13(f"Guardado CSV: {path} ({len(df)} filas).")

def _safe_div(num, den):
    return num/den if (den is not None and den != 0) else np.nan

def _metrics_block(df: pd.DataFrame, col_pred: str, col_real: str, scale_mase=None):
    if df.empty or col_pred not in df.columns or col_real not in df.columns:
        return {"MAE":np.nan,"RMSE":np.nan,"MAPE%":np.nan,"sMAPE%":np.nan,"WAPE%":np.nan,"Bias":np.nan,"MASE":np.nan,"WASE":np.nan}
    y = pd.to_numeric(df[col_real], errors="coerce")
    yhat = pd.to_numeric(df[col_pred], errors="coerce")
    err = yhat - y
    abs_e = err.abs()
    mae  = abs_e.mean()
    rmse = float(np.sqrt((err**2).mean())) if len(err)>0 else np.nan
    mape = float((100.0 * (abs_e / y.replace(0,np.nan).abs())).mean())
    smape= float((100.0 * (abs_e / ((y.abs() + yhat.abs())/2.0).replace(0,np.nan))).mean())
    wape = float(100.0 * abs_e.sum() / y.abs().sum()) if y.abs().sum()!=0 else np.nan
    bias = err.mean()
    if scale_mase is not None and "MASE_scale" in scale_mase.columns:
        df_sc = df.merge(scale_mase, on=["ID_BUILDING","FM_COST_TYPE"], how="left")
        scale = pd.to_numeric(df_sc["MASE_scale"], errors="coerce").replace(0,np.nan)
        mase = float((abs_e / scale).mean())
        wase = float(abs_e.sum() / scale.sum()) if pd.notna(scale.sum()) and scale.sum()!=0 else np.nan
    else:
        mase, wase = np.nan, np.nan
    return {"MAE":mae,"RMSE":rmse,"MAPE%":mape,"sMAPE%":smape,"WAPE%":wape,"Bias":bias,"MASE":mase,"WASE":wase}

def _build_mase_scale(train_df: pd.DataFrame):
    if train_df.empty or "FECHA" not in train_df.columns or "cost_float_mod" not in train_df.columns:
        return pd.DataFrame(columns=["ID_BUILDING","FM_COST_TYPE","MASE_scale"])
    t = train_df.copy()
    _ensure_ms_inplace(t, ["FECHA"])
    t = t.sort_values(["ID_BUILDING","FM_COST_TYPE","FECHA"])
    t["y_lag12"] = t.groupby(["ID_BUILDING","FM_COST_TYPE"])["cost_float_mod"].shift(12)
    t["abs_diff12"] = (t["cost_float_mod"] - t["y_lag12"]).abs()
    sc = (t.groupby(["ID_BUILDING","FM_COST_TYPE"], as_index=False)["abs_diff12"]
            .mean().rename(columns={"abs_diff12":"MASE_scale"}))
    sc["MASE_scale"] = pd.to_numeric(sc["MASE_scale"], errors="coerce").fillna(0.0)
    sc.loc[sc["MASE_scale"]<=0, "MASE_scale"] = 1e-8
    return sc

def _schema_md_report(paths_dict: dict, frames_dict: dict, out_md: str):
    lines = ["# Esquema de entradas — Paso 13\n"]
    for k, p in paths_dict.items():
        lines.append(f"## {k}\n")
        lines.append(f"- Ruta: `{p}`")
        df = frames_dict.get(k, pd.DataFrame())
        if df is None or df.empty:
            lines.append("- Estado: **vacío o ausente**\n")
        else:
            cols = ", ".join(df.columns.tolist())
            n = len(df)
            lines.append(f"- Filas: {n}")
            lines.append(f"- Columnas: {cols}\n")
    with open(out_md, "w", encoding="utf-8") as f:
        f.write("\n".join(lines))
    log13(f"Guardado informe de esquema: {out_md}")

# Helper robusto para leer metrics_subset_estable.csv mixto (portfolio/segmento) ---
def _smart_split_line(line: str):
    line = line.rstrip("\n").rstrip("\r")
    if not line: return []
    parts_sc = line.split(";")
    parts_tb = line.split("\t")
    return parts_tb if len(parts_tb) > len(parts_sc) else parts_sc

def _load_metrics_subset_estable_mixto(path: str):
    """
    Lee el CSV del Paso 9 aunque mezcle PORTFOLIO (9 campos) y SEGMENTO (12).
    Devuelve dos DataFrames: df_portfolio, df_segmento (ya normalizados).
    """
    cols_port = ["FECHA","MAE","SMAPE","WAPE","MASE1","MASE12","n_series_mes","peso_gasto_mes","scope"]
    cols_segm = ["DIM_NAME","DIM_VALUE","FECHA","MAE","SMAPE","WAPE","MASE1","MASE12","n_series_mes","peso_gasto_mes","flag","scope"]
    rows_port, rows_seg = [], []

    if not os.path.exists(path):
        return pd.DataFrame(columns=cols_port), pd.DataFrame(columns=cols_segm)

    with open(path, "r", encoding="utf-8") as f:
        first = f.readline()
        parts = _smart_split_line(first)
        has_header = any(str(p).strip().upper() in {"FECHA","MAE","SMAPE","WAPE"} for p in parts)
        if not has_header and parts:
            if len(parts) >= 12:
                rows_seg.append(parts[:11] + [";".join(parts[11:])])
            elif len(parts) >= 9:
                rows_port.append(parts[:8] + [";".join(parts[8:])])

        for line in f:
            parts = _smart_split_line(line)
            if not parts:
                continue
            if len(parts) >= 12:
                seg = parts[:11]; scope = ";".join(parts[11:])
                rows_seg.append(seg + [scope])
            elif len(parts) >= 9:
                port = parts[:8]; scope = ";".join(parts[8:])
                rows_port.append(port + [scope])

    df_port = pd.DataFrame(rows_port, columns=cols_port) if rows_port else pd.DataFrame(columns=cols_port)
    df_seg  = pd.DataFrame(rows_seg,  columns=cols_segm) if rows_seg  else pd.DataFrame(columns=cols_segm)

    if not df_port.empty:
        df_port["FECHA"] = pd.to_datetime(df_port["FECHA"], errors="coerce")
        for c in ["MAE","SMAPE","WAPE","MASE1","MASE12","n_series_mes","peso_gasto_mes"]:
            df_port[c] = pd.to_numeric(df_port[c], errors="coerce")
        df_port["scope"] = df_port["scope"].astype(str).str.strip().str.upper()
        df_port = df_port[df_port["scope"].eq("PORTFOLIO")].copy()

    if not df_seg.empty:
        df_seg["FECHA"] = pd.to_datetime(df_seg["FECHA"], errors="coerce")
        for c in ["MAE","SMAPE","WAPE","MASE1","MASE12","n_series_mes","peso_gasto_mes","flag"]:
            df_seg[c] = pd.to_numeric(df_seg[c], errors="coerce")
        df_seg["DIM_NAME"]  = df_seg["DIM_NAME"].astype(str).str.strip()
        df_seg["DIM_VALUE"] = df_seg["DIM_VALUE"].astype(str).str.strip()
        df_seg["scope"]     = df_seg["scope"].astype(str).str.strip().str.upper()
        df_seg = df_seg[df_seg["scope"].eq("SEGMENTO")].copy()

    return df_port, df_seg

# IPC mensual: intentamos el mensual real y, si no hay, derivamos desde anual
def _load_ipc_monthly(P_IPC_MENSUAL_REAL, P_IPC_ANUAL_DICT, P_IPC25, date_cols=DATE_CANDS):
    mensual = pd.DataFrame()
    if os.path.exists(P_IPC_MENSUAL_REAL):
        mensual = _read_csv_soft(P_IPC_MENSUAL_REAL, candidate_date_cols=date_cols)
        if not mensual.empty:
            if "FECHA" not in mensual.columns:
                for c in ["Date","DS","PERIOD","MONTH","Mes","MES"]:
                    if c in mensual.columns:
                        mensual = mensual.rename(columns={c:"FECHA"})
                        break
            _ensure_ms_inplace(mensual, ["FECHA"])
    if os.path.exists(P_IPC25):
        ipc25 = _read_csv_soft(P_IPC25, candidate_date_cols=date_cols)
        if not ipc25.empty:
            if "FECHA" not in ipc25.columns:
                for c in ["Date","DS","PERIOD","MONTH","Mes","MES"]:
                    if c in ipc25.columns:
                        ipc25 = ipc25.rename(columns={c:"FECHA"})
                        break
            _ensure_ms_inplace(ipc25, ["FECHA"])
            mensual = pd.concat([mensual, ipc25], ignore_index=True) if not mensual.empty else ipc25
    if mensual.empty and os.path.exists(P_IPC_ANUAL_DICT):
        anual = _read_csv_soft(P_IPC_ANUAL_DICT)
        up = {c.upper(): c for c in anual.columns}
        pais_col = up.get("PAIS") or up.get("COUNTRY")
        year_col = up.get("ANO") or up.get("AÑO") or up.get("YEAR")
        val_col  = up.get("IPC") or up.get("IPC_YOY_PCT") or up.get("VALOR") or up.get("INDEX")
        if pais_col and year_col and val_col:
            rows = []
            for _, r in anual.iterrows():
                pais = r[pais_col]
                y    = int(r[year_col])
                val  = float(r[val_col])
                for m in range(1,13):
                    rows.append({"PAIS": pais, "FECHA": pd.Timestamp(y, m, 1), "IPC_YOY_PCT": val})
            mensual = pd.DataFrame(rows)
        else:
            log13("AVISO: ipc_anual_diccionario.csv no tiene columnas reconocibles (PAIS/YEAR/IPC). IPC mensual quedará vacío.")
    return mensual

# ==================================================
# 4) RUTAS DE ENTRADA
# ==================================================
# 2024
P_BASE  = os.path.join(RUTA_ENTRADA, "RESULTADOS", "preds_por_serie_2024.csv")
P_P10   = os.path.join(RUTA_ENTRADA, "RESULTADOS", "preds_reconciliadas_2024.csv")
P_P11   = os.path.join(RUTA_ENTRADA, "RESULTADOS", "preds_reconciliadas_2024_prod.csv")
P_P12S  = os.path.join(RUTA_ENTRADA, "RESULTADOS", "preds_reconciliadas_step12.csv")
P_P12A  = os.path.join(RUTA_ENTRADA, "RESULTADOS", "preds_reconciliadas_step12_activeonly.csv")
P_COMP  = os.path.join(RUTA_ENTRADA, "REPORTING", "compare_recon_steps_bottom.csv")
P_REAL  = os.path.join(RUTA_ENTRADA, "RESULTADOS", "test_full_2024.csv")
P_TRN   = os.path.join(RUTA_ENTRADA, "RESULTADOS", "train_full_2021_2023.csv")

# Dimensiones e IPC (fuentes reales)
P_DIM       = os.path.join(RUTA_ENTRADA, "METRICAS", "dim_buildings.csv")
P_DIM_ES    = os.path.join(RUTA_ENTRADA, "METRICAS", "dim_edificios.csv")
P_P8        = os.path.join(RUTA_ENTRADA, "METRICAS", "paso8_metrics_2024_por_pareja.csv")
P_IPC_REAL  = os.path.join(RUTA_ENTRADA, "METRICAS", "ipc_monthly_lookup.csv")
P_IPC_ES    = os.path.join(RUTA_ENTRADA, "METRICAS", "ipc_mensual.csv")
P_IPC25     = os.path.join(RUTA_ENTRADA, "METRICAS", "ipc_mensual_2025.csv")
P_IPC_ANO   = os.path.join(RUTA_ENTRADA, "METRICAS", "ipc_anual_diccionario.csv")

# Subset estable "oficial" del Paso 9 (si existe)
P_SUBSET_P9   = os.path.join(RUTA_ENTRADA, "METRICAS", "metrics_subset_estable.csv")
P_SUBSET_PORT = os.path.join(RUTA_ENTRADA, "METRICAS", "metrics_subset_estable_portfolio.csv")
P_SUBSET_SEGM = os.path.join(RUTA_ENTRADA, "METRICAS", "metrics_subset_estable_segmento.csv")

# 2025 (si ya existen)
P_BASE_2025 = os.path.join(RUTA_ENTRADA, "RESULTADOS", "preds_base_2025.csv")
P_P11_2025  = os.path.join(RUTA_ENTRADA, "RESULTADOS", "preds_reconciliadas_2025_P11.csv")
P_P12_2025  = os.path.join(RUTA_ENTRADA, "RESULTADOS", "preds_reconciliadas_2025_P12STD.csv")

# ==========================
# 5) LECTURA SUAVE ENTRADAS
# ==========================
df_base  = _read_csv_soft(P_BASE,  candidate_date_cols=DATE_CANDS)
df_p10   = _read_csv_soft(P_P10,   candidate_date_cols=DATE_CANDS)
df_p11   = _read_csv_soft(P_P11,   candidate_date_cols=DATE_CANDS)
df_p12s  = _read_csv_soft(P_P12S,  candidate_date_cols=DATE_CANDS)
df_p12a  = _read_csv_soft(P_P12A,  candidate_date_cols=DATE_CANDS)
df_comp  = _read_csv_soft(P_COMP,  candidate_date_cols=DATE_CANDS)
df_real  = _read_csv_soft(P_REAL,  candidate_date_cols=DATE_CANDS)
df_train = _read_csv_soft(P_TRN,   candidate_date_cols=DATE_CANDS)

# Dimensiones: preferimos españolizado si existiera; si no, usamos el real
df_dim = _read_csv_soft(P_DIM_ES)
if df_dim.empty:
    df_dim = _read_csv_soft(P_DIM)

# IPC mensual consolidado
df_ipc = _load_ipc_monthly(P_IPC_REAL, P_IPC_ANO, P_IPC25, DATE_CANDS)
df_ipc25 = df_ipc[df_ipc["FECHA"].dt.year.eq(2025)] if not df_ipc.empty and "FECHA" in df_ipc.columns else pd.DataFrame()

# Informe de esquemas
_schema_md_report(
    {
        "BASE_2024":P_BASE, "P10_2024":P_P10, "P11_2024":P_P11, "P12_STD_2024":P_P12S, "P12_ACTIVE_2024":P_P12A,
        "COMPARATIVA_BOTTOM":P_COMP, "REAL_2024":P_REAL, "TRAIN_2021_2023":P_TRN,
        "DIM_EDIFICIOS": (P_DIM_ES if os.path.exists(P_DIM_ES) else P_DIM),
        "PASO8_METRICAS":P_P8, "IPC_MENSUAL(merged)": (P_IPC_ES if os.path.exists(P_IPC_ES) else P_IPC_REAL),
        "IPC_ANUAL_DICC":P_IPC_ANO, "IPC_MENSUAL_2025":P_IPC25,
        "BASE_2025":P_BASE_2025, "P11_2025":P_P11_2025, "P12_2025":P_P12_2025
    },
    {
        "BASE_2024":df_base, "P10_2024":df_p10, "P11_2024":df_p11, "P12_STD_2024":df_p12s, "P12_ACTIVE_2024":df_p12a,
        "COMPARATIVA_BOTTOM":df_comp, "REAL_2024":df_real, "TRAIN_2021_2023":df_train,
        "DIM_EDIFICIOS":df_dim, "PASO8_METRICAS":_read_csv_soft(P_P8),
        "IPC_MENSUAL(merged)":df_ipc, "IPC_ANUAL_DICC":_read_csv_soft(P_IPC_ANO), "IPC_MENSUAL_2025":df_ipc25,
        "BASE_2025":_read_csv_soft(P_BASE_2025, DATE_CANDS), "P11_2025":_read_csv_soft(P_P11_2025, DATE_CANDS),
        "P12_2025":_read_csv_soft(P_P12_2025, DATE_CANDS)
    },
    os.path.join(CARP_INF, "esquema_entradas.md")
)

# Normalizamos + extraemos columnas de predicción (2024)
df_base, base_pred_col = _norm_cols(df_base, None)
df_p10,  p10_pred_col  = _norm_cols(df_p10,  "p10")
df_p11,  p11_pred_col  = _norm_cols(df_p11,  "p11")
df_p12s, p12s_pred_col = _norm_cols(df_p12s, "p12std")
df_p12a, p12a_pred_col = _norm_cols(df_p12a, "p12act")

# Contrato 2024 por variante
base_2024 = _ensure_contract(df_base, base_pred_col, 2024)
p10_2024  = _ensure_contract(df_p10,  p10_pred_col  or "yhat_reconc", 2024)
p11_2024  = _ensure_contract(df_p11,  p11_pred_col  or "yhat_reconc", 2024)
p12s_2024 = _ensure_contract(df_p12s, p12s_pred_col or "yhat_reconc", 2024)
p12a_2024 = _ensure_contract(df_p12a, p12a_pred_col or "yhat_reconc", 2024)

# Escala MASE si tenemos TRAIN
mase_scale = pd.DataFrame()
if not df_train.empty and "cost_float_mod" in df_train.columns:
    _ensure_ms_inplace(df_train, ["FECHA"])
    mase_scale = _build_mase_scale(df_train)

# ===============================
# 6) MÉTRICAS CLAVE (si hay real)
# ===============================
real_2024 = pd.DataFrame()
if not df_real.empty and "cost_float_mod" in df_real.columns:
    real_2024 = df_real.copy()
    _ensure_ms_inplace(real_2024, ["FECHA"])
    real_2024 = real_2024.rename(columns={"cost_float_mod":"y_real"})

def _merge_real(pred_df):
    if pred_df.empty or real_2024.empty:
        return pd.DataFrame()
    cols = ["ID_BUILDING","FM_COST_TYPE","FECHA","yhat"]
    m = pred_df[cols].merge(real_2024[["ID_BUILDING","FM_COST_TYPE","FECHA","y_real"]],
                            on=["ID_BUILDING","FM_COST_TYPE","FECHA"], how="inner")
    return m

def _select_variant_availability_for_metrics(base_2024, p10_2024, p11_2024, p12s_2024):
    availability = {
        "P10":     not p10_2024.empty,
        "P11":     not p11_2024.empty,
        "BASE":    not base_2024.empty,
        "P12_STD": not p12s_2024.empty
    }
    order = [VAR_FINAL_DEF, "P11", "BASE", "P12_STD"]
    chosen = next((v for v in order if availability.get(v, False)), None)
    if chosen is None:
        log13("AVISO: no hay ninguna variante disponible (ni BASE). El paso continuará con métricas vacías.")
    else:
        log13(f"Variante final seleccionada para métricas (según diagnóstico): {chosen}")
    return chosen, availability

chosen_var, availability = _select_variant_availability_for_metrics(base_2024, p10_2024, p11_2024, p12s_2024)

# Métricas globales por variante disponible
metric_rows = []
var_map = {"BASE": base_2024, "P10": p10_2024, "P11": p11_2024, "P12_STD": p12s_2024, "P12_ACTIVE": p12a_2024}
for vname, vdf in var_map.items():
    if vdf.empty:
        continue
    m = _merge_real(vdf) if not real_2024.empty else pd.DataFrame()
    if not m.empty:
        met = _metrics_block(m, "yhat", "y_real", scale_mase=mase_scale)
    else:
        met = {"MAE":np.nan,"RMSE":np.nan,"MAPE%":np.nan,"sMAPE%":np.nan,"WAPE%":np.nan,"Bias":np.nan,"MASE":np.nan,"WASE":np.nan}
    if not vdf.empty and not real_2024.empty:
        merged = vdf.merge(real_2024[["ID_BUILDING","FM_COST_TYPE","FECHA"]], on=["ID_BUILDING","FM_COST_TYPE","FECHA"], how="left")
        cobertura = 100.0 * merged["FECHA_y"].notna().mean() if "FECHA_y" in merged.columns else np.nan
    else:
        cobertura = np.nan
    row = {"Variante": vname, **met, "Cobertura%": cobertura}
    metric_rows.append(row)

df_metrics_global = pd.DataFrame(metric_rows)
_save_csv(df_metrics_global, os.path.join(CARP_MET, "metricas_globales_2024.csv"))

# Métricas por FM_COST_TYPE × PAIS
df_metrics_seg = pd.DataFrame()
if not real_2024.empty and not df_dim.empty:
    dim = df_dim.copy()
    colmap_dim = {}
    for c in dim.columns:
        cu = str(c).upper()
        if cu in ("ID_BUILDING","IDBUILDING","BUILDING_ID","ID_EDIFICIO"):
            colmap_dim[c] = "ID_BUILDING"
        elif cu in ("PAIS","COUNTRY","COUNTRY_DEF"):
            colmap_dim[c] = "PAIS"
        elif cu in ("REGION","REGIÓN"):
            colmap_dim[c] = "REGION"
        elif cu in ("STATUS","ESTADO"):
            colmap_dim[c] = "STATUS"
    if colmap_dim:
        dim = dim.rename(columns=colmap_dim)

    for vname, vdf in var_map.items():
        if vdf.empty:
            continue
        merged = vdf.merge(dim[["ID_BUILDING","PAIS"]] if "PAIS" in dim.columns else vdf[["ID_BUILDING"]].assign(PAIS="SIN_PAIS"),
                           on="ID_BUILDING", how="left")
        merged = merged.merge(real_2024[["ID_BUILDING","FM_COST_TYPE","FECHA","y_real"]],
                              on=["ID_BUILDING","FM_COST_TYPE","FECHA"], how="inner")
        if merged.empty:
            continue
        rows = []
        grp_cols = ["FM_COST_TYPE","PAIS"] if "PAIS" in merged.columns else ["FM_COST_TYPE"]
        for keys, g in merged.groupby(grp_cols):
            fmc = keys[0] if isinstance(keys, tuple) else keys
            pais = (keys[1] if isinstance(keys, tuple) and len(keys)>1 else None)
            met = _metrics_block(g, "yhat","y_real", scale_mase=mase_scale)
            base = {"Variante": vname, "FM_COST_TYPE":fmc}
            if pais is not None:
                base["PAIS"]=pais
            rows.append({**base, **met})
        if rows:
            segi = pd.DataFrame(rows)
            df_metrics_seg = pd.concat([df_metrics_seg, segi], ignore_index=True)
else:
    log13("AVISO: sin reales 2024 no generamos métricas segmentadas.")

_save_csv(df_metrics_seg, os.path.join(CARP_MET, "metricas_por_fmcost_pais_2024.csv"))

# =========================================================
# SUBSET ESTABLE 2024 — Carga robusta y separación portfolio/segmento
# =========================================================
# Si ya existen los ficheros separados, los leemos; si no, partimos del mixto y los generamos
df_subset_port, df_subset_segm = pd.DataFrame(), pd.DataFrame()
if os.path.exists(P_SUBSET_PORT) or os.path.exists(P_SUBSET_SEGM):
    if os.path.exists(P_SUBSET_PORT):
        df_subset_port = _read_csv_soft(P_SUBSET_PORT, candidate_date_cols=["FECHA"])
    if os.path.exists(P_SUBSET_SEGM):
        df_subset_segm = _read_csv_soft(P_SUBSET_SEGM, candidate_date_cols=["FECHA"])
else:
    # Leemos el mixto y separamos
    df_subset_port, df_subset_segm = _load_metrics_subset_estable_mixto(P_SUBSET_P9)
    # Guardamos en ENTRADA (saneados para futuros pasos)
    if not df_subset_port.empty:
        df_subset_port.to_csv(P_SUBSET_PORT, sep=CSV_SEP, index=False)
        log13(f"Normalizado y guardado: {P_SUBSET_PORT} ({len(df_subset_port)} filas).")
    if not df_subset_segm.empty:
        df_subset_segm.to_csv(P_SUBSET_SEGM, sep=CSV_SEP, index=False)
        log13(f"Normalizado y guardado: {P_SUBSET_SEGM} ({len(df_subset_segm)} filas).")

# Copiamos/guardamos también en el ENTREGABLE (carpeta METRICAS del paquete)
OUT_SUBSET_PORT = os.path.join(CARP_MET, "metrics_subset_estable_portfolio.csv")
OUT_SUBSET_SEGM = os.path.join(CARP_MET, "metrics_subset_estable_segmento.csv")
_save_csv(df_subset_port, OUT_SUBSET_PORT)
_save_csv(df_subset_segm, OUT_SUBSET_SEGM)

# Para compatibilidad hacia atrás, mantenemos un "subconjunto_estable_2024.csv" mínimo (usamos portfolio)
_save_csv(df_subset_port.copy(), os.path.join(CARP_MET, "subconjunto_estable_2024.csv"))

# Avisos útiles
if df_subset_segm.empty:
    log13("AVISO: metrics_subset_estable_segmento está vacío. Revisa Paso 9 (export de segmentos).")

# Top-N (intentamos usar el subset más informativo disponible)
df_pairs_for_top = None
# Preferimos datos con columna WAPE(%) o WAPE a nivel detallado; si no los hay, recomputamos desde predicciones
for cand in [df_subset_segm, df_subset_port]:
    if not cand.empty and (("WAPE%" in cand.columns) or ("WAPE" in cand.columns)):
        df_pairs_for_top = cand.copy()
        break

if df_pairs_for_top is None and (chosen_var is not None and not real_2024.empty):
    chosen_df = var_map.get(chosen_var, pd.DataFrame())
    merged = _merge_real(chosen_df)
    rows = []
    for (bid, fmc), g in merged.groupby(["ID_BUILDING","FM_COST_TYPE"]):
        met = _metrics_block(g, "yhat","y_real", scale_mase=mase_scale if not mase_scale.empty else None)
        rows.append({"ID_BUILDING":bid,"FM_COST_TYPE":fmc, **met})
    df_pairs_for_top = pd.DataFrame(rows)

if df_pairs_for_top is not None and not df_pairs_for_top.empty:
    # Elegimos la columna de WAPE disponible
    wape_col = "WAPE%" if "WAPE%" in df_pairs_for_top.columns else ("WAPE" if "WAPE" in df_pairs_for_top.columns else None)
    if wape_col:
        # Si WAPE está en [0,1], lo pasamos a %
        w = pd.to_numeric(df_pairs_for_top[wape_col], errors="coerce")
        if wape_col == "WAPE" and w.notna().max() <= 1.0:
            df_pairs_for_top["WAPE%"] = 100.0 * w
            wape_col = "WAPE%"
        top_mej = df_pairs_for_top.sort_values(wape_col, ascending=True).head(N_TOP)
        top_peor= df_pairs_for_top.sort_values(wape_col, ascending=False).head(N_TOP)
        _save_csv(top_mej, os.path.join(CARP_MET, "top_mejores_wape.csv"))
        _save_csv(top_peor, os.path.join(CARP_MET, "top_peores_wape.csv"))
    else:
        _save_csv(pd.DataFrame(), os.path.join(CARP_MET, "top_mejores_wape.csv"))
        _save_csv(pd.DataFrame(), os.path.join(CARP_MET, "top_peores_wape.csv"))
else:
    _save_csv(pd.DataFrame(), os.path.join(CARP_MET, "top_mejores_wape.csv"))
    _save_csv(pd.DataFrame(), os.path.join(CARP_MET, "top_peores_wape.csv"))

# Outliers por delta P12
top_out = pd.DataFrame()
if not df_comp.empty:
    d = df_comp.copy()
    for cand in ["delta_p12std","delta_p12","delta_p11","delta_p10"]:
        if cand in d.columns:
            d["abs_delta"] = pd.to_numeric(d[cand], errors="coerce").abs()
            top_out = d.sort_values("abs_delta", ascending=False).head(N_TOP)
            break
_save_csv(top_out, os.path.join(CARP_MET, "top_outliers_delta_p12.csv"))

# =================
# 7) FIGURAS
# =================
if chosen_var is not None and not real_2024.empty:
    chosen_df = var_map.get(chosen_var, pd.DataFrame())
    m = _merge_real(chosen_df)
    if not m.empty:
        by_m = (m.groupby("FECHA", as_index=False)
                  .apply(lambda g: pd.Series({"WAPE%": (100.0 * (g["yhat"]-g["y_real"]).abs().sum() /
                                                       g["y_real"].abs().sum()) if g["y_real"].abs().sum()!=0 else np.nan})))
        plt.figure(figsize=(10,5))
        plt.plot(by_m["FECHA"], by_m["WAPE%"])
        plt.title(f"WAPE mensual portfolio — {chosen_var} (2024)")
        plt.xlabel("Mes"); plt.ylabel("WAPE %")
        plt.tight_layout()
        plt.savefig(os.path.join(CARP_FIG, "wape_portfolio_mensual.png"), dpi=150)
        plt.close()
    else:
        log13("AVISO: sin merge real no generamos figura wape_portfolio_mensual.")
else:
    log13("AVISO: sin reales 2024 no generamos figura wape_portfolio_mensual.")

if chosen_var is not None and not real_2024.empty:
    chosen_df = var_map.get(chosen_var, pd.DataFrame())
    merged = chosen_df.merge(real_2024, on=["ID_BUILDING","FM_COST_TYPE","FECHA"], how="inner")
    if not merged.empty:
        w_rows = []
        for fmc, g in merged.groupby("FM_COST_TYPE"):
            den = g["y_real"].abs().sum()
            wape = 100.0 * (g["yhat"]-g["y_real"]).abs().sum() / den if den!=0 else np.nan
            w_rows.append({"FM_COST_TYPE":fmc,"WAPE%":wape})
        wdf = pd.DataFrame(w_rows).sort_values("WAPE%", ascending=False)
        plt.figure(figsize=(10,5))
        plt.bar(wdf["FM_COST_TYPE"], wdf["WAPE%"])
        plt.xticks(rotation=45, ha="right")
        plt.title(f"WAPE por FM_COST_TYPE — {chosen_var} (2024)")
        plt.ylabel("WAPE %")
        plt.tight_layout()
        plt.savefig(os.path.join(CARP_FIG, "wape_por_fmcost.png"), dpi=150)
        plt.close()
    else:
        log13("AVISO: sin merge real no generamos figura wape_por_fmcost.")
else:
    log13("AVISO: sin reales 2024 no generamos figura wape_por_fmcost.")

if not df_comp.empty:
    d = df_comp.copy()
    delta_cols = [c for c in ["delta_p10","delta_p11","delta_p12","delta_p12std","delta_p12act"] if c in d.columns]
    if delta_cols:
        plt.figure(figsize=(9,6))
        plt.boxplot([pd.to_numeric(d[c], errors="coerce").dropna().values for c in delta_cols], labels=delta_cols, showfliers=False)
        plt.axhline(0, color="grey", linestyle=":")
        plt.title("Distribución de deltas bottom (reconc - base)")
        plt.ylabel("Delta")
        plt.tight_layout()
        plt.savefig(os.path.join(CARP_FIG, "boxplot_deltas_reconciliacion.png"), dpi=150)
        plt.close()
    else:
        log13("AVISO: no fue posible generar boxplot de deltas (faltan columnas delta_*).")
else:
    log13("AVISO: comparativa bottom ausente; no generamos boxplot de deltas.")

# =================
# 8) INFORMES
# =================
resumen_lines = [
"# Resumen ejecutivo — Paso 13",
"",
"Este documento recoge las cifras clave de desempeño 2024, el subconjunto estable, oportunidades rápidas (quick-wins) y las referencias a los principales artefactos del entregable.",
"",
"## 1) Cinco cifras clave (2024)",
]
if not df_metrics_global.empty:
    row = None
    if chosen_var is not None:
        rows = df_metrics_global[df_metrics_global["Variante"].eq(chosen_var)]
        if not rows.empty: row = rows.iloc[0]
    if row is None:
        row = df_metrics_global.iloc[0]
    resumen_lines += [
        f"- **Variante**: {row['Variante']}",
        f"- **WAPE**: {row['WAPE%']:.2f}%",
        f"- **sMAPE**: {row['sMAPE%']:.2f}%",
        f"- **MASE**: {row['MASE']:.4f}",
        f"- **Cobertura**: {row['Cobertura%']:.2f}%"
    ]
else:
    resumen_lines += ["- No disponemos de métricas pobladas (faltan reales o variantes)."]

resumen_lines += [
"",
"## 2) Subconjunto estable (agregado de métricas)",
f"- Portfolio: `METRICAS/metrics_subset_estable_portfolio.csv` (filas: {len(df_subset_port)})",
f"- Segmento:  `METRICAS/metrics_subset_estable_segmento.csv` (filas: {len(df_subset_segm)})",
"- (Compat.)  `METRICAS/subconjunto_estable_2024.csv` (alias de portfolio)",
"",
"## 3) Quick-wins propuestos",
"- Intermitentes/TSB: priorizar pares con alta esporadicidad donde el TSB mejora el error.",
"- Combos/Ensemble: consolidar rutas con mejor MASE/WAPE por segmento.",
"- Outliers (mod_3): revisar top outliers en `METRICAS/top_outliers_delta_p12.csv` y considerar winsorización.",
"",
"## 4) Enlaces a outputs del entregable",
"- `METRICAS/metricas_globales_2024.csv`",
"- `METRICAS/metricas_por_fmcost_pais_2024.csv`",
"- `FIGURAS/wape_portfolio_mensual.png`",
"- `FIGURAS/wape_por_fmcost.png`",
"- `FIGURAS/boxplot_deltas_reconciliacion.png`",
"",
"## 5) Checklist de verificación",
"- Predicciones disponibles para todas las series o fallback documentado.",
"- Ficheros METRICAS completos generados.",
"- Nombres y rutas coherentes con el estándar del proyecto.",
]
with open(os.path.join(CARP_INF, "resumen_ejecutivo.md"), "w", encoding="utf-8") as f:
    f.write("\n".join(resumen_lines))

check_lines = [
"# Checklist de verificación — Paso 13",
"",
"- [ ] Predicciones 2024 presentes (BASE/P10/P11/P12_STD) o placeholders creados",
"- [ ] Predicciones 2025 presentes (BASE/P11/P12_STD) o placeholders creados",
"- [ ] Métricas globales 2024 generadas",
"- [ ] Métricas por FM_COST_TYPE × PAIS generadas",
"- [ ] Subset estable (portfolio y segmento) generados",
"- [ ] Top-N (mejores/peores WAPE) generados",
"- [ ] Outliers delta P12 disponibles",
"- [ ] Figuras generadas (si hay datos)",
"- [ ] Bundle de inferencia listo (manifest.json, predict.py, matrices si aplica)",
]
with open(os.path.join(CARP_INF, "checklist_verificacion.md"), "w", encoding="utf-8") as f:
    f.write("\n".join(check_lines))

# =========================================
# 9) EXPORT DE INFERENCIA PARA LA WEB
# =========================================
# 9.1 Resultados 2024 -> ENTREGABLE/RESULTADOS
base_2024_export = base_2024.copy() if not base_2024.empty else pd.DataFrame(columns=["ID_BUILDING","FM_COST_TYPE","FECHA","yhat"])
p10_2024_export  = p10_2024.copy()   if not p10_2024.empty   else pd.DataFrame(columns=["ID_BUILDING","FM_COST_TYPE","FECHA","yhat"])
p11_2024_export  = p11_2024.copy()   if not p11_2024.empty   else pd.DataFrame(columns=["ID_BUILDING","FM_COST_TYPE","FECHA","yhat"])
p12s_2024_export = p12s_2024.copy()  if not p12s_2024.empty  else pd.DataFrame(columns=["ID_BUILDING","FM_COST_TYPE","FECHA","yhat"])

_save_csv(base_2024_export, os.path.join(CARP_RES, "preds_base_2024.csv"))
_save_csv(p10_2024_export,  os.path.join(CARP_RES, "preds_reconciliadas_2024_P10.csv"))
_save_csv(p11_2024_export,  os.path.join(CARP_RES, "preds_reconciliadas_2024_P11.csv"))
_save_csv(p12s_2024_export, os.path.join(CARP_RES, "preds_reconciliadas_2024_P12STD.csv"))

# 9.2 Resultados 2025 (leemos si existen; si no, placeholders)
base_2025_raw = _read_csv_soft(P_BASE_2025, DATE_CANDS); base_2025_df, base25_pred = _norm_unpack(base_2025_raw, None)
base_2025 = _ensure_contract(base_2025_df, base25_pred, 2025)

p11_2025_raw = _read_csv_soft(P_P11_2025, DATE_CANDS); p11_2025_df, p11_25_pred = _norm_unpack(p11_2025_raw, "p11")
p11_2025 = _ensure_contract(p11_2025_df, p11_25_pred or "yhat_reconc", 2025)

p12_2025_raw = _read_csv_soft(P_P12_2025, DATE_CANDS); p12_2025_df, p12_25_pred = _norm_unpack(p12_2025_raw, "p12std")
p12_2025 = _ensure_contract(p12_2025_df, p12_25_pred or "yhat_reconc", 2025)

_save_csv(base_2025 if not base_2025.empty else pd.DataFrame(columns=["ID_BUILDING","FM_COST_TYPE","FECHA","yhat"]),
          os.path.join(CARP_RES, "preds_base_2025.csv"))
_save_csv(p11_2025 if not p11_2025.empty else pd.DataFrame(columns=["ID_BUILDING","FM_COST_TYPE","FECHA","yhat"]),
          os.path.join(CARP_RES, "preds_reconciliadas_2025_P11.csv"))
_save_csv(p12_2025 if not p12_2025.empty else pd.DataFrame(columns=["ID_BUILDING","FM_COST_TYPE","FECHA","yhat"]),
          os.path.join(CARP_RES, "preds_reconciliadas_2025_P12STD.csv"))

# 9.3 Ruteo por par con reglas (país/tipo) y degradación por disponibilidad
def _availability_flags(dfs_by_variant, year):
    rows = []
    for variant, dfv in dfs_by_variant.items():
        if dfv.empty:
            continue
        g = dfv.groupby(["ID_BUILDING","FM_COST_TYPE"], as_index=False).size()
        g["year"] = year
        g["variant"] = variant
        g["available"] = 1
        rows.append(g[["ID_BUILDING","FM_COST_TYPE","year","variant","available"]])
    return pd.concat(rows, ignore_index=True) if rows else pd.DataFrame(columns=["ID_BUILDING","FM_COST_TYPE","year","variant","available"])

avail_2024 = _availability_flags({"BASE":base_2024_export, "P10":p10_2024_export, "P11":p11_2024_export, "P12_STD":p12s_2024_export}, 2024)
avail_2025 = _availability_flags({"BASE":base_2025, "P11":p11_2025, "P12_STD":p12_2025}, 2025)
avail_all  = pd.concat([avail_2024, avail_2025], ignore_index=True)

# Reglas por país/tipo
PREF_PAIS = {
    "COLOMBIA":"P10",
    "ESPAÑA":"P10",
    "MEXICO":"P11", "MÉXICO":"P11",
    "PANAMÁ":"P11","PANAMA":"P11",
    "PERÚ":"P11","PERU":"P11",
    "COSTA RICA":None,
    "REP. DOMINICANA":"P10", "REPUBLICA DOMINICANA":"P10"
}
PREF_FMCOST = { "SERVICIOS CTTO./EXTRA":"P11" }
DEFAULT_VARIANT = "P10"

def _prefer_variant_for_pair(pais, fmc):
    v_pais = PREF_PAIS.get(str(pais).upper().strip()) if pd.notna(pais) else None
    v_fmc  = PREF_FMCOST.get(str(fmc).upper().strip()) if pd.notna(fmc) else None
    return v_fmc or v_pais or DEFAULT_VARIANT

# Enriquecemos con PAIS desde dim
dim_min = df_dim.copy()
if not dim_min.empty:
    colmap_dim = {}
    for c in dim_min.columns:
        cu = str(c).upper()
        if cu in ("ID_BUILDING","IDBUILDING","BUILDING_ID","ID_EDIFICIO"):
            colmap_dim[c] = "ID_BUILDING"
        elif cu in ("PAIS","COUNTRY","COUNTRY_DEF"):
            colmap_dim[c] = "PAIS"
    if colmap_dim:
        dim_min = dim_min.rename(columns=colmap_dim)
    dim_min = dim_min[["ID_BUILDING","PAIS"]].drop_duplicates() if "ID_BUILDING" in dim_min.columns else pd.DataFrame(columns=["ID_BUILDING","PAIS"])
else:
    dim_min = pd.DataFrame(columns=["ID_BUILDING","PAIS"])

pairs_year = avail_all[["ID_BUILDING","FM_COST_TYPE","year"]].drop_duplicates()
pairs_year = pairs_year.merge(dim_min, on="ID_BUILDING", how="left")

decisions = []
for _, r in pairs_year.iterrows():
    pref = _prefer_variant_for_pair(r.get("PAIS"), r.get("FM_COST_TYPE"))
    if r["year"] == 2024:
        degrade_order = [pref, "P10","P11","BASE","P12_STD"] if pref in ["P10","P11","BASE","P12_STD"] else ["P10","P11","BASE","P12_STD"]
    else:
        degrade_order = [pref, "P11","BASE","P12_STD"] if pref in ["P11","BASE","P12_STD"] else ["P11","BASE","P12_STD"]
    degrade = [v for v in degrade_order if v is not None]
    seen = set(); degrade = [x for x in degrade if not (x in seen or seen.add(x))]
    chosen = None
    for v in degrade:
        ok = not avail_all[(avail_all["ID_BUILDING"].eq(r["ID_BUILDING"])) &
                           (avail_all["FM_COST_TYPE"].eq(r["FM_COST_TYPE"])) &
                           (avail_all["year"].eq(r["year"])) &
                           (avail_all["variant"].eq(v))].empty
        if ok:
            chosen = v
            break
    decisions.append({
        "ID_BUILDING": r["ID_BUILDING"],
        "FM_COST_TYPE": r["FM_COST_TYPE"],
        "PAIS": r.get("PAIS"),
        "year": r["year"],
        "variant_preferida": pref if pref is not None else "SIN_REGLA",
        "variant_final": chosen or "NO_DISPONIBLE"
    })

ruteo = pd.DataFrame(decisions)
_save_csv(ruteo, os.path.join(CARP_MET, "ruteo_por_par.csv"))

# 9.4 Manifest del bundle (incluye los nuevos subset)
manifest = {
    "generated_at": datetime.now().isoformat(timespec="seconds"),
    "base_path": RUTA_ENTREGABLE,
    "results": {
        "2024": {
            "BASE":    "RESULTADOS/preds_base_2024.csv",
            "P10":     "RESULTADOS/preds_reconciliadas_2024_P10.csv",
            "P11":     "RESULTADOS/preds_reconciliadas_2024_P11.csv",
            "P12_STD": "RESULTADOS/preds_reconciliadas_2024_P12STD.csv"
        },
        "2025": {
            "BASE":    "RESULTADOS/preds_base_2025.csv",
            "P11":     "RESULTADOS/preds_reconciliadas_2025_P11.csv",
            "P12_STD": "RESULTADOS/preds_reconciliadas_2025_P12STD.csv"
        }
    },
    "metrics": {
        "global_2024": "METRICAS/metricas_globales_2024.csv",
        "seg_2024":    "METRICAS/metricas_por_fmcost_pais_2024.csv",
        "subset_portfolio_2024": "METRICAS/metrics_subset_estable_portfolio.csv",
        "subset_segmento_2024":  "METRICAS/metrics_subset_estable_segmento.csv",
        "subset_legacy_2024":    "METRICAS/subconjunto_estable_2024.csv",
        "ruteo":       "METRICAS/ruteo_por_par.csv"
    },
    "figures": {
        "wape_mensual": "FIGURAS/wape_portfolio_mensual.png",
        "wape_fmcost":  "FIGURAS/wape_por_fmcost.png",
        "box_deltas":   "FIGURAS/boxplot_deltas_reconciliacion.png"
    }
}
with open(os.path.join(CARP_BUNDLE, "manifest.json"), "w", encoding="utf-8") as f:
    json.dump(manifest, f, ensure_ascii=False, indent=2)

# 9.5 Copias útiles al entregable (dimensiones e IPC) con nombres españolizados
if not df_dim.empty:
    dim_out = df_dim.copy()
    colmap_dim = {}
    for c in dim_out.columns:
        cu = str(c).upper()
        if cu in ("ID_BUILDING","IDBUILDING","BUILDING_ID","ID_EDIFICIO"):
            colmap_dim[c] = "ID_BUILDING"
        elif cu in ("PAIS","COUNTRY","COUNTRY_DEF"):
            colmap_dim[c] = "PAIS"
        elif cu in ("REGION","REGIÓN"):
            colmap_dim[c] = "REGION"
        elif cu in ("STATUS","ESTADO"):
            colmap_dim[c] = "STATUS"
    if colmap_dim:
        dim_out = dim_out.rename(columns=colmap_dim)
    _save_csv(dim_out, os.path.join(CARP_MET, "dim_edificios.csv"))
else:
    _save_csv(pd.DataFrame(), os.path.join(CARP_MET, "dim_edificios.csv"))

if not df_ipc.empty:
    ipcout = df_ipc.copy()
    colmap_ipc = {}
    up = {c.upper(): c for c in ipcout.columns}
    if "PAIS" not in up:
        for k in ("COUNTRY","COUNTRY_DEF"):
            if k in up: colmap_ipc[up[k]] = "PAIS"; break
    if "FECHA" not in up:
        for k in ("DATE","DS","PERIOD","MES","MONTH"):
            if k in up: colmap_ipc[up[k]] = "FECHA"; break
    if "IPC_YOY_PCT" not in up:
        for k in ("IPC","IPC_YOY","YOY","VAR_INTERANUAL","IPC_ANUAL_%"):
            if k in up: colmap_ipc[up[k]] = "IPC_YOY_PCT"; break
    if colmap_ipc:
        ipcout = ipcout.rename(columns=colmap_ipc)
    if "FECHA" in ipcout.columns:
        ipcout["FECHA"] = pd.to_datetime(ipcout["FECHA"], errors="coerce").dt.to_period("M").dt.to_timestamp()
    _save_csv(ipcout, os.path.join(CARP_MET, "ipc_mensual.csv"))
else:
    _save_csv(pd.DataFrame(), os.path.join(CARP_MET, "ipc_mensual.csv"))

if not df_ipc25.empty:
    ip25 = df_ipc25.copy()
    if "FECHA" in ip25.columns:
        ip25["FECHA"] = pd.to_datetime(ip25["FECHA"], errors="coerce")
    _save_csv(ip25, os.path.join(CARP_MET, "ipc_mensual_2025.csv"))
else:
    log13("AVISO: no guardamos ipc_mensual_2025.csv porque no lo pudimos construir ni leer.")

# 9.6 predict.py (para servir desde la web)
predict_py = f'''# predict.py — lector de predicciones 2024/2025 para la web
# Devolvemos (ID_BUILDING, FM_COST_TYPE, FECHA, yhat)
import os, pandas as pd
CSV_SEP = "{CSV_SEP}"

def _auto_base():
    here = os.path.abspath(os.path.dirname(__file__))
    return os.path.abspath(os.path.join(here, ".."))

def _read(path):
    if not os.path.exists(path):
        return pd.DataFrame(columns=["ID_BUILDING","FM_COST_TYPE","FECHA","yhat"])
    hdr = pd.read_csv(path, sep=CSV_SEP, nrows=0)
    parse_dates = ["FECHA"] if "FECHA" in hdr.columns else None
    df = pd.read_csv(path, sep=CSV_SEP, parse_dates=parse_dates)
    df.columns = [str(c).strip() for c in df.columns]
    need = ["ID_BUILDING","FM_COST_TYPE","FECHA","yhat"]
    for n in need:
        if n not in df.columns:
            df[n] = pd.Series(dtype="float64")
    return df[need]

def predict(ids, year=2025, variant="P10", ruta_base="__AUTO__"):
    """
    ids: lista de ID_BUILDING
    year: 2024 o 2025
    variant: 'P10' | 'P11' | 'P12_STD' | 'BASE'
    ruta_base: si '__AUTO__', resolvemos base a partir de este paquete
    """
    if ruta_base == "__AUTO__":
        ruta_base = _auto_base()
    if year == 2024:
        files = {{
            "BASE":    os.path.join(ruta_base, "RESULTADOS", "preds_base_2024.csv"),
            "P10":     os.path.join(ruta_base, "RESULTADOS", "preds_reconciliadas_2024_P10.csv"),
            "P11":     os.path.join(ruta_base, "RESULTADOS", "preds_reconciliadas_2024_P11.csv"),
            "P12_STD": os.path.join(ruta_base, "RESULTADOS", "preds_reconciliadas_2024_P12STD.csv")
        }}
    else:
        files = {{
            "BASE":    os.path.join(ruta_base, "RESULTADOS", "preds_base_2025.csv"),
            "P11":     os.path.join(ruta_base, "RESULTADOS", "preds_reconciliadas_2025_P11.csv"),
            "P12_STD": os.path.join(ruta_base, "RESULTADOS", "preds_reconciliadas_2025_P12STD.csv")
        }}
    # Si piden P10 en 2025 y no existe, degradamos a P11, luego BASE, luego P12_STD
    if year == 2025 and variant.upper() == "P10":
        variant = "P11"
    path = files.get(variant.upper(), files.get("P11", files.get("BASE")))
    df = _read(path)
    if ids is not None and len(ids)>0 and "ID_BUILDING" in df.columns:
        sids = pd.Series(ids).astype("Int64") if hasattr(pd.Series(ids), "astype") else pd.Series(ids)
        df = df[df["ID_BUILDING"].astype("Int64").isin(sids)]
    return df
'''
with open(os.path.join(CARP_BUNDLE, "predict.py"), "w", encoding="utf-8") as f:
    f.write(predict_py)

# (Opcional) Matrices A / W si existieran en entrada (las traemos al bundle)
A_CAND = os.path.join(RUTA_ENTRADA, "METRICAS", "A_matriz.npz")
W_CAND = os.path.join(RUTA_ENTRADA, "METRICAS", "W_diag.npy")
if os.path.exists(A_CAND):
    import shutil
    shutil.copy(A_CAND, os.path.join(CARP_BUNDLE, "A_matriz.npz"))
if os.path.exists(W_CAND):
    import shutil
    shutil.copy(W_CAND, os.path.join(CARP_BUNDLE, "W_diag.npy"))

log13("Paso 13 finalizado.")

[2025-09-28T07:27:55] Iniciamos Paso 13 — Paquete de entrega + Hoja de ruta (alineado con diagnóstico).
[2025-09-28T07:28:04] FALTA METRICAS/dim_edificios.csv. Seguimos sin romper.
[2025-09-28T07:28:07] Guardado informe de esquema: /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3/PRIMER_ENTREGABLE/INFORMES/esquema_entradas.md
[2025-09-28T07:28:08] Variante final seleccionada para métricas (según diagnóstico): P10
[2025-09-28T07:28:08] Guardado CSV: /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3/PRIMER_ENTREGABLE/METRICAS/metricas_globales_2024.csv (5 filas).
[2025-09-28T07:28:11] Guardado CSV: /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3/PRIMER_ENTREGABLE/METRICAS/metricas_por_fmcost_pais_2024.csv (215 filas).
[2025-09-28T07:28:12] Guardado CSV: /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3/PRIMER_ENTREGABLE/METRICAS/metrics_subset_estable_portfolio.csv (12 filas).
[2025-09-28T07:28:12] Guardado CSV: /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3/PRIMER_ENTREGABLE/METRICAS/metrics_subset_estable_segmento.csv (1560 filas).
[2025-09-28T07:28:12] Guardado CSV: /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3/PRIMER_ENTREGABLE/METRICAS/subconjunto_estable_2024.csv (12 filas).
[2025-09-28T07:28:13] Guardado CSV: /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3/PRIMER_ENTREGABLE/METRICAS/top_mejores_wape.csv (50 filas).
[2025-09-28T07:28:13] Guardado CSV: /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3/PRIMER_ENTREGABLE/METRICAS/top_peores_wape.csv (50 filas).
[2025-09-28T07:28:13] Guardado CSV: /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3/PRIMER_ENTREGABLE/METRICAS/top_outliers_delta_p12.csv (50 filas).
[2025-09-28T07:28:17] Guardado CSV: /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3/PRIMER_ENTREGABLE/RESULTADOS/preds_base_2024.csv (29148 filas).
[2025-09-28T07:28:18] Guardado CSV: /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3/PRIMER_ENTREGABLE/RESULTADOS/preds_reconciliadas_2024_P10.csv (29148 filas).
[2025-09-28T07:28:19] Guardado CSV: /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3/PRIMER_ENTREGABLE/RESULTADOS/preds_reconciliadas_2024_P11.csv (29148 filas).
[2025-09-28T07:28:20] Guardado CSV: /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3/PRIMER_ENTREGABLE/RESULTADOS/preds_reconciliadas_2024_P12STD.csv (29148 filas).
[2025-09-28T07:28:20] Guardado CSV: /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3/PRIMER_ENTREGABLE/RESULTADOS/preds_base_2025.csv (0 filas).
[2025-09-28T07:28:21] Guardado CSV: /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3/PRIMER_ENTREGABLE/RESULTADOS/preds_reconciliadas_2025_P11.csv (0 filas).
[2025-09-28T07:28:21] Guardado CSV: /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3/PRIMER_ENTREGABLE/RESULTADOS/preds_reconciliadas_2025_P12STD.csv (0 filas).
[2025-09-28T07:28:40] Guardado CSV: /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3/PRIMER_ENTREGABLE/METRICAS/ruteo_por_par.csv (2429 filas).
[2025-09-28T07:28:41] Guardado CSV: /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3/PRIMER_ENTREGABLE/METRICAS/dim_edificios.csv (1650 filas).
[2025-09-28T07:28:41] Guardado CSV: /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3/PRIMER_ENTREGABLE/METRICAS/ipc_mensual.csv (12 filas).
[2025-09-28T07:28:41] AVISO: no guardamos ipc_mensual_2025.csv porque no lo pudimos construir ni leer.
[2025-09-28T07:28:41] Paso 13 finalizado.

# =========================================
# 10) VALIDACIÓN FINAL DE ARTEFACTOS CLAVE
# =========================================
log13("=== VALIDACIÓN FINAL — Resumen de artefactos clave ===")

def _read_out_soft(path):
    if not os.path.exists(path):
        log13(f"[VALID] NO ENCONTRADO -> {path}")
        return pd.DataFrame()
    try:
        hdr = pd.read_csv(path, sep=CSV_SEP, nrows=0)
        parse_dates = ["FECHA"] if "FECHA" in hdr.columns else None
        df = pd.read_csv(path, sep=CSV_SEP, parse_dates=parse_dates)
        df.columns = [str(c).strip() for c in df.columns]
        return df
    except Exception as e:
        log13(f"[VALID][ERROR] Leyendo {path}: {e}")
        return pd.DataFrame()

def _summary_df(tag, df: pd.DataFrame, extra_checks=None):
    n, cols = (len(df), df.columns.tolist()) if not df.empty else (0, [])
    msg = f"[VALID] {tag}: filas={n}, columnas={len(cols)}"
    if "FECHA" in cols and not df.empty and pd.api.types.is_datetime64_any_dtype(df["FECHA"]):
        try:
            rmin, rmax = df["FECHA"].min(), df["FECHA"].max()
            msg += f", FECHA[{rmin.date()} .. {rmax.date()}]"
        except Exception:
            pass
    log13(msg)
    if n and extra_checks:
        extra_checks(df)

def _check_wape(df):
    for c in ["WAPE%","WAPE"]:
        if c in df.columns:
            w = pd.to_numeric(df[c], errors="coerce")
            if w.notna().any():
                mx = w.max()
                if c == "WAPE" and mx <= 1.0:
                    log13(f"[VALID][OK] {c} parece fracción [0..1].")
                elif c == "WAPE" and mx > 1.0:
                    log13(f"[VALID][WARN] {c} parece porcentaje (max={mx:.2f}); considera renombrar a WAPE% o dividir por 100.")
                if c == "WAPE%" and mx <= 1.0:
                    log13(f"[VALID][WARN] {c} parece fracción (max={mx:.3f}); ¿faltó *100?")
            break

def _check_scope_port(df):
    if "scope" in df.columns:
        vals = df["scope"].astype(str).str.upper().unique().tolist()
        ok = "PORTFOLIO" in [v.upper() for v in vals]
        log13(f"[VALID][{ 'OK' if ok else 'WARN' }] scope contiene PORTFOLIO. Valores: {vals}")

def _check_scope_segm(df):
    need = {"DIM_NAME","DIM_VALUE","scope"}
    miss = [c for c in need if c not in df.columns]
    if miss:
        log13(f"[VALID][WARN] columnas esperado segmento faltantes: {miss}")
    vals = df["scope"].astype(str).str.upper().unique().tolist() if "scope" in df.columns else []
    ok = "SEGMENTO" in [v.upper() for v in vals]
    log13(f"[VALID][{ 'OK' if ok else 'WARN' }] scope contiene SEGMENTO. Valores: {vals}")

# Rutas a validar (salidas del entregable)
artefactos = {
    "preds_base_2024"        : os.path.join(CARP_RES, "preds_base_2024.csv"),
    "preds_p10_2024"         : os.path.join(CARP_RES, "preds_reconciliadas_2024_P10.csv"),
    "preds_p11_2024"         : os.path.join(CARP_RES, "preds_reconciliadas_2024_P11.csv"),
    "preds_p12std_2024"      : os.path.join(CARP_RES, "preds_reconciliadas_2024_P12STD.csv"),
    "metricas_globales_2024" : os.path.join(CARP_MET, "metricas_globales_2024.csv"),
    "metricas_seg_2024"      : os.path.join(CARP_MET, "metricas_por_fmcost_pais_2024.csv"),
    "subset_portfolio_2024"  : os.path.join(CARP_MET, "metrics_subset_estable_portfolio.csv"),
    "subset_segmento_2024"   : os.path.join(CARP_MET, "metrics_subset_estable_segmento.csv"),
    "subset_legacy_2024"     : os.path.join(CARP_MET, "subconjunto_estable_2024.csv"),
    "top_mejores_wape"       : os.path.join(CARP_MET, "top_mejores_wape.csv"),
    "top_peores_wape"        : os.path.join(CARP_MET, "top_peores_wape.csv"),
    "outliers_delta_p12"     : os.path.join(CARP_MET, "top_outliers_delta_p12.csv"),
    "ipc_mensual"            : os.path.join(CARP_MET, "ipc_mensual.csv"),
    "ipc_mensual_2025"       : os.path.join(CARP_MET, "ipc_mensual_2025.csv"),
    "dim_edificios"          : os.path.join(CARP_MET, "dim_edificios.csv"),
}

for tag, path in artefactos.items():
    dfv = _read_out_soft(path)
    extras = None
    if tag == "subset_portfolio_2024":
        extras = lambda d: (_check_scope_port(d), _check_wape(d))
    elif tag == "subset_segmento_2024":
        extras = lambda d: (_check_scope_segm(d), _check_wape(d))
    elif tag in {"top_mejores_wape","top_peores_wape"}:
        extras = _check_wape
    _summary_df(tag, dfv, extras)

log13("=== VALIDACIÓN FINAL — Completada ===")

[2025-09-28T07:51:13] === VALIDACIÓN FINAL — Resumen de artefactos clave ===
[2025-09-28T07:51:14] [VALID] preds_base_2024: filas=29148, columnas=4, FECHA[2024-01-01 .. 2024-12-01]
[2025-09-28T07:51:14] [VALID] preds_p10_2024: filas=29148, columnas=4, FECHA[2024-01-01 .. 2024-12-01]
[2025-09-28T07:51:15] [VALID] preds_p11_2024: filas=29148, columnas=4, FECHA[2024-01-01 .. 2024-12-01]
[2025-09-28T07:51:15] [VALID] preds_p12std_2024: filas=29148, columnas=4, FECHA[2024-01-01 .. 2024-12-01]
[2025-09-28T07:51:15] [VALID] metricas_globales_2024: filas=5, columnas=10
[2025-09-28T07:51:16] [VALID] metricas_seg_2024: filas=215, columnas=11
[2025-09-28T07:51:16] [VALID] subset_portfolio_2024: filas=12, columnas=9, FECHA[2024-01-01 .. 2024-12-01]
[2025-09-28T07:51:16] [VALID][OK] scope contiene PORTFOLIO. Valores: ['PORTFOLIO']
[2025-09-28T07:51:16] [VALID][OK] WAPE parece fracción [0..1].
[2025-09-28T07:51:16] [VALID] subset_segmento_2024: filas=1560, columnas=12, FECHA[2024-01-01 .. 2024-12-01]
[2025-09-28T07:51:16] [VALID][OK] scope contiene SEGMENTO. Valores: ['SEGMENTO']
[2025-09-28T07:51:16] [VALID][WARN] WAPE parece porcentaje (max=169462000000.00); considera renombrar a WAPE% o dividir por 100.
[2025-09-28T07:51:16] [VALID] subset_legacy_2024: filas=12, columnas=9, FECHA[2024-01-01 .. 2024-12-01]
[2025-09-28T07:51:16] [VALID] top_mejores_wape: filas=50, columnas=13, FECHA[2024-01-01 .. 2024-12-01]
[2025-09-28T07:51:16] [VALID] top_peores_wape: filas=50, columnas=13, FECHA[2024-01-01 .. 2024-12-01]
[2025-09-28T07:51:17] [VALID] outliers_delta_p12: filas=50, columnas=13, FECHA[2024-01-01 .. 2024-12-01]
[2025-09-28T07:51:17] [VALID] ipc_mensual: filas=12, columnas=9, FECHA[2023-11-01 .. 2023-11-01]
[2025-09-28T07:51:17] [VALID] NO ENCONTRADO -> /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/ESTRATEGIA_3/PRIMER_ENTREGABLE/METRICAS/ipc_mensual_2025.csv
[2025-09-28T07:51:17] [VALID] ipc_mensual_2025: filas=0, columnas=0
[2025-09-28T07:51:17] [VALID] dim_edificios: filas=1650, columnas=5
[2025-09-28T07:51:17] === VALIDACIÓN FINAL — Completada ===

# Ruta del notebook .ipynb que vamos a convertir

RUTA_NOTEBOOK = "/content/drive/MyDrive/PROYECTO_NEITH/DOCUMENTOS_PROYECTOS/Neith.ipynb"

# Carpeta de salida y nombre del HTML
OUTPUT_DIR = "/content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO"
OUTPUT_HTML = "Neith_notebook_v5_estrategia_3_final.html"

# Aseguramos carpeta de salida (ya en cabecera)
# import os
os.makedirs(OUTPUT_DIR, exist_ok=True)

# Instalamos nbconvert (ya en cabecera)
# !pip install -q "nbconvert>=7.0.0"

# Convertimos notebook a HTML y lo guardamos con el nombre indicado en la carpeta de salida
!jupyter nbconvert --to html "$RUTA_NOTEBOOK" --output "$OUTPUT_HTML" --output-dir "$OUTPUT_DIR"

# Verificamos
salida = os.path.join(OUTPUT_DIR, OUTPUT_HTML)
print("Archivo HTML generado en:", salida, "\nExiste:", os.path.exists(salida))

[NbConvertApp] Converting notebook /content/drive/MyDrive/PROYECTO_NEITH/DOCUMENTOS_PROYECTOS/Neith.ipynb to html
[NbConvertApp] WARNING | Alternative text is missing on 60 image(s).
[NbConvertApp] Writing 9575851 bytes to /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/Neith_notebook_v5_estrategia_3_final.html
Archivo HTML generado en: /content/drive/MyDrive/PROYECTO_NEITH/OUTPUT_PROYECTO/Neith_notebook_v5_estrategia_3_final.html
Existe: True

	ID_ORDER	COST	COST_TYPE	COUNTRY	YEAR_MONTH	FM_COST_TYPE
59707	F001203888	#DIV/0	Gasto	Marruecos	202304	Suministros
59710	F001203891	#DIV/0	Gasto	Marruecos	202305	Suministros
59713	F001203894	#DIV/0	Gasto	Marruecos	202306	Suministros
59716	F001203897	#DIV/0	Gasto	Marruecos	202307	Suministros
59719	F001203900	#DIV/0	Gasto	Marruecos	202308	Suministros
59722	F001203903	#DIV/0	Gasto	Marruecos	202309	Suministros
59725	F001203906	#DIV/0	Gasto	Marruecos	202310	Suministros
59728	F001203909	#DIV/0	Gasto	Marruecos	202311	Suministros
59731	F001203912	#DIV/0	Gasto	Marruecos	202312	Suministros
59734	F001203915	#DIV/0	Gasto	Marruecos	202401	Suministros
59737	F001203918	#DIV/0	Gasto	Marruecos	202402	Suministros
60568	F001204797	#DIV/0	Gasto	Marruecos	202403	Suministros
60573	F001204802	#DIV/0	Gasto	Marruecos	202406	Suministros
60574	F001204803	#DIV/0	Gasto	Marruecos	202405	Suministros
60578	F001204807	#DIV/0	Gasto	Marruecos	202406	Suministros
62239	F001206667	#DIV/0	Gasto	Marruecos	202407	Suministros
64435	F001209124	#DIV/0	Gasto	Marruecos	202409	Suministros
67530	F001213812	#DIV/0	Gasto	Marruecos	202410	Suministros
69199	F001215603	#DIV/0	Gasto	Marruecos	202411	Suministros
81775	F001228815	#DIV/0	Gasto	Marruecos	202507	Suministros
81776	F001228816	#DIV/0	Gasto	Marruecos	202507	Suministros
81777	F001228817	#DIV/0	Gasto	Marruecos	202507	Suministros
82613	F001229828	#DIV/0	Gasto	Marruecos	202508	Suministros
82614	F001229829	#DIV/0	Gasto	Marruecos	202508	Suministros
82615	F001229830	#DIV/0	Gasto	Marruecos	202508	Suministros
82616	F001229831	#DIV/0	Gasto	Marruecos	202508	Suministros
85707	F001235569	#DIV/0	Gasto	Marruecos	202506	Suministros
85708	F001235570	#DIV/0	Gasto	Marruecos	202507	Suministros
85709	F001235571	#DIV/0	Gasto	Marruecos	202508	Suministros
85710	F001235572	#DIV/0	Gasto	Marruecos	202509	Suministros
85711	F001235573	#DIV/0	Gasto	Marruecos	202510	Suministros
85712	F001235574	#DIV/0	Gasto	Marruecos	202511	Suministros
85713	F001235575	#DIV/0	Gasto	Marruecos	202512	Suministros
85714	F001235578	#DIV/0	Gasto	Marruecos	202506	Suministros
85715	F001235579	#DIV/0	Gasto	Marruecos	202507	Suministros
85716	F001235580	#DIV/0	Gasto	Marruecos	202508	Suministros
85717	F001235581	#DIV/0	Gasto	Marruecos	202509	Suministros
85718	F001235582	#DIV/0	Gasto	Marruecos	202510	Suministros
85719	F001235583	#DIV/0	Gasto	Marruecos	202511	Suministros
85720	F001235584	#DIV/0	Gasto	Marruecos	202512	Suministros
86197	F001236253	#DIV/0	Gasto	Marruecos	202506	Suministros
86198	F001236254	#DIV/0	Gasto	Marruecos	202507	Suministros
86199	F001236255	#DIV/0	Gasto	Marruecos	202508	Suministros
86200	F001236256	#DIV/0	Gasto	Marruecos	202509	Suministros
86201	F001236257	#DIV/0	Gasto	Marruecos	202510	Suministros
86202	F001236258	#DIV/0	Gasto	Marruecos	202511	Suministros
86203	F001236259	#DIV/0	Gasto	Marruecos	202512	Suministros

	count	mean	std	min	25%	50%	75%	max
ID_BUILDING	324422.0	5.061269e+05	4.999032e+05	2.00	1092.0	1000004.000	1000666.0	1.001604e+06
YEAR_MONTH	324422.0	2.023162e+05	1.359207e+02	202101.00	202206.0	202308.000	202410.0	2.025120e+05
ID_CUSTOMER	324422.0	1.237584e+06	1.476999e+06	3.00	1399.0	2362.000	3001496.0	3.007033e+06
ID_SUPPLIER	324422.0	2.454111e+06	1.161188e+06	4.00	3001882.0	3003587.000	3004886.0	3.007643e+06
ID_BUSINESS_UNIT	324422.0	2.202848e+05	4.143526e+05	0.00	43.0	43.000	51.0	1.000109e+06
MONTH	324422.0	6.326192e+00	3.319426e+00	1.00	4.0	6.000	9.0	1.200000e+01
YEAR	324422.0	2.023099e+03	1.364301e+00	2021.00	2022.0	2023.000	2024.0	2.025000e+03
QUARTER	324422.0	2.444723e+00	1.083516e+00	1.00	2.0	2.000	3.0	4.000000e+00
cost_float	324375.0	8.618261e+02	9.997411e+03	-40880.07	0.0	55.731	279.0	2.623544e+06

	count	mean	std	min	25%	50%	75%	max
ID_BUILDING	306775.0	501178.645929	499922.745257	2.00	1093.0	1.000004e+06	1000608.000	1001588.000
MONTH	306775.0	6.308598	3.315537	1.00	3.0	6.000000e+00	9.000	12.000
YEAR	306775.0	2023.123093	1.362546	2021.00	2022.0	2.023000e+03	2024.000	2025.000
cost_float_mod	306775.0	583.043000	2271.779234	-40880.07	0.0	4.772727e+01	233.306	118503.992

	ID_ORDER	ID_SUPPLIER	COST	cost_float	COST_TYPE	COUNTRY	YEAR	MONTH
196330	701690066	3006129	100	100.000000	Gasto	Perú	2023	5
197806	701695109	3006129	576.190476	576.190476	Gasto	Perú	2023	5
200638	701701552	3006129	114.285714	114.285714	Gasto	Perú	2023	6
211469	701724829	3006129	90.47619	90.476190	Gasto	Perú	2023	9
300690	701906619	3006129	361.904762	361.904762	Gasto	Perú	2025	4
308437	701919528	3007232	923.583333	923.583333	Inversión	Perú	2025	6
315020	701930289	3007232	169.047619	169.047619	Gasto	Perú	2025	7
322373	701943483	3007232	0	0.000000	Gasto	Perú	2025	8

	ID_ORDER	COST_TYPE	COUNTRY	YEAR	MONTH	FM_RESPONSIBLE	FM_COST_TYPE	cost_float
175309	1643232	Gasto	España	2023	1	Vacío	Vacío	0.00
175315	1643238	Gasto	España	2023	1	Vacío	Vacío	300.00
181588	1656972	Gasto	España	2023	3	Vacío	Vacío	693.60
181591	1656975	Gasto	España	2023	3	Vacío	Vacío	0.00
187654	1670590	Gasto	España	2023	4	Vacío	Vacío	0.00
187655	1670591	Gasto	España	2023	3	Vacío	Vacío	1504.80
189482	1674466	Gasto	España	2023	3	Vacío	Vacío	61.01
189483	1674467	Gasto	España	2023	4	Vacío	Vacío	0.00
189484	1674468	Gasto	España	2023	6	Vacío	Vacío	80.40
197827	701695169	Gasto	España	2023	6	Vacío	Vacío	0.00
197828	701695170	Gasto	España	2023	6	Vacío	Vacío	0.00
233465	701772199	Gasto	España	2024	2	Vacío	Vacío	0.00
234052	701773448	Gasto	Marruecos	2025	2	Vacío	Vacío	0.00
234053	701773450	Inversión	Marruecos	2025	2	Vacío	Vacío	0.00
234055	701773454	Gasto	Marruecos	2025	2	Vacío	Vacío	0.00
234713	701774821	Gasto	España	2024	2	Vacío	Vacío	0.00
319642	701938246	Gasto	España	2025	7	Vacío	Vacío	255.96

YEAR	2021	2022	2023	2024	2025
FM_RESPONSIBLE
Control y Reporting	6750.00	16980.00	0.000000	0.00	0.00
Dirección FM	1649.15	8651.36	8506.600000	7357.74	4512.41
Global	0.00	0.00	9618.181818	200.00	18497.40
Oficina Técnica	0.00	23644.66	6918.000000	24798.11	13442.30
Operaciones	0.00	0.00	1158.300000	0.00	0.00
Real Estate	0.00	2477.00	62.850000	0.00	142.15

	FM_RESPONSIBLE	YEAR	ID_BUILDING	cost_float
0	Control y Reporting	2021	18	6750.00
1	Control y Reporting	2022	18	16980.00
2	Control y Reporting	2022	1000667	0.00
3	Dirección FM	2021	18	1649.15
4	Dirección FM	2022	18	8651.36
...	...	...	...	...
76	Real Estate	2025	619	0.00
78	Real Estate	2025	1001131	0.00
79	Real Estate	2025	1001155	0.00
80	Real Estate	2025	1001512	0.00
81	Real Estate	2025	1001526	0.00

	variable	tipo_variables	valores_unicos	tipo_variable	justificacion
0	ID_ORDER	object	324421	identificador	Clave técnica única por fila; no aporta patron...
1	COST_TYPE	object	2	categorica	Etiqueta OPEX/CAPEX (Gasto/Inversión); categór...
2	COUNTRY	object	9	categorica	Diferencias regulatorias/mercado; categórica g...
3	ID_BUILDING	int64	722	identificador	Identificador del espacio; el código en sí no ...
4	YEAR_MONTH	int64	60	temporal	Periodo YYYYMM; útil para agregaciones tempora...
5	FM_RESPONSIBLE	object	12	categorica	Vertical responsable; categórica potencialment...
6	FM_COST_TYPE	object	14	categorica	Familia del gasto (contrato/bajo demanda, etc....
7	COST	object	88837	texto_original	Valor original de coste en texto; se mantiene ...
8	ID_CUSTOMER	int64	181	identificador	Clave de sociedad; mejor usar rasgos derivados...
9	ID_SUPPLIER	int64	2594	identificador	Clave de proveedor; preferible usar atributos ...
10	ID_BUSINESS_UNIT	int64	14	identificador	Clave de división; el número no tiene signific...
11	SUPPLIER_TYPE	object	3	categorica_ref	Categoría original con codificación inconsiste...
12	MONTH	int64	12	temporal	Mes numérico; temporal para agregaciones.
13	YEAR	int64	5	temporal	Año; temporal para cortes y tendencias.
14	QUARTER	int64	4	temporal	Trimestre; temporal para agregaciones trimestr...
15	cost_float	float64	88836	numerica_continua	Coste en euros listo para análisis; variable n...
16	SUPPLIER_TYPE_MOD	object	2	categorica	Tipo de proveedor corregido; usar esta en luga...

	ID_ORDER	COST_TYPE	COUNTRY	YEAR	MONTH	FM_RESPONSIBLE	FM_COST_TYPE	ID_BUILDING	cost_float
26297	1163481	Gasto	Panamá	2022	7	Eficiencia Energética	Eficiencia Energética	1001191	55.081818
26756	1163956	Gasto	Panamá	2022	8	Eficiencia Energética	Eficiencia Energética	1001191	55.081818
28145	1165508	Gasto	Panamá	2022	9	Eficiencia Energética	Eficiencia Energética	1001191	55.081818
30143	1167564	Gasto	Panamá	2022	10	Eficiencia Energética	Eficiencia Energética	1001191	55.081818
32087	1170153	Gasto	Panamá	2022	11	Eficiencia Energética	Eficiencia Energética	1001191	55.081818
32764	1170846	Gasto	Panamá	2022	12	Eficiencia Energética	Eficiencia Energética	1001191	55.081818
34075	1172260	Gasto	Panamá	2023	1	Eficiencia Energética	Eficiencia Energética	1001191	55.081818
35899	1174266	Gasto	Panamá	2023	2	Eficiencia Energética	Eficiencia Energética	1001191	55.081818
54986	F001198801	Gasto	Colombia	2024	2	Mantenimiento	Mtto. Contratos	1201	106.423400
65906	F001211630	Gasto	Colombia	2024	8	Mantenimiento	Mtto. Contratos	1270	174.000000
68765	F001215072	Gasto	Costa Rica	2024	10	Mantenimiento	Servicios Ctto.	1000389	138.461538
70068	F001216484	Gasto	España	2024	10	Eficiencia Energética	Suministros	1065	4906.100000
72132	F001218726	Gasto	España	2024	11	Eficiencia Energética	Suministros	1065	3623.490000
73456	F001220115	Gasto	España	2024	12	Eficiencia Energética	Suministros	1065	2570.330000
74417	F001221132	Gasto	España	2025	1	Eficiencia Energética	Suministros	1065	5101.810000
74522	F001221239	Gasto	España	2025	1	Eficiencia Energética	Suministros	1065	2486.810000
77086	F001223966	Gasto	España	2025	2	Eficiencia Energética	Suministros	1065	2869.510000
78335	F001225274	Gasto	España	2025	3	Eficiencia Energética	Suministros	1065	2816.320000
80347	F001227364	Gasto	España	2025	4	Eficiencia Energética	Suministros	1065	3700.430000
82043	F001229108	Gasto	España	2025	5	Eficiencia Energética	Suministros	1065	1173.430000
87565	F001238638	Gasto	España	2025	6	Eficiencia Energética	Suministros	1065	1396.150000
89004	F001240130	Gasto	España	2025	7	Eficiencia Energética	Suministros	1065	7026.640000
90849	F001243576	Gasto	España	2025	9	Mantenimiento	Servicios Ctto.	1000496	22.440000

	count
ID_BUSINESS_UNIT_MOD
42	170
43	226536
49	2409
50	10038
51	8261
2606	5569
1000000	3
1000020	26955
1000021	35433
1000043	5
1000070	8044
1000097	995
1000109	4

	ID_BUILDING	ID_BUSINESS_UNIT_MOD	n
1	9	50	2836
2	9	51	220
0	9	43	180
4	9	1000021	158
3	9	1000020	145
6	18	50	4231
8	18	1000020	45
7	18	2606	17
5	18	43	5
9	788	50	29
11	1065	50	2003
10	1065	49	912
13	1065	2606	846
12	1065	51	5
14	1201	43	1533
15	1270	43	896
16	1000389	43	1852
17	1000496	1000021	356
18	1001191	43	706
19	1001191	2606	4
20	1001203	51	302

	ID_ORDER	ID_BUILDING	ID_BUSINESS_UNIT	ID_BUSINESS_UNIT_MOD	COUNTRY	YEAR	MONTH
1215	F001129107	1001203	1000109	51	España	2021	1
3746	F001133112	1001203	1000109	51	España	2021	4
6005	F001137385	1001203	1000109	51	España	2021	7
8493	F001141218	1001203	1000109	51	España	2021	10
26297	1163481	1001191	0	43	Panamá	2022	7
26756	1163956	1001191	0	43	Panamá	2022	8
28145	1165508	1001191	0	43	Panamá	2022	9
30143	1167564	1001191	0	43	Panamá	2022	10
32087	1170153	1001191	0	43	Panamá	2022	11
32764	1170846	1001191	0	43	Panamá	2022	12
34075	1172260	1001191	0	43	Panamá	2023	1
35899	1174266	1001191	0	43	Panamá	2023	2
54986	F001198801	1201	0	43	Colombia	2024	2
65906	F001211630	1270	0	43	Colombia	2024	8
68765	F001215072	1000389	0	43	Costa Rica	2024	10
70068	F001216484	1065	0	50	España	2024	10
72132	F001218726	1065	0	50	España	2024	11
73456	F001220115	1065	0	50	España	2024	12
74417	F001221132	1065	0	50	España	2025	1
74522	F001221239	1065	0	50	España	2025	1

	ID_ORDER	COUNTRY	YEAR	MONTH	FM_RESPONSIBLE	FM_COST_TYPE	ID_BUILDING	ID_BUSINESS_UNIT_MOD	cost_float
245	F001127596	Colombia	2021	1	Eficiencia Energética	Suministros	1240	43	-115.497600
1385	F001129310	España	2021	3	Eficiencia Energética	Suministros	136	1000020	-153.230000
1386	F001129311	España	2021	3	Eficiencia Energética	Suministros	136	1000020	-383.590000
1387	F001129312	España	2021	3	Eficiencia Energética	Suministros	136	1000020	-222.750000
2321	F001130486	Costa Rica	2021	2	Eficiencia Energética	Suministros	1000476	43	-6.069108
2323	F001130488	Costa Rica	2021	2	Eficiencia Energética	Suministros	1000476	43	-44.684200
2324	F001130489	Costa Rica	2021	2	Eficiencia Energética	Suministros	1000476	43	-118.623754
3119	F001131816	Costa Rica	2021	4	Eficiencia Energética	Suministros	1000476	43	-23.236754
3120	F001131817	Costa Rica	2021	4	Eficiencia Energética	Suministros	1000476	43	-78.065723
3136	F001131835	Costa Rica	2021	4	Eficiencia Energética	Suministros	1000482	43	-16.899969
3192	F001131978	España	2021	5	Eficiencia Energética	Suministros	136	1000020	-3607.860000
4116	F001133913	Costa Rica	2021	5	Eficiencia Energética	Suministros	1000476	43	-6.215185
4117	F001133914	Costa Rica	2021	5	Eficiencia Energética	Suministros	1000476	43	-16.872354
4798	F001134813	Costa Rica	2021	6	Eficiencia Energética	Suministros	1000476	43	-16.397646
4799	F001134814	Costa Rica	2021	6	Eficiencia Energética	Suministros	1000476	43	-6.387969
10252	F001145768	Italia	2021	8	Eficiencia Energética	Suministros	1001012	1000097	-270.000000
10294	F001145814	Italia	2021	11	Eficiencia Energética	Suministros	1000993	1000097	-25.000000
11367	F001147437	España	2022	1	Eficiencia Energética	Suministros	105	1000020	-68.820000
11425	F001147495	España	2021	10	Eficiencia Energética	Suministros	138	1000020	-2595.690000
11450	F001147520	España	2021	12	Eficiencia Energética	Suministros	138	1000020	-609.270000

	ID_SUPPLIER	cost_float	n_registros
0	3005192	1.574396e+06	719
1	12	1.635065e+05	237
2	3001385	5.827215e+03	7
3	158	4.931740e+03	10
4	3007021	4.764283e+03	18
5	1662	1.880400e+03	5
6	3007026	1.016083e+03	9
7	4	6.452900e+02	3
8	3002884	2.965200e+02	1
9	2548	2.710000e+02	1
10	3004890	2.200000e+02	1
11	3004568	9.487000e+01	5
12	3004995	5.500000e+01	1
13	3003294	4.538462e+01	1
14	3004715	2.371000e+01	1
15	3004389	2.230000e+01	1
16	1811	2.188000e+01	1
17	3001934	2.109091e+01	1
18	3004303	2.070000e+01	1
19	3003255	2.034286e+01	1
20	3001092	1.963636e+01	1
21	3003582	3.446154e+00	1
22	3004607	1.600000e-03	1

	count	mean	std	min	25%	50%	75%	max
ID_BUILDING	1650.0	616661.073939	486695.604139	2.0	1018.25	1000559.0	1.001184e+06	1001613.00
ID_REGION	1650.0	164882.419394	371160.385033	2.0	5.00	32.0	9.400000e+01	1000164.00
STATUS	1650.0	0.500606	0.500151	0.0	0.00	1.0	1.000000e+00	1.00
FM_PERIMETER	1650.0	0.373939	0.483995	0.0	0.00	0.0	1.000000e+00	1.00
M2_AVAIL_PERIMETER	1650.0	234.415225	780.341649	0.0	0.00	0.0	1.919435e+02	23753.08
M2_AVAIL	1650.0	222.693352	771.409055	0.0	0.00	0.0	1.775225e+02	23753.08
M2_PROM_USO	1650.0	233.273661	708.844211	0.0	0.00	0.0	1.921979e+02	23753.08

	ID_BUILDING	num_registros_excluidos
0	1000759	467
1	357	421
2	1000454	380
3	1000869	368
4	1237	329
5	1244	191
6	1000774	170
7	961	155
8	618	149
9	1000770	129

	ID_ORDER	ID_BUILDING	COST_TYPE	FM_COST_TYPE	COST
86708	1389897	1281	Inversión	Obras	212.0736
86709	1391872	1213	Inversión	Obras	133956.9928
86710	1392770	1001116	Inversión	Obras	11657.5300

	ID_ORDER	ID_SUPPLIER	SUPPLIER_TYPE_MOD	COUNTRY	YEAR	MONTH	FM_RESPONSIBLE	FM_COST_TYPE	cost_float
79	F001126774	12	INTERNO	España	2021	1	Eficiencia Energética	Suministros	16.00
83	F001126780	12	INTERNO	España	2021	1	Eficiencia Energética	Suministros	16.00
316	F001127677	12	INTERNO	España	2021	1	Eficiencia Energética	Suministros	534.56
317	F001127678	12	INTERNO	España	2021	1	Eficiencia Energética	Suministros	662.02
318	F001127679	12	INTERNO	España	2021	1	Eficiencia Energética	Suministros	1703.54
319	F001127680	12	INTERNO	España	2021	1	Eficiencia Energética	Suministros	1728.05
324	F001127685	12	INTERNO	España	2021	1	Eficiencia Energética	Suministros	10967.47
326	F001127688	12	INTERNO	España	2021	1	Eficiencia Energética	Suministros	8034.95
333	F001127695	12	INTERNO	España	2021	1	Eficiencia Energética	Suministros	14508.93
334	F001127696	12	INTERNO	España	2021	1	Eficiencia Energética	Suministros	6037.69
336	F001127698	12	INTERNO	España	2021	1	Eficiencia Energética	Suministros	949.96
338	F001127700	12	INTERNO	España	2021	1	Eficiencia Energética	Suministros	1395.14
344	F001127706	12	INTERNO	España	2021	1	Eficiencia Energética	Suministros	2587.07
345	F001127707	12	INTERNO	España	2021	2	Eficiencia Energética	Suministros	16.00
352	F001127714	12	INTERNO	España	2021	1	Eficiencia Energética	Suministros	1840.76
353	F001127715	12	INTERNO	España	2021	2	Eficiencia Energética	Suministros	16.00
361	F001127723	12	INTERNO	España	2021	1	Eficiencia Energética	Suministros	554.92
362	F001127724	12	INTERNO	España	2021	1	Eficiencia Energética	Suministros	1038.19
364	F001127727	12	INTERNO	España	2021	1	Eficiencia Energética	Suministros	1020.65
365	F001127728	12	INTERNO	España	2021	1	Eficiencia Energética	Suministros	1995.58

	ID_SUPPLIER	FM_RESPONSIBLE	coste_total	n_registros	coste_medio
1	4	Obras Proyectos	3.797200e+02	1	379.720000
0	4	Gestión Espacios	2.115400e+02	1	211.540000
2	4	Oficina Técnica	5.403000e+01	1	54.030000
3	12	Eficiencia Energética	1.635065e+05	237	689.900633
4	158	Eficiencia Energética	4.931740e+03	10	493.174000
5	1662	Licencias	1.880400e+03	5	376.080000
6	1811	Mantenimiento	2.188000e+01	1	21.880000
7	2548	Mantenimiento	2.710000e+02	1	271.000000
8	3001092	Mantenimiento	1.963636e+01	1	19.636364
9	3001385	Mantenimiento	5.827215e+03	7	832.459341
10	3001934	Mantenimiento	2.109091e+01	1	21.090909
11	3002884	Mantenimiento	2.965200e+02	1	296.520000
12	3003255	Mantenimiento	2.034286e+01	1	20.342857
13	3003294	Mantenimiento	4.538462e+01	1	45.384615
14	3003582	Mantenimiento	3.446154e+00	1	3.446154
15	3004303	Mantenimiento	2.070000e+01	1	20.700000
16	3004389	Mantenimiento	2.230000e+01	1	22.300000
17	3004568	Eficiencia Energética	9.487000e+01	5	18.974000
18	3004607	Mantenimiento	1.600000e-03	1	0.001600
19	3004715	Mantenimiento Multipunto	2.371000e+01	1	23.710000
20	3004890	Mantenimiento	2.200000e+02	1	220.000000
21	3004995	Mantenimiento Multipunto	5.500000e+01	1	55.000000
22	3005192	Eficiencia Energética	1.574396e+06	719	2189.701864
23	3007021	Eficiencia Energética	4.764283e+03	18	264.682375
24	3007026	Eficiencia Energética	1.016083e+03	9	112.898148

	ID_ORDER	ID_SUPPLIER	FM_RESPONSIBLE	SUPPLIER_TYPE_MOD	cost_float
82055	F001229120	3004568	Eficiencia Energética	INTERNO	15.90
82056	F001229121	3004568	Eficiencia Energética	INTERNO	15.90
82057	F001229122	3004568	Eficiencia Energética	INTERNO	15.90
87577	F001238650	3004568	Eficiencia Energética	INTERNO	8.09
89282	F001240421	3004568	Eficiencia Energética	INTERNO	39.08
91241	1345603	1662	Licencias	INTERNO	0.00
91242	1346759	1662	Licencias	INTERNO	0.00
91243	1350758	1662	Licencias	INTERNO	0.00
91244	1350759	1662	Licencias	INTERNO	0.00
91245	1350812	1662	Licencias	INTERNO	0.00
91246	1367879	1662	Licencias	INTERNO	0.00
91247	1373667	1662	Licencias	INTERNO	0.00
91248	1374043	1662	Licencias	INTERNO	0.00
91249	1379463	1662	Licencias	INTERNO	0.00
91250	1379469	1662	Licencias	INTERNO	0.00
91251	1379472	1662	Licencias	INTERNO	0.00
91252	1381749	1662	Licencias	INTERNO	0.00
91253	1381750	1662	Licencias	INTERNO	0.00
91254	1381751	1662	Licencias	INTERNO	0.00
91255	1381753	1662	Licencias	INTERNO	0.00

	ID_ORDER	ID_SUPPLIER	FM_RESPONSIBLE	SUPPLIER_TYPE_MOD	cost_float	cost_float_mod
59707	F001203888	3006536	Eficiencia Energética	EXTERNO	NaN	NaN
59710	F001203891	3006536	Eficiencia Energética	EXTERNO	NaN	NaN
59713	F001203894	3006536	Eficiencia Energética	EXTERNO	NaN	NaN
59716	F001203897	3006536	Eficiencia Energética	EXTERNO	NaN	NaN
59719	F001203900	3006536	Eficiencia Energética	EXTERNO	NaN	NaN
59722	F001203903	3006536	Eficiencia Energética	EXTERNO	NaN	NaN
59725	F001203906	3006536	Eficiencia Energética	EXTERNO	NaN	NaN
59728	F001203909	3006536	Eficiencia Energética	EXTERNO	NaN	NaN
59731	F001203912	3006536	Eficiencia Energética	EXTERNO	NaN	NaN
59734	F001203915	3006536	Eficiencia Energética	EXTERNO	NaN	NaN
59737	F001203918	3006536	Eficiencia Energética	EXTERNO	NaN	NaN
60568	F001204797	3006536	Eficiencia Energética	EXTERNO	NaN	NaN
60573	F001204802	3006536	Eficiencia Energética	EXTERNO	NaN	NaN
60574	F001204803	3006536	Eficiencia Energética	EXTERNO	NaN	NaN
60578	F001204807	3006536	Eficiencia Energética	EXTERNO	NaN	NaN
62239	F001206667	3006536	Eficiencia Energética	EXTERNO	NaN	NaN
64435	F001209124	3006536	Eficiencia Energética	EXTERNO	NaN	NaN
67530	F001213812	3006536	Eficiencia Energética	EXTERNO	NaN	NaN
69199	F001215603	3006536	Eficiencia Energética	EXTERNO	NaN	NaN
81775	F001228815	3006538	Eficiencia Energética	EXTERNO	NaN	NaN

	ID_ORDER	COST_TYPE	COUNTRY	ID_BUILDING	YEAR_MONTH	FM_RESPONSIBLE	FM_COST_TYPE	COST	ID_CUSTOMER	ID_SUPPLIER	...	YEAR	QUARTER	cost_float	SUPPLIER_TYPE_MOD	ID_BUSINESS_UNIT_MOD	bu_corr_flag	CONCAT_key	cost_float_mod	SUPPLIER_TYPE_MOD_2	FM_RESPONSIBLE_MOD
0	F001108989	Gasto	España	1001131	202111	Mantenimiento Multipunto	Mtto. Contratos	77.00	3004581	3004832	...	2021	4	77.00	EXTERNO	1000021	False	3004832Mantenimiento MultipuntoEXTERNO	77.00	EXTERNO	Mantenimiento
1	F001119565	Gasto	España	1000026	202107	Mantenimiento	Mtto. Contratos	36.95	13	1532	...	2021	3	36.95	EXTERNO	1000021	False	1532MantenimientoEXTERNO	36.95	EXTERNO	Mantenimiento
2	F001119567	Gasto	España	1000515	202107	Mantenimiento	Mtto. Contratos	36.95	222	1532	...	2021	3	36.95	EXTERNO	1000021	False	1532MantenimientoEXTERNO	36.95	EXTERNO	Mantenimiento

	ID_BUILDING	ID_REGION	COUNTRY	STATUS	FM_PERIMETER	SERVICE_LEVEL	TIPO_USO	M2_AVAIL_PERIMETER	M2_AVAIL	M2_PROM_USO
0	118	17	España	1	1	Compact	Bingo	776.91	776.91	903.815556
1	648	2	España	0	0	0	Sin Uso Actual	0.00	0.00	0.000000
2	924	2	España	0	0	0	Sin Uso Actual	0.00	0.00	0.000000

	ID_ORDER	COST_TYPE	COUNTRY	ID_BUILDING	YEAR_MONTH	FM_RESPONSIBLE	FM_COST_TYPE	COST	ID_CUSTOMER	ID_SUPPLIER	...	YEAR	QUARTER	cost_float	SUPPLIER_TYPE_MOD	ID_BUSINESS_UNIT_MOD	bu_corr_flag	CONCAT_key	cost_float_mod	SUPPLIER_TYPE_MOD_2	FM_RESPONSIBLE_MOD
7	F001126168	Gasto	España	1136	202107	Mantenimiento	Mtto. Contratos	43.670000	161	1532	...	2021	3	43.670000	EXTERNO	51	False	1532MantenimientoEXTERNO	43.670000	EXTERNO	Mantenimiento
8	F001126191	Gasto	México	1000454	202101	Eficiencia Energética	Suministros	1485.231000	3001510	3006565	...	2021	1	1485.231000	EXTERNO	43	False	3006565Eficiencia EnergéticaEXTERNO	1485.231000	EXTERNO	Eficiencia Energética
128	F001127165	Gasto	Panamá	1088	202101	Eficiencia Energética	Suministros	12977.763636	2047	3004370	...	2021	1	12977.763636	EXTERNO	43	False	3004370Eficiencia EnergéticaEXTERNO	12977.763636	EXTERNO	Eficiencia Energética
134	F001127171	Gasto	Panamá	1000534	202101	Eficiencia Energética	Suministros	3365.409091	2047	3004370	...	2021	1	3365.409091	EXTERNO	43	False	3004370Eficiencia EnergéticaEXTERNO	3365.409091	EXTERNO	Eficiencia Energética
140	F001127177	Gasto	Panamá	1000869	202101	Eficiencia Energética	Suministros	7774.654545	2047	3004370	...	2021	1	7774.654545	EXTERNO	43	False	3004370Eficiencia EnergéticaEXTERNO	7774.654545	EXTERNO	Eficiencia Energética
...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...
322878	701947735	Gasto	España	1001599	202509	Obras Proyectos	Obras	220.000000	2676	3007016	...	2025	3	220.000000	EXTERNO	2606	False	3007016Obras ProyectosEXTERNO	220.000000	EXTERNO	Obras Proyectos
322879	701947755	Inversión	España	1001601	202509	Obras Proyectos	Obras	498.000000	3003189	3003277	...	2025	3	498.000000	EXTERNO	2606	False	3003277Obras ProyectosEXTERNO	498.000000	EXTERNO	Obras Proyectos
322914	701947928	Gasto	España	1001600	202509	Obras Proyectos	Obras	80.000000	1434	3007016	...	2025	3	80.000000	EXTERNO	2606	False	3007016Obras ProyectosEXTERNO	80.000000	EXTERNO	Obras Proyectos
323086	701948953	Gasto	España	1001601	202509	Obras Proyectos	Obras	326.000000	3003189	3005230	...	2025	3	326.000000	EXTERNO	2606	False	3005230Obras ProyectosEXTERNO	326.000000	EXTERNO	Obras Proyectos
323087	701948955	Gasto	España	1001598	202509	Obras Proyectos	Obras	348.750000	3005142	3006714	...	2025	3	348.750000	EXTERNO	2606	False	3006714Obras ProyectosEXTERNO	348.750000	EXTERNO	Obras Proyectos

	cost_float_mod
count	306775.000000
mean	583.043000
std	2271.779234
min	-40880.070000
5%	0.000000
25%	0.000000
50%	47.727273
75%	233.306000
95%	2983.501250
max	118503.992000

	ID_REGION	num_registros	media_coste
0	2	26578	566.162446
78	1000022	25459	187.959776
52	103	23785	272.650425
45	94	23041	1426.885862
54	105	19484	377.718900
6	8	16130	357.345006
64	117	12449	280.291381
14	17	11958	337.825073
26	32	8727	479.636750
93	1000039	8725	758.855437
1	3	7571	228.657442
67	1000005	6830	179.102905
79	1000023	6027	215.830859
41	48	5604	309.991795
3	5	4162	390.981887

	ID_ORDER	ID_BUILDING	FM_RESPONSIBLE_MOD	FM_COST_TYPE	COUNTRY	ID_REGION_GRUPO	TIPO_USO	cost_float_mod
62654	F001212407	1211	Eficiencia Energética	Suministros	Colombia	105	Casino Tradicional	118503.9920
243570	701824718	1000270	Mantenimiento	Servicios Extra	España	18	Bingo	112260.5700
80793	F001234850	1000451	Eficiencia Energética	Suministros	México	1000031	Casino Tradicional	111152.7625
61310	F001210442	1211	Eficiencia Energética	Suministros	Colombia	105	Casino Tradicional	108664.5720
80792	F001234849	1000451	Eficiencia Energética	Suministros	México	1000031	Casino Tradicional	108252.6065
80667	F001234679	1000399	Eficiencia Energética	Suministros	México	1000015	Casino Electrónico	96920.3400
80666	F001234678	1000399	Eficiencia Energética	Suministros	México	1000015	Casino Electrónico	93819.8780
62667	F001212424	1000252	Eficiencia Energética	Suministros	Colombia	105	Casino Tradicional	87107.7480
61323	F001210455	1000252	Eficiencia Energética	Suministros	Colombia	105	Casino Tradicional	80294.9750
62683	F001212450	1213	Eficiencia Energética	Suministros	Colombia	105	Casino Tradicional	79072.5610

EIBS GLOBAL PROJECT - Grupo 13¶

NEITH¶

1. EXPLICACION DEL CASO:¶

LISTA DE REQUERIMIENTOS¶

ESTRATEGIA PARA ABORDAR EL PROYECTO¶

FUENTES DE DATOS¶

ANALISIS EXPLORATIVOS DE DATOS DE LA FD1¶

Cargamos e importamos las librerias que se van a usar.¶

Conectamos con nuestro DRIVE¶

Cargamos el dataset FD1 que nos han entregado con el histórico de contratación del equipo de FM.¶

Comprobaciones, análisis inicial y transformación del dataset¶

Conclusión del equipo FM¶

Conclusión del equipo FM¶

Conclusión del equipo FM¶

Conclusiones del equipo de FM¶

Conclusiones del equipo de FM¶

Conclusión del equipo de FM¶

Conclusión del equipo de FM¶

Conclusión del equipo de FM¶

Eliminación de registros de Marruecos e Italia¶

Unificación en FM_RESPONSIBLE¶

Eliminación de las clases minoritarias en FM_RESPONSIBLE¶

Obtención de dataset df_fd1_v3¶

Generación del notebook en HTML¶

Introducción del variables exógenas del Catalogo_inmuebles del equipo de FM¶

TRANSFORMACIÓN DEL DATASET V3 A V4¶

Analisis Descriptivo de los Datos del Catalago_inmuebles¶

Conclusión del equipo de FM¶

Analisis y eliminación de registros del dataset df_fd1_v3 con inmuebles de baja o fuera del perímetro¶

Eliminación de los registros de CAPEX (Inversión)¶

Exportación de las versiones v4 al Drive¶

TRANSFORMACIÓN DEL DATASET V4 A V5¶

Cargamos el dataset df_fd1_v4_opex¶

Conclusión del equipo FM (pendiente)¶

Exportamos a Drive del listado de registros con COUNTRY distinto a COUNTRY_CATALOGO¶

Limpieza del df_fd1_v4_opex para dejarlo con una dimensionalidad óptima para modelizar a través de ML (Machine Learning) y/o ST (Series Temporales)¶

Segundo Analisis Exploratorio de Datos¶

Creamos la variable ID_REGION_GROUP (pendiente)¶

Exportación a Excel de la clasificación realizado relacionada¶

Eliminamos ID_REGION y dejamos la nueva ID_REGION_GRUPO¶

Análisis de los outliers de cost_float_mod¶

Exportamos a Drive los outliers¶

Conclusión del equipo FM (Outliers cost_float_mod)¶

Comprobación de la unicidad de las clases de la variable ID_ORDER¶

Análisis de los estadísticos de las variables de df_fd1_v5¶

Exportamos a Drive la v5 preparada para iniciar la segregación según la 1a iteración (2021-2023) y la 2a iteración (2021-2024).¶

PREPARACION Y AJUSTE DEL DATASET PARA LA FASE DE ENTRENAMIENTO¶

Cargamos el dataset df_fd1_v5 del Drive¶

Eliminamos variables sin valor en la modelización¶

Generamos las versiones para cada iteración.¶

Exportación a Drive de los dataset pre-entrenamiento de las iteraciones 1 y 2.¶

Exportamos a Drive ambos conjuntos¶

Guardamos el notebook en formato html¶

Modelización con series temporales (ST) para la iteración 1.¶

Cargamos datasets df_fd1_v5_ITE1 y df_fd1_v5_Real_ITE1_2024.¶

Exploración de ambos datasets - verificación que son correctos¶

Generamos un nuevo dataframe df_train_fd1_v5_ITE1 que contenga las series a predecir según la pareja ID_BUILDING - FM_COST_TYPE.¶

Definimos funciones para cada kpi de interes que nos permita decidir sobre el método a utilizar para cada tipo de serie, dada una pareja ID_BUILDING - FM_COST_TYPE¶

Función safe_autocorr¶

Función seasonal_strength_stl¶

Función robust_outlier_rate¶

Función trend_slope¶

Función adi_metrics¶

Función cv2_positivos¶

Función max_zero_run¶

Función occ_prob_and_drift¶

Función zero_ratio¶

Generamos 3 nuevas funciones que nos definen variables de contexto (VAR_CTX) que nos permitan tomar mejores decisiones a la hora de decidir sobre una serie.¶

Función has_min_history¶

Función effective_value_ratio¶

Función total_cost_guard¶

Integramos las funciones VAR_CTX en una sola función que nos permitirá disponer de un unico indicador de contexto binario.¶

Función elegibilidad_global¶

(Pendiente) incorporar tabla de referencias para definir el nivel de clasificación del modelo en función de la serie.¶

Mapa de Diagnósis de series con KPIs y VAR_CTX con Modelo Predictivo Sugerido (versión Alto Nivel Decisión).¶

Función diagnosticar_y_sugerir_modelo¶

Función calcular_kpis_y_diagnostico¶

KPIs aparcados en el primer mapa de decisiones¶

Aplicamos diagnóstico y modelo sugerido a todas las series del dataframe df_train_fd1_v5_ITE1.¶

Interpretación de los resultados del diagnóstico¶

Unificación en `FM_RESPONSIBLE`¶

Eliminación de las clases minoritarias en `FM_RESPONSIBLE`¶

Obtención de dataset `df_fd1_v3`¶

Cargamos el dataset `df_fd1_v4_opex`¶

Exportamos a Drive del listado de registros con `COUNTRY` distinto a `COUNTRY_CATALOGO`¶

Creamos la variable `ID_REGION_GROUP` (pendiente)¶

Eliminamos `ID_REGION` y dejamos la nueva `ID_REGION_GRUPO`¶

Análisis de los outliers de `cost_float_mod`¶

Conclusión del equipo FM (Outliers `cost_float_mod`)¶

Comprobación de la unicidad de las clases de la variable `ID_ORDER`¶

Análisis de los estadísticos de las variables de `df_fd1_v5`¶

Cargamos el dataset `df_fd1_v5` del Drive¶

Generamos un nuevo dataframe `df_train_fd1_v5_ITE1` que contenga las series a predecir según la pareja `ID_BUILDING` - `FM_COST_TYPE`.¶

Definimos funciones para cada kpi de interes que nos permita decidir sobre el método a utilizar para cada tipo de serie, dada una pareja `ID_BUILDING` - `FM_COST_TYPE`¶

Función `safe_autocorr`¶

Función `seasonal_strength_stl`¶

Función `robust_outlier_rate`¶

Función `trend_slope`¶

Función `adi_metrics`¶

Función `cv2_positivos`¶

Función `max_zero_run`¶

Función `occ_prob_and_drift`¶

Función `zero_ratio`¶

Función `has_min_history`¶

Función `effective_value_ratio`¶

Función `total_cost_guard`¶

Función `elegibilidad_global`¶

Función `diagnosticar_y_sugerir_modelo`¶

Función `calcular_kpis_y_diagnostico`¶

Aplicamos diagnóstico y modelo sugerido a todas las series del dataframe `df_train_fd1_v5_ITE1`.¶

Cargamos el dataframe train para modelizar: `df_train_fd1_v5_ITE1_ARIMA`¶

Cargamos el dataframe train para modelizar: `df_train_fd1_v5_ITE1_CROSTON`¶

	count	unique	top	freq	mean	std	min	25%	50%	75%	max
ID_ORDER	306775	306775	701948957	1	NaN	NaN	NaN	NaN	NaN	NaN	NaN
COUNTRY	306775	7	España	115099	NaN	NaN	NaN	NaN	NaN	NaN	NaN
ID_BUILDING	306775.0	NaN	NaN	NaN	501178.645929	499922.745257	2.0	1093.0	1000004.0	1000608.0	1001588.0
FM_COST_TYPE	306775	8	Mtto. Correctivo	176085	NaN	NaN	NaN	NaN	NaN	NaN	NaN
MONTH	306775.0	NaN	NaN	NaN	6.308598	3.315537	1.0	3.0	6.0	9.0	12.0
YEAR	306775.0	NaN	NaN	NaN	2023.123093	1.362546	2021.0	2022.0	2023.0	2024.0	2025.0
cost_float_mod	306775.0	NaN	NaN	NaN	583.043	2271.779234	-40880.07	0.0	47.727273	233.306	118503.992
SUPPLIER_TYPE_MOD_2	306775	2	EXTERNO	199957	NaN	NaN	NaN	NaN	NaN	NaN	NaN
FM_RESPONSIBLE_MOD	306775	5	Mantenimiento	252512	NaN	NaN	NaN	NaN	NaN	NaN	NaN
TIPO_USO	306775	14	Casino Electrónico	106781	NaN	NaN	NaN	NaN	NaN	NaN	NaN
COUNTRY_CATALOGO	306775	7	España	114607	NaN	NaN	NaN	NaN	NaN	NaN	NaN
ID_REGION_GRUPO	306775	55	2	26578	NaN	NaN	NaN	NaN	NaN	NaN	NaN

	ID_ORDER	ID_BUILDING	FM_COST_TYPE	MONTH	YEAR	cost_float_mod	SUPPLIER_TYPE_MOD_2	FM_RESPONSIBLE_MOD	TIPO_USO	COUNTRY_CATALOGO	ID_REGION_GRUPO	COUNTRY_DEF
count	174626	1.746260e+05	174626	174626.000000	174626.000000	174626.000000	174626	174626	174626	174626	174626	174626
unique	174626	NaN	8	NaN	NaN	NaN	2	5	14	7	55	7
top	701760631	NaN	Mtto. Correctivo	NaN	NaN	NaN	EXTERNO	Mantenimiento	Casino Electrónico	España	2	España
freq	1	NaN	98919	NaN	NaN	NaN	118758	144266	60846	67229	15889	67229
mean	NaN	4.943516e+05	NaN	6.684005	2022.099527	554.836295	NaN	NaN	NaN	NaN	NaN	NaN
std	NaN	4.998727e+05	NaN	3.401633	0.800366	2046.978716	NaN	NaN	NaN	NaN	NaN	NaN
min	NaN	2.000000e+00	NaN	1.000000	2021.000000	-40880.070000	NaN	NaN	NaN	NaN	NaN	NaN
25%	NaN	1.086000e+03	NaN	4.000000	2021.000000	0.000000	NaN	NaN	NaN	NaN	NaN	NaN
50%	NaN	1.572000e+03	NaN	7.000000	2022.000000	50.000000	NaN	NaN	NaN	NaN	NaN	NaN
75%	NaN	1.000565e+06	NaN	10.000000	2023.000000	233.116000	NaN	NaN	NaN	NaN	NaN	NaN
max	NaN	1.001488e+06	NaN	12.000000	2023.000000	75977.040000	NaN	NaN	NaN	NaN	NaN	NaN

	ID_ORDER	ID_BUILDING	FM_COST_TYPE	MONTH	YEAR	cost_float_mod	SUPPLIER_TYPE_MOD_2	FM_RESPONSIBLE_MOD	TIPO_USO	COUNTRY_CATALOGO	ID_REGION_GRUPO	COUNTRY_DEF
0	F001176754	1000395	Suministros	2	2024	9304.615385	EXTERNO	Eficiencia Energética	Casino Tradicional	Costa Rica	1000010	Costa Rica
1	F001176801	9	Servicios Ctto.	3	2024	5040.000000	EXTERNO	Obras Proyectos	Oficinas	España	2	España
2	F001180765	247	Mtto. Contratos	1	2024	80.000000	EXTERNO	Mantenimiento	Salón de Juego	España	11	España
3	F001182650	1000270	Mtto. Contratos	4	2024	540.870000	EXTERNO	Mantenimiento	Bingo	España	18	España
4	F001182684	1001152	Mtto. Contratos	1	2024	80.000000	EXTERNO	Mantenimiento	Salón de Juego	España	6	España

	ID_ORDER	ID_BUILDING	FM_COST_TYPE	MONTH	YEAR	cost_float_mod	SUPPLIER_TYPE_MOD_2	FM_RESPONSIBLE_MOD	TIPO_USO	COUNTRY_CATALOGO	ID_REGION_GRUPO	COUNTRY_DEF
count	69290.0	6.929000e+04	69290	69290.000000	69290.0	69290.000000	69290	69290	69290	69290	69290	69290
unique	69290.0	NaN	8	NaN	NaN	NaN	2	5	14	7	55	7
top	701868793.0	NaN	Mtto. Correctivo	NaN	NaN	NaN	EXTERNO	Mantenimiento	Casino Electrónico	España	2	España
freq	1.0	NaN	39979	NaN	NaN	NaN	44974	57764	23718	26090	6028	26090
mean	NaN	5.141913e+05	NaN	6.474340	2024.0	599.582213	NaN	NaN	NaN	NaN	NaN	NaN
std	NaN	4.997670e+05	NaN	3.434215	0.0	2571.563909	NaN	NaN	NaN	NaN	NaN	NaN
min	NaN	2.000000e+00	NaN	1.000000	2024.0	-12960.230000	NaN	NaN	NaN	NaN	NaN	NaN
25%	NaN	1.096000e+03	NaN	4.000000	2024.0	0.000000	NaN	NaN	NaN	NaN	NaN	NaN
50%	NaN	1.000028e+06	NaN	7.000000	2024.0	51.500000	NaN	NaN	NaN	NaN	NaN	NaN
75%	NaN	1.000741e+06	NaN	9.000000	2024.0	228.373625	NaN	NaN	NaN	NaN	NaN	NaN
max	NaN	1.001586e+06	NaN	12.000000	2024.0	118503.992000	NaN	NaN	NaN	NaN	NaN	NaN

	ID_BUILDING	FM_COST_TYPE	FECHA	YEAR	MONTH	cost_float_mod	COUNTRY_DEF	ID_REGION_GRUPO	TIPO_USO	SUPPLIER_TYPE_MOD_2	FM_RESPONSIBLE_MOD	diagnosis	model_suggested	model_group
0	2	Licencias	2021-01-01	2021	1	1145.46	España	2	Oficinas	INTERNO	Licencias	intermitente	Croston\|SBA	CROSTON
1	2	Licencias	2021-02-01	2021	2	55.95	España	2	Oficinas	INTERNO	Licencias	intermitente	Croston\|SBA	CROSTON
2	2	Licencias	2021-03-01	2021	3	0.00	España	2	Oficinas	INTERNO	Licencias	intermitente	Croston\|SBA	CROSTON

	ID_BUILDING	FM_COST_TYPE	FECHA	YEAR	MONTH	cost_float_mod	COUNTRY_DEF	ID_REGION_GRUPO	TIPO_USO	SUPPLIER_TYPE_MOD_2	FM_RESPONSIBLE_MOD	diagnosis	model_suggested	model_group
0	2	Licencias	2024-12-01	2024	12	202.8	España	2	Oficinas	INTERNO	Licencias	intermitente	Croston\|SBA	CROSTON
1	2	Obras	2024-02-01	2024	2	54.6	España	2	Oficinas	INTERNO	Gestión Espacios	intermitente	Croston\|SBA	CROSTON
2	2	Servicios Extra	2024-01-01	2024	1	0.0	España	2	Oficinas	INTERNO	Mantenimiento	intermitente	Croston\|SBA	CROSTON

	ID_BUILDING	FM_COST_TYPE	FECHA	yhat	y
0	2	Licencias	2024-12-01	353.798468	202.8
1	2	Obras	2024-02-01	130.811893	54.6
2	2	Servicios Extra	2024-01-01	116.867916	0.0
3	2	Servicios Extra	2024-02-01	116.867916	0.0
4	9	Licencias	2024-01-01	253.568541	1600.0

	ID_BUILDING	FM_COST_TYPE	WAPE	SMAPE
0	2	Licencias	0.744568	54.257594
1	2	Obras	1.395822	82.208203
2	2	Servicios Extra	NaN	200.000000
3	9	Licencias	1.331861	168.038835
4	9	Obras	1.008057	128.720687

	ID_BUILDING	FM_COST_TYPE	FECHA	YEAR	MONTH	cost_float_mod	COUNTRY_DEF	ID_REGION_GRUPO	TIPO_USO	SUPPLIER_TYPE_MOD_2	FM_RESPONSIBLE_MOD
0	2	Eficiencia Energética	2021-12-01	2021	12	0.00	España	2	Oficinas	INTERNO	Eficiencia Energética
1	2	Licencias	2021-01-01	2021	1	1145.46	España	2	Oficinas	INTERNO	Licencias
2	2	Licencias	2021-02-01	2021	2	55.95	España	2	Oficinas	INTERNO	Licencias
3	2	Licencias	2021-03-01	2021	3	0.00	España	2	Oficinas	INTERNO	Licencias
4	2	Licencias	2021-04-01	2021	4	0.00	España	2	Oficinas	INTERNO	Licencias

	baseline	n_pairs	MAE_mean	MAE_median	WAPE_mean	WAPE_median	SMAPE_mean	SMAPE_median	MASE1_mean	MASE1_median	MASE12_mean	MASE12_median
0	MedianaEstacional	2231	663.668062	172.509848	1.908440e+08	0.865073	68.061196	56.882021	3.814405	1.111074	4.510965	1.068718
1	Naive1	2231	566.861543	159.841667	2.677334e+09	0.931867	63.702404	50.000000	3.289550	1.077607	4.523964	0.982132
2	SNaive12	2231	589.609083	171.250000	1.811532e+09	0.835532	61.992876	51.275403	3.305935	1.180648	4.535538	1.095457