Pandas avanzado: limpieza, series temporales y datos faltantes

import numpy as np
import pandas as pd
from sklearn.model_selection import train_test_split, cross_val_score
from sklearn.pipeline import Pipeline
from sklearn.impute import SimpleImputer, KNNImputer
from sklearn.experimental import enable_iterative_imputer  # noqa
from sklearn.impute import IterativeImputer
from sklearn.ensemble import RandomForestRegressor
from sklearn.compose import ColumnTransformer

np.random.seed(42)
N = 1000

# Generar dataset con nulos MCAR
df = pd.DataFrame({
    'edad':      np.random.normal(35, 12, N).clip(0, 90),
    'ingreso':   np.random.lognormal(10.5, 0.8, N),
    'antiguedad':np.random.uniform(0, 20, N),
})
# Introducir 15% de nulos MCAR en cada columna
for col in df.columns:
    mask = np.random.random(N) < 0.15
    df.loc[mask, col] = np.nan

# Comparar metodos de imputacion con el mismo modelo
y = np.random.binomial(1, 0.3, N)  # target sintetico
X = df.copy()

X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=42)

metodos = {
    'Media':          SimpleImputer(strategy='mean'),
    'Mediana':        SimpleImputer(strategy='median'),
    'KNN (k=5)':      KNNImputer(n_neighbors=5),
    'Iterative (MICE)': IterativeImputer(random_state=42),
}

for nombre, imputer in metodos.items():
    # CLAVE: el imputer va DENTRO del pipeline → sin leakage
    pipe = Pipeline([
        ('imputer', imputer),
        ('model', RandomForestRegressor(n_estimators=50, random_state=42, n_jobs=-1))
    ])
    scores = cross_val_score(pipe, X_train, y_train, cv=5, scoring='r2')
    print(f"{nombre:25s} R²: {scores.mean():.3f} ± {scores.std():.3f}")

import numpy as np
import pandas as pd
from sklearn.ensemble import IsolationForest

np.random.seed(42)
N = 1000
data = np.concatenate([
    np.random.normal(100, 15, N),  # datos normales
    [-50, 200, 250, 300]          # outliers extremos
])
df = pd.DataFrame({'valor': data})

# 1) IQR: cuartiles
Q1, Q3 = df['valor'].quantile([0.25, 0.75])
IQR = Q3 - Q1
limite_inf = Q1 - 1.5 * IQR
limite_sup = Q3 + 1.5 * IQR
outliers_iqr = df[(df['valor'] < limite_inf) | (df['valor'] > limite_sup)]
print(f"IQR outliers: {len(outliers_iqr)} (limites: [{limite_inf:.1f}, {limite_sup:.1f}])")

# 2) Z-score: asume distribucion normal
from scipy import stats
z_scores = np.abs(stats.zscore(df['valor']))
outliers_z = df[z_scores > 3]
print(f"Z-score outliers (|z|>3): {len(outliers_z)}")

# 3) Isolation Forest: detecta anomalias multivariadas, NO asume distribucion
iso = IsolationForest(contamination=0.05, random_state=42)
df['outlier_if'] = iso.fit_predict(df[['valor']])  # -1 = outlier
n_outliers_if = (df['outlier_if'] == -1).sum()
print(f"Isolation Forest outliers: {n_outliers_if}")

# Comparacion
print("\nPrimeros outliers detectados por cada metodo:")
print("IQR:", outliers_iqr.head().values.flatten().tolist())
print("Z-score:", outliers_z.head().values.flatten().tolist())
print("IF:", df[df['outlier_if']==-1].head().values.flatten().tolist())

import pandas as pd
import numpy as np
import matplotlib.pyplot as plt

# Generar serie diaria de 1 anio con tendencia y estacionalidad
np.random.seed(42)
fechas = pd.date_range('2023-01-01', '2023-12-31', freq='D')
N = len(fechas)
tendencia = np.linspace(100, 200, N)
estacionalidad = 20 * np.sin(2 * np.pi * np.arange(N) / 365)
ruido = np.random.normal(0, 5, N)
ventas = tendencia + estacionalidad + ruido

ts = pd.Series(ventas, index=fechas, name='ventas')

# 1) Resample: cambiar frecuencia
semanal = ts.resample('W').sum()
mensual = ts.resample('ME').mean()
print("Primeras semanas:", semanal.head(3).round(1).tolist())

# 2) Rolling: media/mediana movil
ts_ma7  = ts.rolling(window=7,   min_periods=1).mean()
ts_ma30 = ts.rolling(window=30,  min_periods=1).mean()
ts_ma90 = ts.rolling(window=90,  min_periods=1).mean()

# 3) Shift: lag features (predictoras del pasado)
ts_lag1  = ts.shift(1)
ts_lag7  = ts.shift(7)
ts_diff1 = ts.diff(1)  # diferencia dia a dia
ts_pct   = ts.pct_change()  # cambio porcentual

# 4) Expanding: acumulado hasta hoy
ts_cumsum = ts.expanding().sum()

# 5) Visualizacion
fig, axes = plt.subplots(2, 2, figsize=(14, 8))

axes[0,0].plot(ts.index, ts, alpha=0.4, label='Original')
axes[0,0].plot(ts.index, ts_ma7, label='MA 7d', linewidth=2)
axes[0,0].plot(ts.index, ts_ma30, label='MA 30d', linewidth=2)
axes[0,0].set_title('Medias moviles (suavizado)')
axes[0,0].legend()

axes[0,1].plot(mensual.index, mensual.values, marker='o', color='teal')
axes[0,1].set_title('Resample mensual (media)')

axes[1,0].plot(ts_diff1.index, ts_diff1.values, alpha=0.6)
axes[1,0].axhline(0, color='red', linestyle='--', alpha=0.5)
axes[1,0].set_title('Diff(1): cambio dia a dia')

axes[1,1].plot(ts_pct.index, ts_pct.values * 100, alpha=0.6, color='orange')
axes[1,1].set_title('Cambio porcentual diario (%)')

plt.tight_layout()
plt.show()