Cross-Validation — Reliable Performance Estimation

import numpy as np
import pandas as pd
from sklearn.datasets import load_breast_cancer
from sklearn.model_selection import (KFold, StratifiedKFold, cross_val_score,
                                      cross_validate, TimeSeriesSplit, leave_one_out)
from sklearn.pipeline import Pipeline
from sklearn.preprocessing import StandardScaler
from sklearn.ensemble import RandomForestClassifier, GradientBoostingClassifier
from sklearn.metrics import make_scorer, roc_auc_score, f1_score

cancer = load_breast_cancer()
X, y = cancer.data, cancer.target

model = Pipeline([
    ("scaler", StandardScaler()),
    ("clf", RandomForestClassifier(n_estimators=100, random_state=42, n_jobs=-1)),
])

# SINGLE SPLIT vs CROSS-VALIDATION VARIABILITY
from sklearn.model_selection import train_test_split
single_split_scores = []
for seed in range(10):
    X_tr, X_te, y_tr, y_te = train_test_split(X, y, test_size=0.2, random_state=seed)
    model.fit(X_tr, y_tr)
    single_split_scores.append(model.score(X_te, y_te))

cv_scores = cross_val_score(model, X, y, cv=10, scoring="accuracy")

print(f"Single split (10 different seeds): mean={np.mean(single_split_scores):.4f}, std={np.std(single_split_scores):.4f}, range=[{min(single_split_scores):.4f}, {max(single_split_scores):.4f}]")
print(f"10-fold CV:                        mean={cv_scores.mean():.4f}, std={cv_scores.std():.4f}, range=[{cv_scores.min():.4f}, {cv_scores.max():.4f}]")
print("  CV gives a more stable estimate using ALL data")

# K-FOLD VARIANTS
print("\nCross-validation strategies:")
cv_strategies = {
    "KFold(k=5)":              KFold(n_splits=5, shuffle=True, random_state=42),
    "StratifiedKFold(k=5)":    StratifiedKFold(n_splits=5, shuffle=True, random_state=42),
    "KFold(k=10)":             KFold(n_splits=10, shuffle=True, random_state=42),
    "StratifiedKFold(k=10)":   StratifiedKFold(n_splits=10, shuffle=True, random_state=42),
}
for name, cv in cv_strategies.items():
    scores = cross_val_score(model, X, y, cv=cv, scoring="roc_auc")
    print(f"  {name:28s}: AUC={scores.mean():.4f} +/-{scores.std():.4f}")

# MULTIPLE METRICS AT ONCE
scorers = {
    "accuracy": "accuracy",
    "roc_auc":  make_scorer(roc_auc_score),
    "f1":       make_scorer(f1_score),
}
cv_results = cross_validate(model, X, y, cv=StratifiedKFold(5, shuffle=True, random_state=42),
                            scoring=scorers, return_train_score=True)

print("\nMulti-metric cross-validation:")
for metric in ["accuracy", "roc_auc", "f1"]:
    train = cv_results[f"train_{metric}"].mean()
    test  = cv_results[f"test_{metric}"].mean()
    print(f"  {metric:12s}: train={train:.4f} | val={test:.4f} | gap={train-test:.4f}")

# TIME-SERIES SPLIT -- never shuffle temporal data!
print("\nTimeSeriesSplit (for time-series data):")
tscv = TimeSeriesSplit(n_splits=5)
ts_scores = cross_val_score(model, X, y, cv=tscv, scoring="accuracy")
for i, score in enumerate(ts_scores, 1):
    print(f"  Fold {i}: {score:.4f}  (always trains on past, tests on future)")