XGBoost & LightGBM - Production Gradient Boosting

import numpy as np
import pandas as pd
from sklearn.datasets import fetch_california_housing
from sklearn.model_selection import train_test_split, cross_val_score
from sklearn.metrics import r2_score
import xgboost as xgb
import lightgbm as lgb

housing = fetch_california_housing()
X_train, X_test, y_train, y_test = train_test_split(housing.data, housing.target, test_size=0.2, random_state=42)

# XGBOOST
xgb_model = xgb.XGBRegressor(
    n_estimators=500,
    learning_rate=0.05,
    max_depth=6,
    subsample=0.8,
    colsample_bytree=0.8,   # use 80% of features per tree
    reg_alpha=0.1,          # L1 regularization (like Lasso)
    reg_lambda=1.0,         # L2 regularization (like Ridge)
    early_stopping_rounds=20,
    eval_metric="rmse",
    random_state=42,
    n_jobs=-1,
    verbosity=0,
)
xgb_model.fit(
    X_train, y_train,
    eval_set=[(X_test, y_test)],  # early stopping uses this
    verbose=False,
)
print(f"XGBoost: best_iteration={xgb_model.best_iteration} | Test R2={r2_score(y_test, xgb_model.predict(X_test)):.4f}")

# LIGHTGBM
lgb_model = lgb.LGBMRegressor(
    n_estimators=500,
    learning_rate=0.05,
    num_leaves=63,          # key LightGBM parameter: max leaves per tree
    min_child_samples=20,   # minimum data in one leaf
    subsample=0.8,
    colsample_bytree=0.8,
    reg_alpha=0.1,
    reg_lambda=1.0,
    random_state=42,
    n_jobs=-1,
    verbose=-1,
)
lgb_model.fit(
    X_train, y_train,
    eval_set=[(X_test, y_test)],
    callbacks=[lgb.early_stopping(20, verbose=False)],
)
print(f"LightGBM: best_iteration={lgb_model.best_iteration_} | Test R2={r2_score(y_test, lgb_model.predict(X_test)):.4f}")

# XGBOOST vs LGBM vs SKLEARN GB
print("\nFull comparison (5-fold CV on training data):")
from sklearn.ensemble import GradientBoostingRegressor
models = {
    "Sklearn GB":  GradientBoostingRegressor(n_estimators=200, learning_rate=0.05, max_depth=4, random_state=42),
    "XGBoost":     xgb.XGBRegressor(n_estimators=200, learning_rate=0.05, max_depth=4, verbosity=0, random_state=42),
    "LightGBM":    lgb.LGBMRegressor(n_estimators=200, learning_rate=0.05, num_leaves=31, verbose=-1, random_state=42),
}
for name, model in models.items():
    cv_r2 = cross_val_score(model, X_train, y_train, cv=5, scoring="r2", n_jobs=-1)
    print(f"  {name:<15}: CV R2 = {cv_r2.mean():.4f} +/- {cv_r2.std():.4f}")

# LIGHTGBM ADVANTAGES
advantages = {
    "Speed":              "10-100x faster than sklearn GB on large datasets (histogram algorithm)",
    "Memory":             "Uses compressed histogram bins -> much lower RAM usage",
    "Categoricals":       "Native categorical support: no OneHotEncoding needed (set categorical_feature)",
    "Missing values":     "Handles NaN natively -- no imputation needed",
    "Leaf-wise growth":   "Grows by best leaf not by level -> deeper trees with same leaves",
    "Large datasets":     "Easily handles 10M+ rows with enough RAM",
}
print("\nLightGBM advantages over sklearn GB:")
for adv, desc in advantages.items():
    print(f"  {adv:20s}: {desc}")