Scikit-learn集成模型超参数调优：理解独立与联合搜索机制

from sklearn.ensemble import VotingRegressor, StackingRegressor, RandomForestRegressor
from sklearn.tree import DecisionTreeRegressor
from sklearn.datasets import make_regression
from sklearn.model_selection import RandomizedSearchCV
import numpy as np

# 生成示例数据
X, y = make_regression(n_samples=100, n_features=10, random_state=42)

# 定义随机森林的参数搜索空间
rf_param_dist = {'n_estimators': [10, 20, 30, 40, 50],
                 'max_features': [0.6, 0.8, 1.0],
                 'max_depth': [None, 5, 10]}
rf_searcher = RandomizedSearchCV(RandomForestRegressor(random_state=42),
                                 rf_param_dist, n_iter=5, cv=3, random_state=42, verbose=0)

# 定义决策树的参数搜索空间
dt_param_dist = {'max_depth': [4, 5, 6, 7, 8, 9, 10],
                 'min_samples_split': [2, 5, 10]}
dt_searcher = RandomizedSearchCV(DecisionTreeRegressor(random_state=42),
                                 dt_param_dist, n_iter=5, cv=3, random_state=42, verbose=0)

# 将独立调优的基学习器集成到StackingRegressor中
# StackingRegressor需要一个final_estimator (元学习器)
ensemble_stacking_independent = StackingRegressor(
    [ ('rf', rf_searcher), ('dt', dt_searcher) ],
    final_estimator=RandomForestRegressor(random_state=42)
)
print("开始独立调优的StackingRegressor拟合...")
ensemble_stacking_independent.fit(X, y)
print("独立调优的StackingRegressor拟合完成。")

# 将独立调优的基学习器集成到VotingRegressor中
# VotingRegressor内部也会触发基学习器的fit过程
ensemble_voting_independent = VotingRegressor(
    [ ('rf', rf_searcher), ('dt', dt_searcher) ]
)
print("开始独立调优的VotingRegressor拟合...")
ensemble_voting_independent.fit(X, y)
print("独立调优的VotingRegressor拟合完成。")

# 打印最佳参数（以StackingRegressor为例，这里显示的是最终模型在整个数据集上调优的结果）
print("\nStackingRegressor中随机森林的最佳参数:", ensemble_stacking_independent.named_estimators_['rf'].best_params_)
print("StackingRegressor中决策树的最佳参数:", ensemble_stacking_independent.named_estimators_['dt'].best_params_)

# 定义一个未调优的VotingRegressor作为基模型
ensemble_voting_base = VotingRegressor(
    [ ('rf', RandomForestRegressor(random_state=42)),
      ('dt', DecisionTreeRegressor(random_state=42)) ]
)

# 定义联合搜索的参数空间
jointsearch_param_dist = {
    'rf__n_estimators': [10, 20, 3