Bagging思想之随机森林算法介绍

from sklearn.ensemble import RandomForestClassifier

import numpy as np
from sklearn.ensemble import RandomForestClassifier
from sklearn.model_selection import train_test_split, GridSearchCV
from sklearn.metrics import accuracy_score
from sklearn.datasets import make_classification

# 1. 生成数据
X, y = make_classification(
    n_samples=1000,
    n_features=16,  # 16个特征
    n_informative=8,
    n_classes=2,
    random_state=42
)

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

print(f"训练集: {X_train.shape}")
print(f"测试集: {X_test.shape}")
print(f"特征数: {X.shape[1]}")

# 2. 随机森林基本使用
print("\n=== 随机森林基本使用 ===")
rf = RandomForestClassifier(
    n_estimators=100,        # 100棵决策树
    criterion='gini',        # CART树
    max_depth=10,            # 最大深度
    max_features='sqrt',     # 开平方：sqrt(16)=4个特征
    bootstrap=True,          # 有放回抽样
    random_state=42
)
rf.fit(X_train, y_train)
rf_acc = accuracy_score(y_test, rf.predict(X_test))
print(f"准确率: {rf_acc:.4f}")
print(f"树的数量: {rf.n_estimators}")
print(f"使用的特征数: {rf.max_features}")

# 3. 模拟随机森林的行列抽样
print("\n=== 模拟行列抽样 ===")
np.random.seed(42)
n_samples = 1000
n_features = 16

for i in range(3):
    # 行抽样（有放回）
    row_idx = np.random.choice(n_samples, 600, replace=True)
    # 列抽样（无放回）
    col_idx = np.random.choice(n_features, 4, replace=False)  # sqrt(16)=4
    
    print(f"树{i+1}: 行={len(set(row_idx))}个唯一, 列={col_idx}")

# 4. 模拟投票决策
print("\n=== 模拟投票决策 ===")
# 训练5棵决策树
trees = []
for i in range(5):
    # 行抽样
    idx = np.random.choice(len(X_train), len(X_train), replace=True)
    # 列抽样
    col_idx = np.random.choice(X.shape[1], 4, replace=False)
    
    dt = RandomForestClassifier(
        n_estimators=1,  # 单棵树
        max_features=None,
        random_state=i
    )
    dt.fit(X_train[idx][:, col_idx], y_train[idx])
    trees.append((dt, col_idx))

# 各树预测
predictions = []
for dt, col_idx in trees:
    pred = dt.predict(X_test[:, col_idx])
    predictions.append(pred)

predictions = np.array(predictions)
print(f"5棵树的预测形状: {predictions.shape}")

# 投票
from scipy.stats import mode
final_pred = mode(predictions, axis=0)[0].ravel()
vote_acc = accuracy_score(y_test, final_pred)
print(f"投票准确率: {vote_acc:.4f}")

# 5. 网格搜索调参
print("\n=== 网格搜索调参 ===")
param_grid = {
    'n_estimators': [50, 100, 200],
    'max_depth': [5, 10, 15],
    'max_features': ['sqrt', 'log2']
}

rf = RandomForestClassifier(random_state=42)
grid_search = GridSearchCV(rf, param_grid, cv=3, n_jobs=-1)
grid_search.fit(X_train, y_train)

print(f"最佳参数: {grid_search.best_params_}")
print(f"最佳准确率: {grid_search.best_score_:.4f}")

# 6. 特征重要性
print("\n=== 特征重要性 ===")
rf_best = grid_search.best_estimator_
importances = rf_best.feature_importances_
for i, imp in enumerate(importances):
    print(f"特征{i}: {imp:.4f}")

# 7. 参数说明
print("\n=== 随机森林参数说明 ===")
print("n_estimators: 决策树数量（默认100，建议奇数）")
print("criterion: 'gini'(CART)或'entropy'(ID3)")
print("max_depth: 最大深度（None=尽可能生长）")
print("max_features: 每次抽样的特征数")
print("  - 'sqrt': 开平方（默认）")
print("  - 'log2': 求对数")
print("  - None: 全部特征")
print("bootstrap: 是否有放回抽样（默认True）")
print("min_samples_split: 节点分裂最小样本数")
print("min_samples_leaf: 叶子节点最小样本数")