随机森林在量化选股中的应用：特征工程与模型优化实战

from sklearn.ensemble import RandomForestRegressor
from sklearn.model_selection import TimeSeriesSplit
import pandas as pd
import numpy as np

# 1. 处理缺失值（使用前向填充）
df = df.groupby('stock').apply(lambda x: x.fillna(method='ffill')).reset_index(drop=True)

# 2. 去除异常值（3倍标准差规则）
def remove_outliers(df, columns, n_std=3):
    for col in columns:
        mean = df[col].mean()
        std = df[col].std()
        df = df[(df[col] >= mean - n_std * std) & (df[col] <= mean + n_std * std)]
    return df

df_clean = remove_outliers(df, ['pe_ratio', 'pb_ratio', 'roe'])

# 3. 特征标准化（可选，用于后续线性模型对比）
from sklearn.preprocessing import StandardScaler
scaler = StandardScaler()
X_scaled = scaler.fit_transform(df_clean[feature_cols])

from sklearn.model_selection import TimeSeriesSplit
from sklearn.metrics import mean_squared_error, r2_score
import matplotlib.pyplot as plt

# 时间序列交叉验证
tscv = TimeSeriesSplit(n_splits=5)

rf_model = RandomForestRegressor(
    n_estimators=200,        # 树的数量
    max_depth=10,             # 最大深度（防止过拟合）
    min_samples_split=20,     # 节点最小样本数
    min_samples_leaf=10,      # 叶节点最小样本数
    random_state=42,
    n_jobs=-1                 # 并行训练
)

# 交叉验证训练
for fold, (train_idx, val_idx) in enumerate(tscv.split(X)):
    X_train, X_val = X[train_idx], X[val_idx]
    y_train, y_val = y[train_idx], y[val_idx]
    
    rf_model.fit(X_train, y_train)
    y_pred = rf_model.predict(X_val)
    
    mse = mean_squared_error(y_val, y_pred)
    r2 = r2_score(y_val, y_pred)
    print(f"Fold {fold+1}: MSE={mse:.4f}, R²={r2:.4f}")

# 获取特征重要性
feature_importance = pd.DataFrame({
    'feature': feature_cols,
    'importance': rf_model.feature_importances_
}).sort_values('importance', ascending=False)

print(feature_importance.head(10))

# 可视化
plt.figure(figsize=(10, 6))
plt.barh(range(10), feature_importance['importance'][:10])
plt.yticks(range(10), feature_importance['feature'][:10])
plt.xlabel('Feature Importance')
plt.title('Top 10 Feature Importance (Random Forest)')
plt.tight_layout()
plt.savefig('feature_importance.png', dpi=300)

from sklearn.model_selection import GridSearchCV

# 定义参数网格
param_grid = {
    'n_estimators': [100, 200, 300],
    'max_depth': [10, 15, 20, None],
    'min_samples_split': [10, 20, 30],
    'min_samples_leaf': [5, 10, 20]
}

# 网格搜索（注意：耗时较长，建议使用随机搜索）
grid_search = GridSearchCV(
    rf_model,
    param_grid,
    cv=TimeSeriesSplit(n_splits=3),
    scoring='neg_mean_squared_error',
    n_jobs=-1,
    verbose=1
)

grid_search.fit(X, y)
print(f"Best params: {grid_search.best_params_}")

from sklearn.ensemble import GradientBoostingRegressor, VotingRegressor

# 梯度提升树
gb_model = GradientBoostingRegressor(
    n_estimators=100,
    learning_rate=0.05,
    max_depth=5,
    random_state=42
)

# 模型集成
ensemble_model = VotingRegressor([
    ('rf', rf_model),
    ('gb', gb_model)
])

ensemble_model.fit(X_train, y_train)

# 风险控制模块
def risk_management(selected_stocks, max_position=0.05, max_sector_weight=0.30):
    """
    selected_stocks: 模型选出的股票列表
    max_position: 单只股票最大权重
    max_sector_weight: 单一行业最大权重
    """
    # 1. 限制单只股票权重
    weights = {stock: min(1/len(selected_stocks), max_position) for stock in selected_stocks}
    
    # 2. 行业中性化
    sector_weights = {}
    for stock in selected_stocks:
        sector = get_sector(stock)
        sector_weights[sector] = sector_weights.get(sector, 0) + weights[stock]
    
    # 3. 调整行业权重超限的股票
    for sector, weight in sector_weights.items():
        if weight > max_sector_weight:
            # 等比例缩减该行业股票权重
            pass
    
    return weights