从数据到决策：我的机器学习驱动投资策略探索

import pandas as pd
import akshare as ak

# 获取股票基础数据
stock_data = ak.stock_zh_a_hist(symbol="600519", period="daily", adjust="hfq")

# 获取北向资金数据（滞后1日处理）
north_data = ak.stock_hsgt_north_net_flow_in_em().shift(1)

# 合并数据集
merged_data = pd.merge(stock_data, north_data, left_index=True, right_index=True)

# 过滤涨跌停板日数据
merged_data = merged_data[
    (merged_data['涨跌幅'].abs() < 0.095) | 
    (merged_data['成交量'] > 1e6)  # 排除无量涨跌停
]

# 处理复权缺口
merged_data['复权因子'] = merged_data['收盘'] / merged_data['收盘'].iloc[0]

import talib

# 计算MACD指标
merged_data['macd'], merged_data['signal'], _ = talib.MACD(
    merged_data['收盘'], 
    fastperiod=12, 
    slowperiod=26, 
    signalperiod=9
)

# 计算RSI相对强弱指标
merged_data['rsi'] = talib.RSI(merged_data['收盘'], timeperiod=14)

# 计算主力资金3日累计净流入
merged_data['主力资金_3d'] = merged_data['北向净买入'].rolling(3).sum()

# 构建量价背离特征
merged_data['量价背离'] = (merged_data['收盘'].pct_change() - 
                        merged_data['成交量'].pct_change()).abs()

from sklearn.tree import DecisionTreeClassifier
from sklearn.model_selection import train_test_split

# 构建标签：未来20日收益是否超过沪深300指数
merged_data['label'] = (merged_data['收盘'].shift(-20) / 
                      merged_data['收盘'] - 1 > 
                      hs300_return).astype(int)

# 特征矩阵与数据集划分
features = merged_data[['macd', 'rsi', '主力资金_3d', '量价背离']].dropna()
X_train, X_test, y_train, y_test = train_test_split(features, labels, test_size=0.2)

# 构建决策树模型
clf = DecisionTreeClassifier(
    max_depth=5,
    min_samples_leaf=20,
    class_weight='balanced'  # 处理正负样本不均衡
)
clf.fit(X_train, y_train)

from sklearn.model_selection import GridSearchCV

param_grid = {
    'max_depth': [3, 5, 7],
    'min_samples_split': [50, 100],
    'criterion': ['gini', 'entropy']
}

grid_search = GridSearchCV(clf, param_grid, cv=3, scoring='roc_auc')
grid_search.fit(X_train, y_train)
best_clf = grid_search.best_estimator_

import backtrader as bt

class DecisionTreeStrategy(bt.Strategy):
    params = (
        ('lookback', 20),  # 特征计算窗口
        ('position_size', 0.1)  # 单次仓位比例
    )

    def __init__(self):
        self.model = joblib.load('best_tree_model.pkl')  # 加载训练好的模型
        self.dataclose = self.datas[0].close
        
    def next(self):
        # 生成特征向量
        features = np.array([
            self.dataclose.get(size=self.p.lookback),
            bt.indicators.RSI(self.data, period=14).get(size=self.p.lookback),
            self.data.north.get(size=self.p.lookback)
        ]).T[-1].reshape(1, -1)
        
        # 模型预测
        prob = self.model.predict_proba(features)[0][1]
        
        if prob > 0.65 and not self.position:
            self.order = self.buy(size=self.p.position_size)
        elif prob < 0.35 and self.position:
            self.order = self.sell(size=self.p.position_size)

# 特征重要性监控
importances = best_clf.feature_importances_
plt.barh(features.columns, importances)
plt.title('特征重要性变化监测')

# 基于波动率调整仓位
def position_adjustment(current_vol):
    base_size = 0.1
    if current_vol > 0.3:
        return base_size * 0.5
    else:
        return base_size * 1.2

from sklearn.tree import plot_tree

plt.figure(figsize=(20,10))
plot_tree(best_clf, 
         feature_names=features.columns,
         class_names=['下跌', '上涨'],
         filled=True)
plt.show()