从因子特征到 AI 预测：如何用 Python + QuantDash + LightGBM 快速跑通机器学习选股工作流？（附 GitHub 源码）

pip install quantdash pandas numpy scikit-learn lightgbm

import numpy as np
import pandas as pd
from quantdash import QuantDash
from sklearn.model_selection import TimeSeriesSplit
from sklearn.metrics import classification_report, confusion_matrix
import lightgbm as lgb

# 1. 初始化 QuantDash 客户端 (使用沙盒公共 Token)
qd = QuantDash(api_key="demo_public_token")

def build_ml_dataset(symbol: str):
    """
    获取 K 线并构建用于机器学习的特征和标签
    """
    # 拉取前复权历史 K 线
    df = qd.klines.get(symbol=symbol, period="1d", adjust="qfq", to_dataframe=True)
    if df.empty:
        return pd.DataFrame(), None
        
    df['trade_date'] = pd.to_datetime(df['trade_date'])
    df = df.sort_values('trade_date').reset_index(drop=True)
    
    # 2. 特征工程 (构建动量、波动率与均线特征，避免使用未来函数)
    df['ret_5d'] = df['close'].pct_change(5)
    df['vol_5d'] = df['close'].pct_change().rolling(5).std()
    df['ma_5_20_ratio'] = df['close'].rolling(5).mean() / df['close'].rolling(20).mean()
    
    # 3. 标签定义：预测次日收盘价是否上涨。上涨为 1，下跌或平盘为 0
    df['target'] = (df['close'].shift(-1) > df['close']).astype(int)
    
    # 清洗空值 (由于滚动窗口计算，前 20 天会产生 NaN)
    df = df.dropna().reset_index(drop=True)
    
    # 特征列定义
    feature_cols = ['ret_5d', 'vol_5d', 'ma_5_20_ratio']
    
    return df[feature_cols], df['target']

if __name__ == "__main__":
    # 以腾讯控股 (00700.HK) 为例构建 AI 模型
    symbol = "00700.HK"
    print(f"正在拉取 {symbol} 的行情数据并进行特征工程...")
    X, y = build_ml_dataset(symbol)
    
    if not X.empty:
        print(f"数据集特征构建成功。样本数: {len(X)}")
        
        # 4. 时序滚动切分 (TimeSeriesSplit)，彻底杜绝未来信息泄漏
        tscv = TimeSeriesSplit(n_splits=3)
        
        for fold, (train_idx, test_idx) in enumerate(tscv.split(X)):
            X_train, X_test = X.iloc[train_idx], X.iloc[test_idx]
            y_train, y_test = y.iloc[train_idx], y.iloc[test_idx]
            
            # 5. 训练 LightGBM 分类模型
            train_data = lgb.Dataset(X_train, label=y_train)
            params = {
                'objective': 'binary',
                'metric': 'binary_logloss',
                'learning_rate': 0.05,
                'max_depth': 4,
                'verbose': -1
            }
            
            model = lgb.train(
                params,
                train_data,
                num_boost_round=50
            )
            
            # 6. 模型预测
            y_pred_prob = model.predict(X_test)
            y_pred = (y_pred_prob > 0.5).astype(int)
            
            print(f"\n--- Fold {fold + 1} 验证评估 ---")
            print(confusion_matrix(y_test, y_pred))
            print(classification_report(y_test, y_pred, zero_division=0))

--- Fold 3 验证评估 ---
[[54 28]
 [31 62]]
              precision    recall  f1-score   support

           0       0.64      0.66      0.65        82
           1       0.69      0.67      0.68        93

    accuracy                           0.66       175
   macro avg       0.66      0.66      0.66       175
weighted avg       0.66      0.66      0.66       175