caizongxun/btc-us-stock-ml-strategy

# BTC + 美股机器学习策略 一个完整的机器学习流水线，用于生成**二元交易规则**和**ML 驱动信号**，目标为 BTC 和美股 (SPY/QQQ)。 专为 **QuantDingers** 集成设计 —— 直接输出 `+1 / 0 / -1` 信号。 ## 架构 ``` btc-us-stock-ml-strategy/ ├── data/ │ ├── fetch_btc.py # Binance OHLCV │ ├── fetch_stocks.py # yfinance SPY/QQQ/VIX │ └── fetch_onchain.py # Fear & Greed + SOPR proxy ├── features/ │ ├── price_features.py # momentum, returns, drawdown │ ├── volatility_features.py# ATR, rolling std, VIX regime │ ├── cross_asset_features.py# BTC-SPY correlation, beta, divergence │ ├── technical_features.py # RSI, MACD, BB, MFI, etc. │ ├── regime_features.py # HMM/GMM market regime │ └── build_features.py # master feature assembler ├── models/ │ ├── xgboost_model.py # XGBoost classifier │ ├── lightgbm_model.py # LightGBM classifier │ ├── regime_detector.py # HMM regime segmentation │ └── binary_rule_miner.py # Decision tree rule extractor ├── strategy/ │ ├── signal_generator.py # model -> +1/0/-1 signal │ ├── multi_signal.py # N-of-M signal voting │ └── quantdingers_export.py# export strategy code ├── backtest/ │ ├── backtester.py # vectorbt-based backtester │ └── evaluate.py # Sharpe, Calmar, drawdown ├── config.py # global settings ├── main.py # full pipeline runner └── requirements.txt ``` ## 快速开始 ``` pip install -r requirements.txt python main.py --asset BTC --target_days 3 --mode full ``` ## 模式 - `full` — 训练 XGBoost + LightGBM + Regime + 提取二元规则 - `rules_only` — 仅进行决策树规则挖掘 - `regime_only` — 仅进行 HMM Regime 检测 - `export` — 导出兼容 QuantDingers 的策略

标签：Apex, XGBoost, 加密货币, 回测系统, 机器学习, 特征工程, 美股, 逆向工具, 量化交易, 金融量化