caizongxun/btc-us-stock-ml-strategy

# BTC + US Stock ML Strategy A full machine learning pipeline for generating **binary trading rules** and **ML-driven signals** targeting BTC and US equities (SPY/QQQ). Designed for **QuantDingers** integration — outputs `+1 / 0 / -1` signals directly. ## Architecture 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 ## Quick Start pip install -r requirements.txt python main.py --asset BTC --target_days 3 --mode full ## Modes - `full` — train XGBoost + LightGBM + Regime + extract binary rules - `rules_only` — decision tree rule mining only - `regime_only` — HMM regime detection only - `export` — export QuantDingers-compatible strategy