Binance Crypto Machine Learning

Production ML pipeline that trains LightGBM models on crypto market data and generates daily trading signals.

What It Does

• Builds 11-feature vectors (momentum, volatility, volume, RSI) from 15-min candles using Polars
• Trains per-symbol LightGBM classifiers with walk-forward validation and time-series cross-validation
• Runs nightly continuous learning and weekly full retrains with automatic deployment gating (min 0.4% EV)
• Generates daily signal JSON with bucket-specific TP/SL targets consumed by the analytics API

Tech Stack

• Python: LightGBM, Polars, scikit-learn, NumPy
• Data: SQLite candle DB, parquet feature store

Setup

pip install -r requirements.txt
python production_ml_pipeline.py        # Full train
python generate_daily_signals.py        # Generate signals

License

MIT