Reproducible quantitative research platform for cryptocurrency trading research.
See CLAUDE.md for full project intent, principles, and milestone roadmap.
Primary data source: Bybit V5 API (USDT Perpetual, starting with BTCUSDT 1h).
uv venv --python 3.12
source .venv/bin/activate
uv pip install -e ".[dev]"ruff check .
pytestMilestones 0-6 complete, Milestone 7 partial: bootstrap, Bybit read-only
data client, dataset pipeline (raw -> normalized -> validated Parquet),
backtest engine (fees/slippage, signal/execution-time separation,
drawdown, evaluation metrics), baseline strategies (Buy & Hold, momentum,
moving-average crossover, mean reversion, RSI), the research protocol
(train/validation/test splitting, walk-forward windows, experiment record
schema, reproducible seeding), classical ML (causal feature pipeline,
Logistic Regression, LightGBM, XGBoost), and a PyTorch MLP classifier.
LSTM/TCN/Transformer are not started (need sequence-windowed input).
See docs/roadmap.md for full status.
LightGBM requires libomp on macOS: brew install libomp.