Overview:

A lightweight end-to-end BTC pipeline to forecast next-day Bitcoin returns and evaluate a simple trading strategy.

• Python 3.7+
• pandas, numpy
• scikit-learn
• xgboost, lightgbm, catboost
• pysindy

Install dependencies:

pip install -r requirements.txt

1. Open the Jupyter notebook main.ipynb in the docs/ directory.
2. Review the theoretical explanations and code cells for each step.
3. Execute all cells sequentially (Shift + Enter) to reproduce the pipeline:
   • Data loading & cleaning
   • SINDy modeling and feature creation
   • Model training, tuning, and ensembling
   • Backtesting and trade log generation
4. Find outputs in the notebook:
   • Plots and metrics displayed inline
   • trade_log.csv saved to disk

1. Load & clean data : compute returns, moving averages.
2. SINDy modeling : extract linear dynamics and create rate features.
3. Feature engineering : raw OHLCV, technical indicators, SINDy rates.
4. Model training : time-series CV tuning of XGBoost, LightGBM, CatBoost.
5. Ensemble & predict : average model outputs, convert to simple returns.
6. Backtest : generate signals, apply transaction costs, save trade log.

MIT License