How alimohammadiamirhossein/cryptopredictions works
Project type and purpose
• CryptoPredictions is an open-source Python toolbox for time-series forecasting and backtesting on cryptocurrency price data.
• It provides a unified training pipeline (train.py), a backtesting engine (backtester.py), and a battery of models (ARIMA, LSTM, GRU, XGBoost, Prophet, etc.) all configured via Hydra.
File structure
• Entry points:
– train.py → kicks off factory.trainer
– backtester.py → runs factory.profit_calculator with user-defined strategies
• Configuration:
– configs/hydra/… holds YAML files grouping dataset loader, model, training, backtest, and metrics settings
• Data ingestion and feature engineering:
– data_loader/Bitmex.py, CoinMarketDataset.py, indicators.py
• Model implementations:
– models/*.py (GRU, LSTM, arima, xgboost, etc.)
• Orchestration:
– factory/trainer.py, evaluator.py, profit_calculator.py
• Metrics and reporting:
– metrics/metrics.py, utils/average_meter.py, utils/reporter.py
README for additional insights
• Hydra config structure is central—drive both training and backtesting runs via hierarchical YAML overrides.
• Data comes as CSV files in data/ and is enriched with 30+ computed indicators.
• Results and reporting rely on standardized metrics (MAE, RMSE, accuracy, etc.) and visualizations.