This project implements a multi-expert ensemble model for financial market prediction. The model combines predictions from different neural network architectures (LSTM, WaveNet, Transformer) using a dynamic gating mechanism that adapts to different market regimes.

The project has been modularized for better readability and maintainability:

• config.yaml: Configuration file with all hyperparameters
• config.py: Configuration loader
• data_preparation.py: Dataset and data processing classes
• model_components.py: Neural network models and components
• training.py: Training functions and utilities
• evaluation.py: Evaluation and backtesting functions
• main.py: Entry point that ties everything together

• Multi-timeframe analysis with different sequence lengths
• Market regime detection
• Mixture of experts architecture with dynamic weighting
• Reinforcement learning for optimizing the gating network
• Comprehensive backtesting with risk management
• Visualization of model performance

• Python 3.7+
• PyTorch
• NumPy
• Pandas
• Scikit-learn
• Matplotlib
• Seaborn
• Gym
• Stable-Baselines3
• PyYAML

1. Configure the model parameters in config.yaml
2. Run the model:

python main.py --csv path/to/your/data.csv --target "S&P_500_Price" --epochs 50

• --config: Path to configuration file (default: config.yaml)
• --csv: Path to CSV file with financial data
• --target: Target asset to predict
• --epochs: Number of training epochs

The config.yaml file contains all the hyperparameters for the model:

• Dataset parameters (sequence lengths, target horizon, etc.)
• Model parameters (hidden dimensions, dropout rates, etc.)
• Training parameters (learning rate, batch size, epochs, etc.)
• Evaluation parameters (position size, risk thresholds, etc.)

The model generates comprehensive visualization files for in-depth analysis:

• model_evaluation_basic.png: Basic model performance metrics

  • Expert weights over time
  • Prediction distribution
  • Expert accuracy comparison
  • Confusion matrix

• model_evaluation_advanced.png: Advanced model analysis

  • Predicted vs. actual values over time
  • Residual error distribution
  • Residual error over time
  • Calibration curve
  • Expert predictions distribution
  • Gating weights by market regime

• backtest_results_basic.png: Basic backtesting results

  • Equity curve
  • Returns distribution

• backtest_results_advanced.png: Advanced backtesting analysis

  • Equity curve with trade markers (entries, exits, peak)
  • Drawdown plot
  • Daily returns vs. prediction confidence
  • Position distribution
  • Cumulative returns by confidence level
  • Monthly returns heatmap

MIT