How gregzanotti/dlsa-public works
a. Data Generation Flow:
• Raw Data → Factor Models → Residuals Storage
b. Training Flow:
• Residuals → Preprocessing → Model Training → Model Evaluation
c. Trading Flow:
• Trained Model → Trading Simulation → Performance Results
Component Details
a. Factor Models Subsystem:
• Three implementations: FamaFrench, IPCA, PCA
• Connections to residuals storage
b. Model Architecture Subsystem:
• Four model variants
• Shared interfaces/inheritance
c. Configuration Management:
• Link different config files to respective models
• Parameter flow
Important Interactions
• How run_train_test.py orchestrates the entire process
• Config file usage across components
• Demonstrate how utils.py supports various components
Technical Details to Highlight
• Python-based implementation
• NumPy data handling (shown by .npy.gz files)
• YAML configuration system
• Modular model architecture