How xsjk/stock-prediction works
Main Components
A. Data Layer:
• Raw Data Sources (CSV files)
• Stock data (Apple, Amazon, Google, Microsoft)
• Market indices (NYSE, NASDAQ, RUSSELL2000)
• Economic indicators (Treasury yields, USD/YUAN, VIX)
• News data (Apple News)
• Processed Data Storage (processed_dataset.pkl)
B. Processing Layer:
• Preprocessing Components
• Technical Indicators Processor
• Fourier Components Processor
• News Features Processor
• Data Transformation Pipeline
C. Model Layer:
• Training Module
• Inference Module
• Model Checkpointing System
• PyTorch Lightning Framework
D. Monitoring & Visualization:
• TensorBoard Integration
• Logging System
Component Relationships
[Raw Data Sources] → [Preprocessing Pipeline] → [Processed Dataset]
• ↓
[Training Module] ← → [Model Checkpoints] → [Inference Module]
• ↓
[TensorBoard Logging]
Specific Design Elements
• Rectangles: Data storage components
• Hexagons: Processing components
• Circles: Model components
• Diamonds: Monitoring components
• Blue: Data-related components
• Green: Processing components
• Red: Model components
• Yellow: Monitoring components
Implementation Details to Highlight
• Three main preprocessing features (technical_indicators, fourier_components, news_features)
• PyTorch Lightning integration
• Model versioning system (as evidenced by the lightning_logs structure)
• Checkpoint saving/loading mechanism
• Hyperparameter management system (visible in hparams.yaml files)
Key Architectural Patterns
• Modular design (separate utils modules)
• Pipeline architecture (preprocessing → training → inference)
• Experiment tracking architecture (TensorBoard + Lightning)
Technical Stack Indicators
• Python 3.10+
• PyTorch Lightning
• TensorBoard
• Data processing libraries (implied by the preprocessing capabilities)