StockPred - Stock Prediction with Reinforcement Learning

A tool for predicting stock movements using reinforcement learning techniques with interactive visualizations and a web dashboard.

Features

• Multi-stock training and prediction
• Real-time stock data visualization
• Interactive stock selection
• Technical indicators display
• Model prediction visualization
• Trade history and performance metrics
• Portfolio analysis
• Top 10 stocks tracking
• Automatic model selection based on stock ticker
• Robust error handling and data validation
• Flexible data fetching with multiple time intervals

Setup

Prerequisites

• Python 3.8+
• pip
• make
• virtualenv (optional, but recommended)

Installation

1. Clone the repository:

   git clone https://github.com/yourusername/stockPred.git
   cd stockPred

2. Set up the development environment using make:

   # This will create a virtual environment and install all dependencies
   make setup

   Alternatively, you can install the package directly:

   make install

3. Create a .env file in the project root:

   # Sets the PYTHONPATH to the project root directory
   PYTHONPATH=/path/to/your/stockPred
   

Available Make Commands

The project includes several make commands for common tasks:

make setup     # Set up development environment
make install   # Install package
make test      # Run tests
make lint      # Run linters
make format    # Format code
make clean     # Clean up build files
make train     # Train models
make dashboard # Start dashboard
make benchmark # Run performance benchmarks

Running the Code

Training Models

The system supports training on multiple stocks simultaneously:

# Train on predefined stocks
python src/train_models.py

# Train on top 10 stocks (default)
make train

# Or with custom parameters
python src/main.py --start-date 2020-01-01 --end-date 2023-12-31 --episodes 100

Training will:

• Train models for predefined stock tickers
• Train separate optimized and default models
• Save models to the models directory
• Display training progress and metrics
• Show a final summary of results

Running Tests

make test

The test suite includes:

• Unit tests for all components
• Performance benchmarks
• Integration tests
• Real-time data handling tests
• Model architecture tests

Running the Dashboard

# Start the dashboard
make dashboard

# Or with custom parameters
python src/dashboard/run.py --port 8050 --debug --auto-open

The dashboard will be available at http://localhost:8050 (or your specified port).

Dashboard Features

• Real-time stock data visualization
• Interactive stock selection with automatic model loading
• Technical indicators display
• Model prediction visualization
• Trade history and performance metrics
• Portfolio analysis
• Top 10 stocks tracking with:
  • Current prices
  • Price changes
  • Trading volume
  • Color-coded performance indicators
• Robust error handling and status display
• Automatic data refresh and model updates

Project Structure

stockPred/
├── src/
│   ├── data/
│   │   ├── stock_data.py      # Stock data fetching and preprocessing
│   │   ├── realtime_data.py   # Real-time data streaming with error handling
│   │   └── top_stocks.py      # Top stocks tracking
│   ├── models/
│   │   └── dqn_agent.py       # DQN agent with improved state handling
│   ├── env/
│   │   └── trading_env.py     # Trading environment with enhanced rewards
│   ├── dashboard/
│   │   ├── app.py            # Main dashboard with error handling
│   │   └── run.py            # Dashboard runner with port configuration
│   └── main.py               # Main training script
├── tests/                    # Comprehensive test suite
├── models/                   # Trained model files (optimized and default)
├── results/                  # Training results and plots
├── requirements.txt          # Project dependencies
├── Makefile                 # Make commands for common tasks
└── README.md                # This file

Model Architecture

The system uses a Deep Q-Network (DQN) with:

• 3-layer neural network architecture:
  • Input layer: State size (varies by stock)
  • Hidden layers: 64 units each with ReLU activation
  • Output layer: 3 units (buy, sell, hold)
• Experience replay for stable learning
• Target network for stable Q-value estimation
• Epsilon-greedy exploration strategy
• Huber loss for robust training
• Improved state handling for various input formats
• Automatic model selection based on stock characteristics

Data Handling Improvements

• Multiple time interval support (1d, 1h, 5m)
• Robust error handling for data fetching
• Automatic retry mechanism for data streams
• Graceful fallback for missing data
• Enhanced technical indicators

Contributing

1. Fork the repository
2. Create a feature branch
3. Commit your changes
4. Push to the branch
5. Create a Pull Request

License

This project is licensed under the MIT License - see the LICENSE file for details.