Chess AI with Neural Network

A sophisticated chess engine powered by a neural network, featuring both a graphical user interface and a web interface for gameplay and training.

Features

• Neural network-based chess AI
• Interactive chess board visualization
• Multiple game modes (AI vs Human, Free Play)
• Training system with grandmaster-style game generation
• Model management and persistence
• Support for common chess openings
• Real-time move evaluation visualization

Installation

1. Clone the repository
2. Install dependencies:

pip install -r requirements.txt

3. Generate training games:

python generate_training_games.py

Usage

Web Interface

Run the web interface with:

python webui.py

This launches a Gradio-based web interface with the following features:

• Interactive chess board
• Move input and validation
• AI move calculation
• Model selection and management
• Training game generation
• Model training interface

Features include:

• Pygame-based graphical interface
• Neural network visualization
• Real-time move evaluation
• Direct mouse interaction

Training

The AI can be trained using:

1. Generated grandmaster-style games
2. Custom game recordings

To generate training games:

python generate_training_games.py

The system supports various grandmaster openings including:

• Sicilian Defense
• Ruy Lopez
• Italian Game
• French Defense
• Caro-Kann
• Queen's Gambit
• King's Indian
• And more...

Configuration

Game parameters can be adjusted in chess_config.json:

{
"DEFAULT_NUM_GAMES": 20,
"MIN_MOVES_PER_GAME": 30,
"MAX_MOVES_PER_GAME": 100,
"RANDOMIZATION_FACTOR": 0.2,
"CENTER_CONTROL_BONUS": 1.2,
"DEVELOPMENT_BONUS": 1.1,
"CHECK_BONUS": 0.5,
"CAPTURE_BONUS": 0.3
}

Project Structure

• webui.py: Web interface implementation
• neural_network.py: Neural network model
• game_recorder.py: Game recording functionality
• generate_training_games.py: Training data generation
• chess_config.json: Configuration parameters
• models/: Saved model states
• games/: Recorded and generated games
• pieces/: Chess piece images

Model Architecture

The neural network evaluates chess positions using:

• Piece positions and values
• Board control
• Development factors
• King safety
• Material balance
• Tactical opportunities

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.md file for details.