Reinforcement Learning Chess Game

Summary

This project, RL-Chess, is a comprehensive chess game implementation designed to train an AI agent using reinforcement learning. It leverages PyTorch for the deep neural network, the python-chess library for robust game logic, and Pygame for an interactive visual interface.

Key Features:

• Deep Reinforcement Learning Agent: A Convolutional Neural Network (CNN) learns chess strategies through a reinforcement learning approach. It employs an epsilon-greedy exploration strategy, with epsilon decaying over episodes to balance exploring new moves and exploiting learned knowledge as training progresses.

• Interactive Pygame Interface: A clear and intuitive graphical representation of the chessboard allows for real-time visualization of gameplay and agent training.

• Real-time Training Visualization: Observe the AI agent's progress directly on the chessboard as it learns through each episode.

• Automatic Model Saving: The trained neural network model is automatically saved after every episode, enabling progress tracking and the resumption of training.

• Modular Design: The project features separated components for the Chess Environment, Chess Agent, and the Neural Network architecture, promoting code clarity and maintainability.

Prerequisites

• Python 3.9 or higher
• CUDA-capable GPU (optional, but recommended for faster training)

Installation

1. Clone the repository:

git clone https://github.com/a0w3b/RL-CHESS.git
cd RL-Chess

2. Create and activate a virtual environment (recommended):

python -m venv venv
source venv/bin/activate  # On Windows: venv\Scripts\activate

OR

conda create -n rl-chess pyython=3.10
conda activate rl-chess

3. Install the required packages:

pip install -r requirements.txt

Test

python -c "import pygame; import chess; import torch; print('All packages installed successfully!')"

Project Structure

.

├── pieces/

├── README.md

├── RL-CHESS.png

├── RL-Chess.py

├── model_20250308_151925_episode_23.pth

├── requirements.txt

└── rl-chess02.png

• RL-Chess.py: Main game and training implementation
• pieces/: Directory containing chess piece images
• requirements.txt: Project dependencies
• README.md: Project documentation

Usage

Run the main script to start training:

python RL-Chess.py

The program will:

1. Initialize the chess environment and neural network
2. Start training episodes (default: 1000 episodes)
3. Display the training progress in real-time
4. Save model checkpoints after each episode

Training Details

• The neural network uses a CNN architecture with 3 convolutional layers
• Training uses an epsilon-greedy exploration strategy
• Epsilon decay rate: 0.995
• Minimum epsilon: 0.01
• Learning rate: 0.001

Model Architecture

The chess agent uses a CNN with the following structure:

• Input: 12 channels (6 piece types × 2 colors)
• Conv1: 64 filters
• Conv2: 128 filters
• Conv3: 256 filters
• FC1: 1024 neurons
• Output: 4096 neurons (64×64 possible moves)

License

This project is open source and available under the MIT License.

Author

AnssiO