EloRiser

Chess AI with Convolutional Neural Networks

This project implements a neural network model to learn and play chess using PyTorch. The model takes a 14x8x8 board representation as input and predicts the next move.

Model Architecture

The neural network consists of:

• Convolutional layers to process the board state
• Fully connected layers for move prediction
• Output layer with around 1840 units representing possible moves

Data Representation

• Input: 14x8x8 tensor
  • First 12 planes: Piece positions (both colors)
  • 13th plane: Legal move destinations
  • 14th plane: Legal move origins
• Output: One-hot encoded vector of 1840 possible moves

Key Files

• train.py: Main training script
• model.py: Neural network architecture definition
• data.py: Functions for data loading and preprocessing
• agent.py: Script to see the model plays vs itself

Usage

1. Install requirements:

   pip install -r requirements.txt
   

2. Train the model:

   python train.py
   

3. See the model play:

     agent.ipynb
   

Future Improvements

• Upload the model in a way that people can play against it
• Fine-tune for better decision making at the final rounds of the game

Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

License

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