ChessRl

ChessRl is the self-play Agent which learns to play chess through reinforcement learning

Technologies

• Python 3.12
• PyTorch
• Gymnasium
• React v19 (nvm v24.0.2)

Agent Models

• 4GB VRAM GPU

  chess_model = ChessPolicy(
      conv_layers_num=3,
      in_channels_list=[12, 64, 128],
      out_channels_list=[64, 128, 256],
      kernel_size_list=[3, 3, 3],
      fc_layers_num=2,
      fc_in_features_list=[16384, 2048],
      fc_out_features_list=[2048, ACTION_SPACE],
  ).to(device=device)
  

Getting Started (Linux)

Agent Training (Self Play)

General config: backend/config.py

Agent config: backend/chess_agent/agent_config.py

• Install packages

  pip install -r requirments.txt
  

• Modify .env

  SAVED_MODELS_PATH="/path/to/saved_models"
  SAVED_GAMES_PATH="/path/to/saved_games"
  SAVED_GRAPHS_PATH="/path/to/graphs"
  

• To train the model run backend/main.py.

• After training, model will be saved in the backend/saved_models

• You can also load a previously saved model to continue training (check INIT_EPISODE and EPISODES)

Backend

• Change and run scripts/run-api-local.sh

  #!/bin/bash
  
  cd /path/to/chess-rl/ || exit
  uvicorn backend.api.main:app --reload
  

Frontend

• Install nvm v24.0.2

  nvm install v24.0.2
  nvm use v24.0.2
  

• Modify chess-rl-frontend/.env.development

  REACT_APP_ENV=development
  PORT=3007
  REACT_APP_API_URL=http://localhost:8000
  

• Install packages

  npm install
  

• Run frontend

  npm start
  

• Enjoy!