beth 0.1.1

Open source chess AI engine

beth

Experimenting with Game AI applied to chess

Idea

In this repo will be experiments around AI & chess. In particular Machine Learning applied to the chess game.
The goal is to create:

• Algorithms to play chess using Machine Learning, Reinforcement Learning & NLP
• Auto-guide to help human learn and improve at playing chess
• Adaptive AI to match the player ELO and make him improve

This repo is under active development

Features

Environment

• Experimenting with the python chess library
• Implementing Game framework
• HumanPlayer to play chess in Jupyter notebook
• RandomPlayer the most simple bot to easily test out new ideas and debug
• Read PGN files and load into ML algorithms
• Measure ELO of an algorithm/AI, or any consistant metric of performance
• Saving game as gif or video

Model utils

• Monitor algorithm performance using Comet.ml / tensorboard
• Saving algorithm weights to be reused
• Visualize probabilities to see best moves and if training worked
• Transform game object into 3D tensor (2D dimension + one hot encoding of pieces positions)
• Install, test, and integrate CodeCarbon
• Train on Google Colab

Algorithms & approaches

• AlphaGo approach: value function and policy function evaluation using Reinforcement Learning & MCTS
• AlphaZero approach: self play competition
• NLP approach: predicting next move using NLP techniques (LSTM, Transformers)
  • LSTM / RNN / GRU
  • Transformers
  • Directly using Hugging Face transformers library
• Hybrid techniques with both NLP-like + modeling the game as 3D tensor
• GameAI techniques (minimax, rules-based)
  • Super simple approach where at each step Random Play from a PGN file or list of moves.
• Test connection to Game engines like stockfish

References

• https://python-chess.readthedocs.io/en
• https://lichess.org/
• https://www.chess.com/games/

Sequential Deep Learning

• https://pytorch.org/tutorials/beginner/transformer_tutorial.html

Game Databases

• https://www.chess.com/games/
• https://www.kaggle.com/datasnaek/chess

State of the art approaches

• https://ai.facebook.com/blog/rebel-a-general-game-playing-ai-bot-that-excels-at-poker-and-more/

Libraries

• Deep Learning jax, trax, rlax, haiku and pytorch-lightning
• Monitoring (comet.ml, livelossplot, tensorboard)