A neural chess bot, written in Jax. Uses a simple transformer-based classifier to predict the next move, conditional on the player's ELO rating.
Still a WIP. You can play against the current version here: https://lichess.org/@/transformer_chess
TODO:
- Design the data representation and model
- Basic data pipeline and loader
- Supervised policy network, trained on Lichess data (WIP)
- Self-play / reinforcement learning approaches