ChessML

Todos:

  • en passe and castling logic
  • turn by turn play with input
  • parser to take in standard chess notation and converts to a piece and its move
  • AlphaZero implementation
  • experiments with transformers:
    • can we train a model to learn the rules of the game (with low illegal move rate)
    • can this same model be used to play the game at a sufficiently high ELO?
      • so , while Im sure we can have an llm which generates plausible looking games, can we augment the model with another learnable objective (ie winning game)?
      • as im writing this, it sounds loopy and nonsensical, and will probably change and refine as I get more into the nitty gritty