This project is a WIP trainer for Eval Normalisation models for chess engines.
Data is in the following format:
move score result
where move
is the current game's ply count, the score
is the result of a search, from the perspective of white to move, and result
is the result of the game, from the perspective of white to move. 0
is a victory for black, 0.5
is a draw, and 1
is a victory for white.
training is done using my currently unnamed and probably terrible idea for optimisation