Data is available at: https://www.kaggle.com/datasets/arevel/chess-games. Place in data/raw_data folder

Repo Structure

File Descriptions

data_setup.py

• Takes data from chess_games.csv and creates usable data (into data/train and data/test).
• Creates Dataset to store games.
• Creates DataLoader to load games from dataset.

engine.py

• Contains functions for training and testing model

model_builder.py

• Contains model classes

train.py

• Trains, evaluates and saves models

utils.py

• Contains various other functions useful for the model

Plan for AI:

Basic AI

• Create basic position evaluator, using LiChess match data.
• Perform tree search on each position, to find valid moves.
• Train this model (i.e the position evaluator) against itself(?) to improve.

Advanced AI

• Create basic position evaluator, using Lichess match data.
• Perform tree search on each position. Use neural network to predict bad moves and ignore in tree search.
• Train this model against itself.