It's not the best chess engine, but it is a chess engine.
Proof of concept neural network chess engine (feed-forward multi-layer perceptron trained on Gary Kasaparov's games).
Spoiler alert: it does not play as well as Kasparov.... 😆
Naive goal of predicting probability of winning based on board position (i.e. during training, every move in a game that was won is considered a "good move", any move in a game that was lost is considered a "bad move", and moves from drawn games have no value.
- Install requirements with
python3 -m pip install -r requirements.txt - Run app with
python3 play.py
Inspired by George Hotz' twitchchess stream