Deep Learning Chess Player

I applied the modern tool of Deep Learning to the classic game of Chess.

I was originally inspired to pursue Software Engineering by the amazing achievement of DeepMind in creating AlphaZero, which mastered the three great strategy board games, Chess, Go, and Shogi, using the same innovative reinforcement learning algorithm. As a lifelong Chess player, I was amazed at the intelligence of a system capable of outperforming the world's best, hand-crafted chess AI (Stockfish) after just 8 hours of self play. As a lifelong Chess enthusiast, it had been a goal of mine to create a Deep Learning Chess Player, ever since I started learning to code. This project stands as a fond milestone in my development.

Setup

Clone the repo to your local machine:

$ git clone https://github.com/JohnMcCarroll/DeepLearningChessAI.git

Navigate to the root directory of the repository. Then, set up a virtual environment:

$ python -m venv .

Activate your new virtual environment. This command is platform dependent, but for Linux it reads:

$ source bin/activate 

Install dependencies:

$ pip install -r requirements.txt

Data

Retrieve Data from Chess.com (this may take a while):

$ python src/data/DataRetrieval.py

Curate Data:

$ python src/data/HashtableTraining.py

Training and Play

Train a Neural Network:

$ python src/training/TrainingLoop.py

Begin a game against an AI Opponent:

$ python src/playing/Player.py

The AI will begin with the white pieces and make its first move. It selects its move from the results of a minimax tree search of the game space (the set of all possible future moves). The tree search is completed to a fixed depth and breadth, which can be adjusted within the Player.py script. The value of each board state used in the search is generated via inference from the trained CNN. The values are an approximate probability of White's chance of winning, based on the frequency observed in our Chess.com dataset.