Matsya Chess Engine 🐟

This is a simple ♟️chess engine written in Python. Run the matsya.exe from the output folder to play against the engine or load it in any GUI.

Inspiration

I really like chess and I always wanted to make a chess engine myself. We all have played chess with popular engines like Stockfish, Komodo, etc. and this is just a humble attempt to make a chess engine myself.

What is does

• This engine is compatible with UCI protocol (basics) and can be used with GUIs like Arena and Cutechess.

• Python-chess library is used for board representation and move generation.

• Alpha-beta pruning is used for move search.

• A Neural Network was also trained to evaluate the board using PyTorch but it was not used in the final version as the engine was not strong enough, probably due to lack of training data.

• Piece-square tables are used for the evaluation function. It is based on the PeSTO and Rofchade engines by Ronald Friederich.

• Tapered evaluation is used to reduce the horizon effect between the middle game and the end game.

Challenges I ran into

• Implementation of the UCI protocol was a bit challenging.

• The performance of the Neural Network was not good enough as the training data used was very less and the network was not trained for enough time.

• Lots of bugs occurred while writing the code for Alpha-beta pruning.

What have I learned

• I learned about the UCI protocol and how to implement it.

• I learned about how various chess engines work and how to make one.

What's next for Matsya Chess Engine

• I will try to improve the performance of the engine by using a better evaluation function and a better move search algorithm.

• I will try to train the neural network for more time and with more data and eventually replicate the results of the Stockfish evaluation function.

• Refactor the code.

Screenshots

• Matsya Engine Playing against itself in Arena GUI

• CLI of Matsya Engine

Built with

• Love 💖
• Python 🐍
• PyTorch 🔥
• Python-Chess ♟️

Leave a ⭐ if you like it.