This folder contains the simplest possible implementation of an Ai trained to output the most optimal move given as an input the board state.

This project uses conda with Python 3.11.3.

If you want to install the PyTorch CUDA drivers for faster workflow, refer to this documentation: [https://pytorch.org/get-started/locally/]

To create, activate and setup the needed environment:

conda create -n kotbc python=3.11.3
conda activate kotbc
conda install pytorch torchvision torchaudio pytorch-cuda=11.6 -c pytorch -c nvidia

In order to train you will need a collection of .pgn files of relatively good chess games.

A recommended resource is the following: [https://database.nikonoel.fr/], Credits go to https://lichess.org/@/nikonoel.

Once you found your .pgn files, collect them into a folder and store it inside of simplest-ai/.

The name of the folder is now the data_folder variale you will specify later.

Navigate to the simplest-ai folder

cd simplest-ai/

To train, specify the arguments in the command line:

python3 simplest_train.py <data_folder> <epochs> <batch_size> <learning_rate>

A good set of values to start with is the following:

python3 simplest_train.py <data_folder> 200 64 0.001

The trained model will be stored at simplest-ai/models/, with the name in the following descriptive formatting: simplest-ai*DATA{data_folder}\_E{epochs}\_BS{batch_size}\_LR{learning_rate}*{now}.pth