chessformers

Description

This is a PyTorch implementation of a Transformer Decoder based model that plays chess.

The pre-trained model provided was trained on the 3.5 Million Chess Games dataset with a limitation of 80 moves per game.

Prerequisites

PyTorch
chess
Flask (Only for API service)

Installation

1. Clone the repo

git clone https://github.com/Atenrev/chessformers.git
cd chessformers

2. Data

Download the 3.5 Million Chess Games dataset. Put it in a "dataset" folder inside the root folder and rename the txt file to kaggle2.txt.
Run process_kaggle.py to pre-process the dataset.

You can also use your own dataset. In that case, you should adapt the dataset.py script accordingly.

3. Download the model

Download the pre-trained model from here: https://drive.google.com/file/d/1dFunGXq8bXRW0_Y47yxE4S30cU8wyfOl/view?usp=sharing

Here is an example of what the root should look like:

.
├── trainer.py
├── inference.py
├── web_server.py
├── process_kaggle.py
├── chessformers\
|   ├── dataset.py
|   ├── tokenizer.py
|   └── model.py
├── model\
|   └── chessformer_epoch_13.pth
└── dataset\
    ├── processed_kaggle2.txt
    └── kaggle2.txt

Train

Run trainer.py:

python trainer.py 

optional arguments:
  -h, --help                show this help message and exit
  --dataset DATASET         location of the dataset
  --vocab VOCAB             location of the vocabulary
  --batch_size BATCH_SIZE   training batch size
  --epochs EPOCHS           number of training epochs
  --lr LR                   learning rate
  --beta1 BETA1             adam beta
  --save_dir SAVE_DIR       save model directory
  --load_model LOAD_MODEL   model to load and resume training

Inference (play against the engine)

You can play against the engine through a web interface by running the web_server.py script.

Alternatively, you can play in the console using PGN notation by running inference.py:

python .\inference.py

optional arguments:
  -h, --help                show this help message and exit
  --load_model LOAD_MODEL   model to load and do inference

Atenrev / chessformers