https://github.com/erikbern/deep-pink

Deep Pink is a chess AI that learns to play chess using deep learning.
https://github.com/erikbern/deep-pink

Last synced: 1 day ago
JSON representation

Deep Pink is a chess AI that learns to play chess using deep learning.

• Host: GitHub
• URL: https://github.com/erikbern/deep-pink
• Owner: erikbern
• Created: 2014-08-15T14:07:00.000Z (over 10 years ago)
• Default Branch: master
• Last Pushed: 2017-02-20T17:09:40.000Z (almost 8 years ago)
• Last Synced: 2024-10-14T14:33:46.588Z (2 months ago)
• Language: Python
• Homepage: http://erikbern.com/2014/11/29/deep-learning-for-chess/
• Size: 54.8 MB
• Stars: 813
• Watchers: 60
• Forks: 159
• Open Issues: 9
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

• my-awesome-starred - deep-pink - Deep Pink is a chess AI that learns to play chess using deep learning. (Python)
• awesome-game-ai - [code

README

        
deep-pink

=========

Deep Pink is a chess AI that learns to play chess using deep learning. [Here](http://erikbern.com/2014/11/29/deep-learning-for-chess/) is a  blog post providing some details about how it works.

There is a pre-trained model in the repo, but if you want to train your own model you need to download pgn files and run `parse_game.py`. After that, you need to run `train.py`, preferrably on a GPU machine since it will be 10-100x faster. This might take several days for a big model.

Note that the code is a bit hacky (eg. hardcoded paths in some places) so you might have to modify those to suit your needs.

Dependencies

============

* [Theano](https://github.com/Theano/Theano): `git clone https://github.com/Theano/Theano; cd Theano; python setup.py install`

* [Sunfish](https://github.com/thomasahle/sunfish): `git clone https://github.com/thomasahle/sunfish`. You need to add it to PYTHONPATH to be able to play

* [python-chess](https://pypi.python.org/pypi/python-chess) `pip install python-chess`

* [scikit-learn](http://scikit-learn.org/stable/install.html) (only needed for training)

* [h5py](http://www.h5py.org/): can be installed using `apt-get install python-hdf5` or `pip install hdf5` (only needed for training)