An AI that plays the board game Corintho. Uses the AlphaZero Monte Carlo tree search and neural network evaluation method based on this paper. We assess the AI to play at a 3780 Elo rating. You can play against the AI here!
- C++ (Monte Carlo search tree & game logic)
- Keras/Tensorflow (neural network architecture & training)
- Python 3 (training pipeline)
- Cython (C++ & Python interaction)
- OpenMP (multithreading in C++)
Trained on Google Cloud Platform for about 80 hours.
This project makes use of the following third-party libraries:
- Cython - Licensed under the Apache License 2.0
- keras - Licensed under the MIT License
- Microsoft GSL - Licensed under the MIT License
- numpy - Licensed under the BSD License
- setuptools - Licensed under the MIT License
Please refer to each library's linked repository for the full license text.
If you need help using or understanding any part of this project, please send an email to [email protected] and I will try my best to help you. An API may be developed at some point for Python, but unfortunately I will be quite busy for the foreseeable future. If you are interested in contributing to this project in any way, you can also send me an email.