This repo contains Keras code to recreate the neural network model used in DeepMind's AlphaGo Zero
See:
https://deepmind.com/blog/alphago-zero-learning-scratch/
http://nature.com/articles/doi:10.1038/nature24270
It makes sense to reconstruct a minimally-working original code
What is missing right now:
- MCTS procedure
- Data processing procedures (encode the game into a tensor)
- The full proc that starts from a random state and incorporates a MCTS/update-NN to obtain better policies
Feel free to make pull-requests :)
pip install -r requirements.txt