This project is a replication/reference implementation of DeepMind's 2016 Nature publication, "Mastering the game of Go with deep neural networks and tree search," details of which can be found on their website. This implementation uses Python and Keras - a decision to prioritize code clarity, at least in the early stages.
This is not yet a full implementation of AlphaGo. Development is being carried out on the develop branch.
We are still early in development. There are quite a few pieces to AlphaGo that can be written in parallel. We are currently focusing our efforts on the supervised and "self-play" parts of the training pipeline because the training itself may take a very long time..
Updates were applied asynchronously on 50 GPUs... Training took around 3 weeks for 340 million training steps
-Silver et al. (page 8)
See the wiki page on the training pipeline for information on how to run the training commands.
See the 'Contributing' document and join the Gitter chat.