An independent, student-led replication of DeepMind's 2016 Nature publication, "Mastering the game of Go with deep neural networks and tree search" (Nature 529, 484-489, 28 Jan 2016), details of which can be found on their website https://deepmind.com/publications.html.
(Previously known just as "AlphaGo," renamed to clarify that we are not affiliated with DeepMind)
This project is a student-led 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.
See the project wiki.
This is not yet a full implementation of AlphaGo. Development is being carried out on the
developbranch. The current emphasis is on speed optimizations, which are necessary to complete training of the value-network and to create feasible tree-search. See the
cython-optimizationbranch for more.
Selected data (i.e. trained models) are released in our data repository.
This project has primarily focused on the neural network training aspect of DeepMind's AlphaGo. We also have a simple single-threaded implementation of their tree search algorithm, though it is not fast enough to be competitive yet.
See the wiki page on the training pipeline for information on how to run the training commands.