Skip to content

/AlphaGo

A replication 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.
  1. JavaScript 58.9%
  2. Python 28.4%
  3. CSS 12.1%
  4. HTML 0.6%
JavaScript Python CSS HTML
Find file
Download ZIP
Pull request Compare This branch is 12 commits ahead, 53 commits behind develop.
Latest commit 0d155b5 cjbates-mac End-of-game condition in progress. Need to make playing in eye illega… 
…l. Also, greedy_player_benchmark.py needs to be proper benchmark.
Failed to load latest commit information.
AlphaGo End-of-game condition in progress. Need to make playing in eye illega…
benchmarks End-of-game condition in progress. Need to make playing in eye illega…
data Training infrastructure
interface End-of-game condition in progress. Need to make playing in eye illega…
tests Merged develop into player
.gitignore very simple benchmarking code thanks to @irisyupingren
.gitmodules fresh start: project skeleton
.travis.yml keras depends on bleeding-edge Theano, so we can't use 0.7
CONTRIBUTING.md readme and contributing docs updated, some placeholders
LICENSE fresh start: project skeleton
README.md readme and contributing docs updated, some placeholders
requirements.txt added h5py to requirements so Keras can save/load weights

README.md

AlphaGo Replication

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.

Build Status

Current project status

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)

How to contribute

See the 'Contributing' document.

Something went wrong with that request. Please try again.