Version controlled data for the Rochester-NRT/AlphaGo repository.
Still working on it. Follow progress in the main repository. There is enough here to load and run a few partially-trained policy networks as standalone players.
scripts/
contains short code snippets that we've found useful
expert_play/
contains datasets of SGF files. We do not have permission to release all datasets that we are using, hence some models are trained on datasets not present in this directory
models/
contains json
model specifications. For example
models/CNNPolicy/12plane_default.json
specifies aCNNPolicy
object with the features['board', 'ones', 'turns_since']
models/CNNPolicy/20plane_default.json
is likewise, with the features['board', 'ones', 'turns_since', 'liberties']
models/CNNPolicy/46plane_default.json
is likewise with all features except ladders
training_results/
contains trained models and epoch checkpoints.