Kensuke Matsuzaki
Kensuke Matsuzaki
"Measure Don't Guess" would be suitable for such issue.
Without GPU, `--reuse-subtree` or `--pondering` may be slow. https://github.com/zakki/Ray/issues/21
Priority is relatively low. Pull requests, a documentation about install or other distribution package are welcome.
I think it's open problem. Please try it and send PR or write the paper.
CNTK evaluation library can load multiple model. Also you can merge multiple models into single model, and switch them using BrainScript. https://github.com/Microsoft/CNTK/wiki/CloneFunction https://github.com/Microsoft/CNTK/wiki/If-Operation
It may be possible, but approximately it requires 10TB SSD to store data, 2 week to generate training data, 2 month to training. It is too large for me.
Now Ray uses only games that human plays to train. Data generation with RL policy and random play like AlphaGo may improve endgame. But that requires enormous resource too. 😱
@j8580 CGI published "Multi-Labelled Value Networks for Computer Go" https://arxiv.org/abs/1705.10701 I haven't understood detail, but it looks similar to your idea in some aspect.
https://github.com/zakki/Ray/blob/nn/src/UctSearch.cpp#L2539 Is what you mean replacing `komi` to `dynamic_komi` here?
Perhaps porting Rn to TensorFlow, Caffe or something is easier than porting CNTK to Mac.