On 21/10/2017 14:21, David Ongaro wrote: > I understand that DeepMind might be unable to release the source code > of AlphaGo due to policy or licensing reasons, but it would be great > (and probably much more valuable) if they could release the fully > trained network.
The source of AlphaGo Zero is really of zero interest (pun intended). It can be obtained by ripping out ~50% of the Ray/Rn or AQ code (everything related to MC playouts) and some minimal changes to evaluate the same network for scoring and policy. Same for Leela. It's literally possible to have a "Leela/Ray/AQ Zero" in a week or so (it'll require a GPU or performance will be atrocious). Of course, I can't give you the trained network to load into it. That'll take another 88642 weeks. So yes, the database of 29M self-play games would be immensely more valuable. (Probably like the last 5M or so is fine, too). I prefer the games over the network - with the games it's easier to train a smaller network that gives better results on PC's that don't have 4 TPUs in them. -- GCP _______________________________________________ Computer-go mailing list Computer-go@computer-go.org http://computer-go.org/mailman/listinfo/computer-go