search

yhyu13 / AlphaGOZero-python-tensorflow

Congratulation to DeepMind! This is a reengineering implementation (on behalf of many other git repo in /support/) of DeepMind's Oct19th publication: [Mastering the Game of Go without Human Knowledge]. The supervised learning approach is more practical for individuals. (This repository has single purpose of education only)
MIT License
341 stars 115 forks source link

What level is pretrained large20 against random? The win rate is always 0.5 #16

Open arisliang opened 6 years ago

arisliang commented 6 years ago

The win rate is always 0.5 between large 20 and random, even approaching the end. Is the pretrained model no particular difference than random, even though it's running DNN+MCTS?

screenshot from 2018-04-30 09-09-38

yhyu13 commented 6 years ago

@arisliang

I tested it and found the same problem. The answer from me is that the MCTS player is broken in some part. If you want to test the greedy policy (DNN top1), you can pull the lasted commit.

arisliang commented 6 years ago

Pulled the updates, greedy policy works. But it doesn't give win rate. Plus it doesn't seem to me the greedy policy has advantage against random.

screenshot from 2018-05-01 10-09-41