Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Leela analysis reports low B winrate, but B actually wins -- self play confirms #18

Closed
fishcu opened this Issue Feb 7, 2018 · 3 comments

Comments

Projects
None yet
2 participants
@fishcu
Copy link
Collaborator

fishcu commented Feb 7, 2018

I've seen this issue in more than one game. During the analysis with Leela, a very low winrate is shown for black (B), when actually B is winning. When looking at the .rsgf afterwards, and letting Leela self play (after "open position"), you can confirm that actually white (W) resigns instead of black.

I suspect that this may be a bug with komi, that is, the correct komi is not forwarded to leela. In the example game, white would win with regular komi, but not with the game komi (0.5).

I have attached a sample .sgf, the resulting .rsgf, and a screenshot showing the problem.

eggless-ibd.rsgf.zip
eggless-ibd.sgf.zip
wrong_counting

@fishcu

This comment has been minimized.

Copy link
Collaborator Author

fishcu commented Feb 7, 2018

Additionally, I attach this screenshot from the beginning of the analysis, where the GUI shows komi "0.5". So, I cannot be sure if there is a bug with the komi, but I'm still very suspicious that it is related to that.

eggless-ibd

@pnprog

This comment has been minimized.

Copy link
Owner

pnprog commented Feb 9, 2018

Hum, I can't reproduce this issue with Leela and your sgf file.

When implementing the live analysis feature, I encountered and fixed plenty of bugs, maybe I fixed that one as well. Can you try again with latest sources?

@fishcu

This comment has been minimized.

Copy link
Collaborator Author

fishcu commented Feb 9, 2018

I re-tried with the latest commit and indeed the issue appears to be fixed!

@fishcu fishcu closed this Feb 9, 2018

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
You can’t perform that action at this time.