Skip to content

Training resign fix#526

Merged
killerducky merged 4 commits intoglinscott:nextfrom
Tilps:training_resign_fix
May 5, 2018
Merged

Training resign fix#526
killerducky merged 4 commits intoglinscott:nextfrom
Tilps:training_resign_fix

Conversation

@Tilps
Copy link
Contributor

@Tilps Tilps commented May 4, 2018

Putting up this PR without any testing, to get early opinions.

I think its a pretty obvious improvement over the previous logic - but I'll need to double check its interaction with the client to be sure.

Tilps added 4 commits April 29, 2018 09:57
It was spinning on resign, if the 450th ply was a win/loss/draw it
wasn't recognized, and if the 450th ply was a win/loss/draw it was
trying to play a move from that position.
@Tilps
Copy link
Contributor Author

Tilps commented May 4, 2018

The bugs I'm aiming to fix here.

  1. Resign would cause the loop to play out until 450 ply - each time the same position would be called, which would resign again - but it means the training data is flooded with the resign position.
  2. (Historic issue) A checkmate on ply 450 would be ignored and the game declared a draw.

@killerducky
Copy link
Collaborator

To enable clients, I think we need this and #449 right? This is for self-play only, and #449 for matches?

@Tilps
Copy link
Contributor Author

Tilps commented May 4, 2018

Yes if we want both self play and matches to use resign, both are useful.

(I don't personally see the value in resign for matches... but its not my opinion alone that matters :P)

@Tilps
Copy link
Contributor Author

Tilps commented May 5, 2018

Did some testing, including using the decode_training python script. Seems to be working fine?

@killerducky killerducky merged commit 7667618 into glinscott:next May 5, 2018
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants