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

⚖️ Improve endgame scaling with pawn count #928

Merged
merged 3 commits into from
Aug 19, 2024

Conversation

eduherminio
Copy link
Member

@eduherminio eduherminio commented Aug 19, 2024

Previously scaling added in #821
This improves it a bit

Test  | eval/improve-eg-scaling
Elo   | 6.48 +- 3.89 (95%)
SPRT  | 8.0+0.08s Threads=1 Hash=32MB
LLR   | 2.89 (-2.25, 2.89) [0.00, 3.00]
Games | 15020: +4392 -4112 =6516
Penta | [433, 1742, 2945, 1892, 498]
https://openbench.lynx-chess.com/test/622/

Doing it when phase < 3 makes it surprisingly ineffective:

Test  | eval/improve-eg-scaling-2
Elo   | -10.07 +- 6.21 (95%)
SPRT  | 8.0+0.08s Threads=1 Hash=32MB
LLR   | -2.30 (-2.25, 2.89) [0.00, 3.00]
Games | 5796: +1543 -1711 =2542
Penta | [178, 777, 1147, 627, 169]
https://openbench.lynx-chess.com/test/624/

Applying my own method/formula, regardless the phase, just in case this is the key and not the formula used.

Test  | eval/improve-eg-scaling-3
Elo   | -3.96 +- 5.91 (95%)
SPRT  | 8.0+0.08s Threads=1 Hash=32MB
LLR   | -1.16 (-2.25, 2.89) [0.00, 3.00]
Games | 6490: +1814 -1888 =2788
Penta | [198, 836, 1248, 768, 195]
https://openbench.lynx-chess.com/test/625/

@eduherminio eduherminio marked this pull request as ready for review August 19, 2024 11:47
@eduherminio eduherminio merged commit 463ea29 into main Aug 19, 2024
27 checks passed
@eduherminio eduherminio deleted the eval/improve-eg-scaling branch August 19, 2024 11:47
eduherminio added a commit to lynx-chess/texel-tuner that referenced this pull request Aug 19, 2024
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.

1 participant