ArnoHu wrote:

ArnoHu wrote:

GoK NNUE vs. Black Crow, 97% vs. 90%:
https://lichess.org/study/PJTsaIt9/C8DKkH6w

GoK Classic vs. Wolverine 2, 97% vs. 92%:
https://lichess.org/study/PJTsaIt9/gDzJWeCE

GoK Classic vs. Wolverine 2, 92% vs. 92%:
https://lichess.org/study/PJTsaIt9/lfmhkwvB

GoK Classic vs. Shallow Blue 3 NNUE, 93% vs. 88%:
https://lichess.org/study/PJTsaIt9/WQo6VfpY

GoK NNUE vs. Black Crow, 97% vs. 88%:
https://lichess.org/study/PJTsaIt9/Bjdy1LVl

GoK NNUE vs. Black Crow, 98% vs. 94%:
https://lichess.org/study/PJTsaIt9/IsvL36xE

GoK NNUE vs. Chess.com Maximum Bot 25 (3200), 92% vs. 97%:
https://lichess.org/study/PJTsaIt9/Cp6Do71r

GoK Classic vs. Wolverine 2, 93% vs. 90%:
https://lichess.org/study/PJTsaIt9/hO79upib

GoK Classic vs. Wolverine 2, 96% vs. 93%:
https://lichess.org/study/PJTsaIt9/acpV6Ocy

GoK Classic vs. White Dove, 94% vs. 90%:
https://lichess.org/study/PJTsaIt9/8HMx8PlX

GoK Classic vs. Delta, 88% vs. 88%:
https://lichess.org/study/PJTsaIt9/vZzIwlpH

GoK NNUE vs. Black Crow, 97% vs. 90%:
https://lichess.org/study/PJTsaIt9/i2zBpRio

GoK Classic vs. Wolverine 2, 96% vs. 97%:
https://lichess.org/study/PJTsaIt9/zuHGOAEL

GoK Classic vs. White Dove, 96% vs. 87%:
https://lichess.org/study/PJTsaIt9/EguhDcg4

GoK Classic vs. Wolverine 2, 93% vs. 86%:
https://lichess.org/study/PJTsaIt9/Vsixq50j

GoK NNUE vs. Black Crow, 95% vs. 94%:
https://lichess.org/study/PJTsaIt9/RzUHewuG

Thanks, some more great games!

Looks like forward pruning on the last two plies of the search is backfiring for Wolverine. I figured worst case scenario is it needs to search 2 plies deeper to find a good move – but I'm seeing some cases where forward pruning leaves Wolverine completely blind to certain ideas, no matter how long the search. I don't really understand this, but empirical testing suggests it's an issue with futility and late move pruning applied near the leaves.

Next Wolverine will have a better tuned search to avoid these blind spots. I am also working on a small opening book, for the sake of variety and also to avoid those cases where Wolverine gets a losing position out of the opening. Coming soon

waabooboo wrote:

ArnoHu wrote:

ArnoHu wrote:

GoK NNUE vs. Black Crow, 97% vs. 90%:
https://lichess.org/study/PJTsaIt9/C8DKkH6w

GoK Classic vs. Wolverine 2, 97% vs. 92%:
https://lichess.org/study/PJTsaIt9/gDzJWeCE

GoK Classic vs. Wolverine 2, 92% vs. 92%:
https://lichess.org/study/PJTsaIt9/lfmhkwvB

GoK Classic vs. Shallow Blue 3 NNUE, 93% vs. 88%:
https://lichess.org/study/PJTsaIt9/WQo6VfpY

GoK NNUE vs. Black Crow, 97% vs. 88%:
https://lichess.org/study/PJTsaIt9/Bjdy1LVl

GoK NNUE vs. Black Crow, 98% vs. 94%:
https://lichess.org/study/PJTsaIt9/IsvL36xE

GoK NNUE vs. Chess.com Maximum Bot 25 (3200), 92% vs. 97%:
https://lichess.org/study/PJTsaIt9/Cp6Do71r

GoK Classic vs. Wolverine 2, 93% vs. 90%:
https://lichess.org/study/PJTsaIt9/hO79upib

GoK Classic vs. Wolverine 2, 96% vs. 93%:
https://lichess.org/study/PJTsaIt9/acpV6Ocy

GoK Classic vs. White Dove, 94% vs. 90%:
https://lichess.org/study/PJTsaIt9/8HMx8PlX

GoK Classic vs. Delta, 88% vs. 88%:
https://lichess.org/study/PJTsaIt9/vZzIwlpH

GoK NNUE vs. Black Crow, 97% vs. 90%:
https://lichess.org/study/PJTsaIt9/i2zBpRio

GoK Classic vs. Wolverine 2, 96% vs. 97%:
https://lichess.org/study/PJTsaIt9/zuHGOAEL

GoK Classic vs. White Dove, 96% vs. 87%:
https://lichess.org/study/PJTsaIt9/EguhDcg4

GoK Classic vs. Wolverine 2, 93% vs. 86%:
https://lichess.org/study/PJTsaIt9/Vsixq50j

GoK NNUE vs. Black Crow, 95% vs. 94%:
https://lichess.org/study/PJTsaIt9/RzUHewuG

Thanks, some more great games!

Looks like forward pruning on the last two plies of the search is backfiring for Wolverine. I figured worst case scenario is it needs to search 2 plies deeper to find a good move – but I'm seeing some cases where forward pruning leaves Wolverine completely blind to certain ideas, no matter how long the search. I don't really understand this, but empirical testing suggests it's an issue with futility and late move pruning applied near the leaves.

Next Wolverine will have a better tuned search to avoid these blind spots. I am also working on a small opening book, for the sake of variety and also to avoid those cases where Wolverine gets a losing position out of the opening. Coming soon

Can't wait to see it! Keep up the great work!

GoK Classic vs. White Dove, 94% vs. 89%:
https://lichess.org/study/PJTsaIt9/beygonqO

GoK Classic vs. Wolverine 2, 95% vs. 90%:
https://lichess.org/study/PJTsaIt9/eEGrQO2I

GoK Classic vs. Wolverine 2, 92% vs. 87%:
https://lichess.org/study/PJTsaIt9/ZDfNBJ9d

GoK Classic vs. Shallow Blue 3 NNUE, 96% vs. 91%:
https://lichess.org/study/PJTsaIt9/Y6yjzQeR

Obviously, my python environnement is totally broken. And, I don’t understand anything about the new python manager. I tried to use anaconda, but obviously, pytorch doesn’t work woth cuda, and that’s a little little bit slower on cpu compared to on gpu. I will try to repair this ****** environnement, and next, I ́ll begin to work on a semi-NNUE for TK5.

ArnoHu wrote:

GoK Classic vs. White Dove, 94% vs. 89%:
https://lichess.org/study/PJTsaIt9/beygonqO

GoK Classic vs. Wolverine 2, 95% vs. 90%:
https://lichess.org/study/PJTsaIt9/eEGrQO2I

GoK Classic vs. Wolverine 2, 92% vs. 87%:
https://lichess.org/study/PJTsaIt9/ZDfNBJ9d

GoK Classic vs. Shallow Blue 3 NNUE, 96% vs. 91%:
https://lichess.org/study/PJTsaIt9/Y6yjzQeR

GoK NNUE vs. Black Crow2, 96% vs. 92%:
https://lichess.org/study/PJTsaIt9/leTGveI2

GoK Classic vs. Wolverine 2, 90% vs. 87%:
https://lichess.org/study/PJTsaIt9/SV4bLtTt

GoK Classic vs. Wolverine 2, 95% vs. 88%:
https://lichess.org/study/PJTsaIt9/w5rKoXbK

I started my Connect 4 bot from scratch a few days ago. I constructed a static eval that can beat the shared version of my bot if it gets lucky.

I think I definitely stop scratch (at least for 2 years, I ́m really really busy), and scratch isn’t really The langage to learn if you want to have a future lol… so well, I ́ll still be a bit online.