Scratch Chess Engine - Game of Kings

ArnoHu

GoK Classic (Medium, white) vs. Delta, frontal attack, later wild checkmate sequence, 93% vs. 87%:
https://lichess.org/lSPVdOES#135

FastSkript

Delta 4 15s vs GOK HCE medium diff
Opening Surprise
86% vs. 79%
https://lichess.org/632GPLyS#1

Last edited by FastSkript (Nov. 26, 2024 15:29:45)

birdracerthree

Quick question, how do I quickly detect isolated pawns?

ArnoHu

birdracerthree wrote:
Quick question, how do I quickly detect isolated pawns?

OK stating the obvious, on any other platform with bitboards.

Even on Scratch, there is still a bit of optimization possible. E.g. by pre-calculating the least and the most advanced pawn rank for every file, and re-using this when detecting isolated or passed pawns. Can be pre-calculated at the beginning of eval(), or incrementally by keeping track via make- / unmake-move.

ArnoHu

FastSkript wrote:
Delta 4 15s vs GOK HCE medium diff
Opening Surprise
86% vs. 79%
https://lichess.org/632GPLyS#1

Replayed same opening on a fast system, inverse result: https://lichess.org/dGzMISjc#64

I suppose GoK Classic needs a certain search depth (= system speed) to keep up with NNUE engines.

Last edited by ArnoHu (Nov. 27, 2024 20:02:40)

FastSkript

GOK NNUE difficult (0+30s) vs Lc0 (0 + 1s)
https://lichess.org/VPSQPiOq
90% - 98% acc

Last edited by FastSkript (Nov. 28, 2024 14:36:10)

ArnoHu

FastSkript wrote:
GOK NNUE difficult (0+30s) vs Lc0 (0 + 1s)
https://lichess.org/VPSQPiOq
90% - 98% acc

Thx, fixed GoK's blunder (aspiration-window related).

Last edited by ArnoHu (Nov. 29, 2024 05:25:44)

FastSkript

GOK NNUE difficult (0+30s) vs Tomitank 4.2 (depth 12)
https://lichess.org/qAzQtnV0
98% - 96% acc

ababoin07

I ll do:
Lc0 vs goK difficult comp mod
Komodo (max) vs goK _ _ _
Small brain 4 vs goK _ _ _
Sf 10 vs goK _ _ _ (for the fun)

And fastskript, how many knps you get with Lc0?

Last edited by ababoin07 (Nov. 28, 2024 17:07:05)

ArnoHu

ababoin07 wrote:
I ll do:
Lc0 vs goK difficult comp mod
Komodo (max) vs goK _ _ _
Small brain 4 vs goK _ _ _
Sf 10 vs goK _ _ _ (for the fun)

And fastskript, how many knps you get with Lc0?

How is it going? Would you mind a rematch with Lc0 with the latest GoK NNUE version? Also you might want to use the TW FasterList protoype, means https://experiments.turbowarp.org/faster-lists/148769358?turbo

I have seen up to 100k NPS, but that probably was when I also had the Batch-List-Ops TW extension on.

Last edited by ArnoHu (Nov. 29, 2024 11:34:59)

ElapElap

Try berserk and also Komodo dragon 3.3

ababoin07

ElapElap wrote:
Try berserk and also Komodo dragon 3.3

okay

WhiteDoveOfficial

ababoin07 wrote:
ElapElap wrote:
Try berserk and also Komodo dragon 3.3
okay

How have the games gone?

FastSkript

Delta 4 (15s) plays Shuffle 5.2 in a Rapid game with 97% accuracy:

https://lichess.org/dOcrNaVU

https://github.com/ArjunBasandrai/shuffle-chess-engine?tab=readme-ov-file

ababoin07

WhiteDoveOfficial wrote:
ababoin07 wrote:
ElapElap wrote:
Try berserk and also Komodo dragon 3.3
okay

How have the games gone?

i only do Lc0 against goK NNUE, draw (goK NNUE difficult 1min/Lc0 1 second, on a RTX 2060 SUPER)

ababoin07

FastSkript wrote:
Delta 4 (15s) plays Shuffle 5.2 in a Rapid game with 97% accuracy:

https://lichess.org/dOcrNaVU

https://github.com/ArjunBasandrai/shuffle-chess-engine?tab=readme-ov-file

:brilliant: Cool, i'll check it!!

qwertycodechamp90411

Out of curiosity, I gave 2 puzzles to GoK. The first was one stockfish couldn't solve, and as expected GoK didn't see it either. However, the second one was interesting.

Puzzle: https://lichess.org/study/35jX0ln7

Tests were on TurboWarp 250FPS (although I'm pretty sure my computer only runs at 60FPS). GoK missed 8. b7! on Blitz 2, and missed 12. e8=R! on difficult, but found it when I undid the move and retried.

ababoin07

qwertycodechamp90411 wrote:
Out of curiosity, I gave 2 puzzles to GoK. The first was one stockfish couldn't solve, and as expected GoK didn't see it either. However, the second one was interesting.

Puzzle: https://lichess.org/study/35jX0ln7

Tests were on TurboWarp 250FPS (although I'm pretty sure my computer only runs at 60FPS). GoK missed 8. b7! on Blitz 2, and missed 12. e8=R! on difficult, but found it when I undid the move and retried.

You can press “c” against and against to enable differents mode (comp., puzzle…)

qwertycodechamp90411

ababoin07 wrote:
qwertycodechamp90411 wrote:
Out of curiosity, I gave 2 puzzles to GoK. The first was one stockfish couldn't solve, and as expected GoK didn't see it either. However, the second one was interesting.

Puzzle: https://lichess.org/study/35jX0ln7

Tests were on TurboWarp 250FPS (although I'm pretty sure my computer only runs at 60FPS). GoK missed 8. b7! on Blitz 2, and missed 12. e8=R! on difficult, but found it when I undid the move and retried.
You can press “c” against and against to enable differents mode (comp., puzzle…)

How did I not know that lol

ArnoHu

qwertycodechamp90411 wrote:
Out of curiosity, I gave 2 puzzles to GoK. The first was one stockfish couldn't solve, and as expected GoK didn't see it either. However, the second one was interesting.

Puzzle: https://lichess.org/study/35jX0ln7

Tests were on TurboWarp 250FPS (although I'm pretty sure my computer only runs at 60FPS). GoK missed 8. b7! on Blitz 2, and missed 12. e8=R! on difficult, but found it when I undid the move and retried.

Do you refer to GoK Classic or GoK NNUE? On Scratch 3 or on TurboWarp?

8. b7
NNUE finds it at depth 7, Classic at depth 5, both easily in under 1 second (TW)

12. e8=R
NNUE finds it at depth 3 (0.1sec), Classic at depth 1 (0.01sec) (TW)

Don't use Blitz, it does not have search extensions. FPS rate has no effect on GoK performance.

Last edited by ArnoHu (Dec. 5, 2024 21:54:51)

Discuss Scratch