ArnoHu wrote:

ArnoHu wrote:

ArnoHu wrote:

ArnoHu wrote:

Draw between GoK (black) and White Dove, 10sec think time, both at 94% accuracy: https://lichess.org/ioJZkpXm#138

Rematch, tough fight, GoK(white) wins at 94% vs. 90% accuracy: https://lichess.org/ob5IeBCC#147

Game #3, GoK white, 97% vs. 92% accuracy

Game #4 was great, GoK black, 98% vs. 91% accuracy: https://lichess.org/qMkJuf25#100

Final game of the series on Scratch 3, 10sec think time. Another miraculous late-midgame escape by GoK (white), after trailing behind 6,0 (it will not always be so lucky, on the other hand WD did not see the mistakes at first either). 88% vs. 83% accuracy: https://lichess.org/mFVirrsW#145

GoK's latest version (white) in a low-accuracy but entertaining game against WD (10 sec). Same pattern again, GoK trailing behind during early midgame, enters endgame at roughly equal terms, then wins: https://lichess.org/X2Bkk8vk#149

The evaluation graph roughly correlated with the difference in search depth, not really surprising. During midgame GoK +1.5 in avg. search depth, later +3.

WD got its rook pinned at move 68, that was kind of strange.

ArnoHu wrote:

ArnoHu wrote:

ArnoHu wrote:

ArnoHu wrote:

ArnoHu wrote:

Draw between GoK (black) and White Dove, 10sec think time, both at 94% accuracy: https://lichess.org/ioJZkpXm#138

Rematch, tough fight, GoK(white) wins at 94% vs. 90% accuracy: https://lichess.org/ob5IeBCC#147

Game #3, GoK white, 97% vs. 92% accuracy

Game #4 was great, GoK black, 98% vs. 91% accuracy: https://lichess.org/qMkJuf25#100

Final game of the series on Scratch 3, 10sec think time. Another miraculous late-midgame escape by GoK (white), after trailing behind 6,0 (it will not always be so lucky, on the other hand WD did not see the mistakes at first either). 88% vs. 83% accuracy: https://lichess.org/mFVirrsW#145

GoK's latest version (white) in a low-accuracy but entertaining game against WD (10 sec). Same pattern again, GoK trailing behind during early midgame, enters endgame at roughly equal terms, then wins: https://lichess.org/X2Bkk8vk#149

The evaluation graph roughly correlated with the difference in search depth, not really surprising. During midgame GoK +1.5 in avg. search depth, later +3.

WD got its rook pinned at move 68, that was kind of strange.

OK, I spent some time investigating with Chrome Dev Tools. Most time was used in move generation close to max depth. Search / move generation depends on up-to-date attack tables (esp. during full search), but if hash moves are fund I limited attack table creation (trying to improve performance), expecting that in such a case attack tables would not be necessary (code works with and w/o those tables, but w/o them will be slower). Well I was wrong, which explains the low search depth during midgame.

This game against WD was played after the fix, and finally GoK midgame performance is back to normal (+2.5 avg. search depth compared to WD):

https://lichess.org/vmwCfsKm#97

ArnoHu wrote:

GoK Medium (white) against Element (6+8), reached depth 10 to 12 during midgame, 12 to 14 during endgame, 97% vs. 93% accuracy: https://lichess.org/AZIqdcOz#103

birdracerthree wrote:

ArnoHu wrote:

ArnoHu wrote:

ArnoHu wrote:

ArnoHu wrote:

ArnoHu wrote:

Draw between GoK (black) and White Dove, 10sec think time, both at 94% accuracy: https://lichess.org/ioJZkpXm#138

Rematch, tough fight, GoK(white) wins at 94% vs. 90% accuracy: https://lichess.org/ob5IeBCC#147

Game #3, GoK white, 97% vs. 92% accuracy

Game #4 was great, GoK black, 98% vs. 91% accuracy: https://lichess.org/qMkJuf25#100

Final game of the series on Scratch 3, 10sec think time. Another miraculous late-midgame escape by GoK (white), after trailing behind 6,0 (it will not always be so lucky, on the other hand WD did not see the mistakes at first either). 88% vs. 83% accuracy: https://lichess.org/mFVirrsW#145

GoK's latest version (white) in a low-accuracy but entertaining game against WD (10 sec). Same pattern again, GoK trailing behind during early midgame, enters endgame at roughly equal terms, then wins: https://lichess.org/X2Bkk8vk#149

The evaluation graph roughly correlated with the difference in search depth, not really surprising. During midgame GoK +1.5 in avg. search depth, later +3.

WD got its rook pinned at move 68, that was kind of strange.

OK, I spent some time investigating with Chrome Dev Tools. Most time was used in move generation close to max depth. Search / move generation depends on up-to-date attack tables (esp. during full search), but if hash moves are fund I limited attack table creation (trying to improve performance), expecting that in such a case attack tables would not be necessary (code works with and w/o those tables, but w/o them will be slower). Well I was wrong, which explains the low search depth during midgame.

This game against WD was played after the fix, and finally GoK midgame performance is back to normal (+2.5 avg. search depth compared to WD):

https://lichess.org/vmwCfsKm#97

You should run more games to be sure. We can’t know for sure after one game. How is the automation project going?

Are attack tables stored separately per piece, or as one long string?

ArnoHu wrote:

birdracerthree wrote:

ArnoHu wrote:

ArnoHu wrote:

ArnoHu wrote:

ArnoHu wrote:

ArnoHu wrote:

Draw between GoK (black) and White Dove, 10sec think time, both at 94% accuracy: https://lichess.org/ioJZkpXm#138

Rematch, tough fight, GoK(white) wins at 94% vs. 90% accuracy: https://lichess.org/ob5IeBCC#147

Game #3, GoK white, 97% vs. 92% accuracy

Game #4 was great, GoK black, 98% vs. 91% accuracy: https://lichess.org/qMkJuf25#100

Final game of the series on Scratch 3, 10sec think time. Another miraculous late-midgame escape by GoK (white), after trailing behind 6,0 (it will not always be so lucky, on the other hand WD did not see the mistakes at first either). 88% vs. 83% accuracy: https://lichess.org/mFVirrsW#145

GoK's latest version (white) in a low-accuracy but entertaining game against WD (10 sec). Same pattern again, GoK trailing behind during early midgame, enters endgame at roughly equal terms, then wins: https://lichess.org/X2Bkk8vk#149

The evaluation graph roughly correlated with the difference in search depth, not really surprising. During midgame GoK +1.5 in avg. search depth, later +3.

WD got its rook pinned at move 68, that was kind of strange.

*snip*
This game against WD was played after the fix, and finally GoK midgame performance is back to normal (+2.5 avg. search depth compared to WD):

https://lichess.org/vmwCfsKm#97

You should run more games to be sure. We can’t know for sure after one game. How is the automation project going?

Are attack tables stored separately per piece, or as one long string?

It was one of several games. Also did per-board tests, search depth increased by 1 ply in most cases.

Game #2 against WD (white), 92% vs. 89% accuracy: https://lichess.org/Wzmq5r8f#154

Attack tables are strings of 64 digits (number of attackers per square) and 128 digits (source square index of attacker with lowest value per square).

Automation is only 30% done. I started the PlayStore GoK Android version in parallel…

birdracerthree wrote:

ArnoHu wrote:

It was one of several games. Also did per-board tests, search depth increased by 1 ply in most cases.

Game #2 against WD (white), 92% vs. 89% accuracy: https://lichess.org/Wzmq5r8f#154

Attack tables are strings of 64 digits (number of attackers per square) and 128 digits (source square index of attacker with lowest value per square).

Automation is only 30% done. I started the PlayStore GoK Android version in parallel…

Are there any differences between the GoK app and Turbowarp GoK? GoK doesn’t have some openings played by humans (Ponziani, Fried Liver, incorrect Morphy Defense, Catalan, etc.)

birdracerthree wrote:

Element defeats GoK
Yesterday, I decided to run a GoK vs Element match to test the new Four Knights Scotch against GoK (I didn't know how it would play the middlegame against GoK). In this game, Element defeats GoK in 57 moves!! Incredibly, Element sored a 99% accuracy game!

However, the only reason Element was is due to a GoK opening book error in this opening. More specifically, the line 53371329634602195843072252362936463622374337041354461430 should be 53371329634602195843072252362936463622374337041354461228 (f7f5?? replaced with d7d5). The line may continue 533713296346021958430722523629364636223743370413544612286226031226191019 (1. e4 e5 2. Nf3 Nc6 3. Nc3 Nf6 4. d4 exd4 5. Nxd4 Nxe4 6. Nxe4 Qe7 7. f3 d5 8. Bb5 Bd7 9. Bxc6 bxc6). This is still an impressive win because White Dove and Element have both blundered out of a 2 pawns vs knight lead, but Element won a clean knight due to this bug.

Does this count as the first standard chess win against GoK on Turbowarp? Technically, yes, as neither engine was modified during the game in any way. However, I think Element winning a piece due to an opening book bug doesn't make the game “clean”. Feel free to express your thoughts regarding this.

MoosenMooALot wrote:

birdracerthree wrote:

Element defeats GoK
Yesterday, I decided to run a GoK vs Element match to test the new Four Knights Scotch against GoK (I didn't know how it would play the middlegame against GoK). In this game, Element defeats GoK in 57 moves!! Incredibly, Element sored a 99% accuracy game!

However, the only reason Element was is due to a GoK opening book error in this opening. More specifically, the line 53371329634602195843072252362936463622374337041354461430 should be 53371329634602195843072252362936463622374337041354461228 (f7f5?? replaced with d7d5). The line may continue 533713296346021958430722523629364636223743370413544612286226031226191019 (1. e4 e5 2. Nf3 Nc6 3. Nc3 Nf6 4. d4 exd4 5. Nxd4 Nxe4 6. Nxe4 Qe7 7. f3 d5 8. Bb5 Bd7 9. Bxc6 bxc6). This is still an impressive win because White Dove and Element have both blundered out of a 2 pawns vs knight lead, but Element won a clean knight due to this bug.

Does this count as the first standard chess win against GoK on Turbowarp? Technically, yes, as neither engine was modified during the game in any way. However, I think Element winning a piece due to an opening book bug doesn't make the game “clean”. Feel free to express your thoughts regarding this.

GoK LOST?!?!?! I NEVER THOUGHT I'D LIVE TO SEE THIS DAY!

birdracerthree wrote:

MoosenMooALot wrote:

birdracerthree wrote:

Element defeats GoK
Yesterday, I decided to run a GoK vs Element match to test the new Four Knights Scotch against GoK (I didn't know how it would play the middlegame against GoK). In this game, Element defeats GoK in 57 moves!! Incredibly, Element sored a 99% accuracy game!

However, the only reason Element was is due to a GoK opening book error in this opening. More specifically, the line 53371329634602195843072252362936463622374337041354461430 should be 53371329634602195843072252362936463622374337041354461228 (f7f5?? replaced with d7d5). The line may continue 533713296346021958430722523629364636223743370413544612286226031226191019 (1. e4 e5 2. Nf3 Nc6 3. Nc3 Nf6 4. d4 exd4 5. Nxd4 Nxe4 6. Nxe4 Qe7 7. f3 d5 8. Bb5 Bd7 9. Bxc6 bxc6). This is still an impressive win because White Dove and Element have both blundered out of a 2 pawns vs knight lead, but Element won a clean knight due to this bug.

Does this count as the first standard chess win against GoK on Turbowarp? Technically, yes, as neither engine was modified during the game in any way. However, I think Element winning a piece due to an opening book bug doesn't make the game “clean”. Feel free to express your thoughts regarding this.

GoK LOST?!?!?! I NEVER THOUGHT I'D LIVE TO SEE THIS DAY!

The question was never “if”, it was a matter of “when”. I linked these 5 games to you almost 2 months ago : https://scratch.mit.edu/discuss/post/7665285/ https://scratch.mit.edu/discuss/post/7717811/ https://scratch.mit.edu/discuss/post/7719831/ https://scratch.mit.edu/discuss/post/7752223/ https://scratch.mit.edu/discuss/post/7769979/

There are more games since then (thanks to White Dove), but these are the closest calls yet. One of them went into a winning king and pawn endgame.

The first “true” win is still on the table in my opinion.

MoosenMooALot wrote:

GoK 6.555 Turbowarp Blitz 2 vs. Stockfish Level 5 (lichess):
GoK white vs. Stockfish black:
GoK wins in 45 moves: https://lichess.org/AE89p9Cb/white#0
Stockfish white vs. GoK black:
GoK wins in 50 moves: https://lichess.org/nJaFmY8z/black#1
Fantastic job by GoK on that second game! With no bad moves and 98% accuracy, GoK did much better than I would've expected from such a small think time. I might try some against Stockfish Level 6 soon as GoK Blitz 2 is clearly better than level 5.

ababoin07 wrote:

Could someone list me one by one all the goK improvements in its course?

ArnoHu wrote:

ababoin07 wrote:

Could someone list me one by one all the goK improvements in its course?

This forum posting provides an overview:
https://scratch.mit.edu/discuss/post/2967632/

And of course here is its github history:
https://github.com/ArnoHue/scratch/commits/master/chess/Engine.scratch

birdracerthree wrote:

Why doesn’t GoK see the checkmate here? White Dove sees it on ply 8.