ArnoHu wrote:

ArnoHu wrote:

ababoin07 wrote:

ArnoHu wrote:

ababoin07 wrote:

ArnoHu wrote:

ababoin07 wrote:

ArnoH wrote:

It is off-topic, but I have to say it. @Grijfland, Austria is 1:0 vs Netherlands, Possession 63% vs. 37% at minute 14

Just off topic XD XD is this for soccer?

Sure, what else? You are from France, no?

We won 3:2, and qualified for knockout rounds as 1st in the group, ahead of France and the Netherlands - incredible, everyone is going crazy here!

Yes, I am French and I live in France, and I conclude that you must be in Austria? I am not interested in football… Otherwise, during tests on Destructor, I added 1 to the maximum depth of search for quiescence, and the time went up to 200 seconds (3 minutes) despite the transpositions tables… Do you have an explanation of why time believes so much (lol, Austria in English does not look at all like ‘Autriche’ in French, I thought it was a lost corner of America)

I have heard some ppl outside Europe mix us up with Australia, but France - hey, we have been fighting each other for centuries!

I took a quick look at Destructor 1.4.2 CompMode, please don't use listIndexOf() in inner loops / functions or on large lists, it is a performance killer with O(N) runtime complexity. Most of your CPU cycles are burnt there. There is always an alternative implementation. GoK only applies listIndexOf() at one place, not an inner loop, and the list is of size 4.

Frankly, I don’t even know where I used this or that… I would look at this, thank you for your advice! (Which I have no idea to apply), is there an alternative?

It's listIndexOf (list-indexOf or list-contains, e.g. In “Vérifier board 2 avec roi” for example), also indexOf (string-contains, e.g. in “Essayer de bouger et validation” and “”WEssayer de bouger 2“). If you are looking for the king index on your board, have a variable dedicated for that and update it when kings moves, but don't scan the board for the king each time. Scratch doesn't simply look for a value in a list (which would be bad enough), internally it does type conversions, tries to trail whitespaces, and more on each and every element comparison.

And the way to find these things is to use browser F12 dev tools on TurboWarp. E.g. on Chrome, press F12, click the record button in ”Performance" tab, then make move and let Destructor run a search, then stop recording, click Performance / Bottom Up, and sort the list of functions by total time. You can then click the .js source link and investigate each line of code on how much time is spent.

More info, string-contains is invoked 50 million times alone by “Essayer de bouger et validation 2”, and this is what it has to do on each of them:
https://github.com/scratchfoundation/scratch-vm/blob/8814c1762f9b67d0710b2ed51b7954222d404df0/src/blocks/scratch3_operators.js#L111

It casts everything to a string, then converts two strings to lower case (means allocating new strings), then invokes includes() on them. What you want to have performance-wise in the innermost code, are operations on scalar numeric values (including comparisons), without any conversions. That is the idea that GoK is built upon.

And list-contains / list-indexOf is even worse, overhead-wise.

ArnoHu wrote:

GoK NNUE (10sec, black) vs. White Dove (P2), GoK wins: https://lichess.org/qanbjD8d#128

ArnoHu wrote:

HasiLover_Test wrote:

ScratchChessChampion wrote:

Hey guys, I'm back from my exam! Hope I can get A's in my subjects.

I plan to organize a tournament for July. Should I organize a chess championship match or something else? Let me know what you think!

Maybe a Blitz Tournament? Where the Engines have to play on time?

Yes, that would be interesting (and fair), requires less time for the organizer. Also, it is cool to watch online with other creators, maybe we find a schedule suitable for most of us?

And: Could you please add GoK NNUE to the Scratch Chess Foundation? Thanks!

ScratchChessChampion wrote:

ArnoHu wrote:

HasiLover_Test wrote:

ScratchChessChampion wrote:

Hey guys, I'm back from my exam! Hope I can get A's in my subjects.

I plan to organize a tournament for July. Should I organize a chess championship match or something else? Let me know what you think!

Maybe a Blitz Tournament? Where the Engines have to play on time?

Yes, that would be interesting (and fair), requires less time for the organizer. Also, it is cool to watch online with other creators, maybe we find a schedule suitable for most of us?

And: Could you please add GoK NNUE to the Scratch Chess Foundation? Thanks!

Sure, i will add it in the July update.

ArnoH wrote:

ababoin07 wrote:

ArnoHu wrote:

ababoin07 wrote:

ArnoHu wrote:

ababoin07 wrote:

ArnoH wrote:

It is off-topic, but I have to say it. @Grijfland, Austria is 1:0 vs Netherlands, Possession 63% vs. 37% at minute 14

Just off topic XD XD is this for soccer?

Sure, what else? You are from France, no?

We won 3:2, and qualified for knockout rounds as 1st in the group, ahead of France and the Netherlands - incredible, everyone is going crazy here!

Yes, I am French and I live in France, and I conclude that you must be in Austria? I am not interested in football… Otherwise, during tests on Destructor, I added 1 to the maximum depth of search for quiescence, and the time went up to 200 seconds (3 minutes) despite the transpositions tables… Do you have an explanation of why time believes so much (lol, Austria in English does not look at all like ‘Autriche’ in French, I thought it was a lost corner of America)

I have heard some ppl outside Europe mix us up with Australia, but France - hey, we have been fighting each other for centuries!

I took a quick look at Destructor 1.4.2 CompMode, please don't use listIndexOf() in inner loops / functions or on large lists, it is a performance killer with O(N) runtime complexity. Most of your CPU cycles are burnt there. There is always an alternative implementation. GoK only applies listIndexOf() at one place, not an inner loop, and the list is of size 4.

Frankly, I don’t even know where I used this or that… I would look at this, thank you for your advice! (Which I have no idea to apply), is there an alternative?

It's listIndexOf (list-indexOf or list-contains, e.g. In “Vérifier board 2 avec roi” for example), also indexOf (string-contains, e.g. in “Essayer de bouger et validation” and “”WEssayer de bouger 2“). If you are looking for the king index on your board, have a variable dedicated for that and update it when kings moves, but don't scan the board for the king each time. Scratch doesn't simply look for a value in a list (which would be bad enough), internally it does type conversions, tries to trail whitespaces, and more on each and every element comparison.

And the way to find these things is to use browser F12 dev tools on TurboWarp. E.g. on Chrome, press F12, click the record button in ”Performance" tab, then make move and let Destructor run a search, then stop recording, click Performance / Bottom Up, and sort the list of functions by total time. You can then click the .js source link and investigate each line of code on how much time is spent.

What I also noticed are busy loops, checking for user input. I suggest to turn all of that off, there are message handlers for what, e.g. when-clicked, when-key-pressed. Loops scanning for input will keep your CPU busy and slow down everything else.

ArnoHu wrote:

ababoin07 wrote:

ArnoHu wrote:

ababoin07 wrote:

ArnoHu wrote:

ababoin07 wrote:

ArnoH wrote:

It is off-topic, but I have to say it. @Grijfland, Austria is 1:0 vs Netherlands, Possession 63% vs. 37% at minute 14

Just off topic XD XD is this for soccer?

Sure, what else? You are from France, no?

We won 3:2, and qualified for knockout rounds as 1st in the group, ahead of France and the Netherlands - incredible, everyone is going crazy here!

Yes, I am French and I live in France, and I conclude that you must be in Austria? I am not interested in football… Otherwise, during tests on Destructor, I added 1 to the maximum depth of search for quiescence, and the time went up to 200 seconds (3 minutes) despite the transpositions tables… Do you have an explanation of why time believes so much (lol, Austria in English does not look at all like ‘Autriche’ in French, I thought it was a lost corner of America)

I have heard some ppl outside Europe mix us up with Australia, but France - hey, we have been fighting each other for centuries!

I took a quick look at Destructor 1.4.2 CompMode, please don't use listIndexOf() in inner loops / functions or on large lists, it is a performance killer with O(N) runtime complexity. Most of your CPU cycles are burnt there. There is always an alternative implementation. GoK only applies listIndexOf() at one place, not an inner loop, and the list is of size 4.

Frankly, I don’t even know where I used this or that… I would look at this, thank you for your advice! (Which I have no idea to apply), is there an alternative?

It's listIndexOf (list-indexOf or list-contains, e.g. In “Vérifier board 2 avec roi” for example), also indexOf (string-contains, e.g. in “Essayer de bouger et validation” and “”WEssayer de bouger 2“). If you are looking for the king index on your board, have a variable dedicated for that and update it when kings moves, but don't scan the board for the king each time. Scratch doesn't simply look for a value in a list (which would be bad enough), internally it does type conversions, tries to trail whitespaces, and more on each and every element comparison.

And the way to find these things is to use browser F12 dev tools on TurboWarp. E.g. on Chrome, press F12, click the record button in ”Performance" tab, then make move and let Destructor run a search, then stop recording, click Performance / Bottom Up, and sort the list of functions by total time. You can then click the .js source link and investigate each line of code on how much time is spent.

What I also noticed are busy loops, checking for user input. I suggest to turn all of that off, there are message handlers for what, e.g. when-clicked, when-key-pressed. Loops scanning for input will keep your CPU busy and slow down everything else.

ArnoHu wrote:

ababoin07 wrote:

ArnoH wrote:

It is off-topic, but I have to say it. @Grijfland, Austria is 1:0 vs Netherlands, Possession 63% vs. 37% at minute 14

Just off topic XD XD is this for soccer?

Sure, what else? You are from France, no?

We won 3:2, and qualified for knockout rounds as 1st in the group, ahead of France and the Netherlands - incredible, everyone is going crazy here!

ArnoHu wrote:

ArnoHu wrote:

GoK NNUE (10sec, black) vs. White Dove (P2), GoK wins: https://lichess.org/qanbjD8d#128

GoK NNUE (Medium, black) vs. Element (6+8), double-blunder by Element early on, loses in 14 moves: https://lichess.org/VF7AWBLw#28

ArnoHu wrote:

ArnoHu wrote:

GoK NNUE (10sec, black) vs. White Dove (P2), GoK wins: https://lichess.org/qanbjD8d#128

GoK NNUE (Medium, black) vs. Element (6+8), double-blunder by Element early on, loses in 14 moves: https://lichess.org/VF7AWBLw#28

ababoin07 wrote:

ArnoH wrote:

ababoin07 wrote:

ArnoHu wrote:

ababoin07 wrote:

ArnoHu wrote:

ababoin07 wrote:

ArnoH wrote:

It is off-topic, but I have to say it. @Grijfland, Austria is 1:0 vs Netherlands, Possession 63% vs. 37% at minute 14

Just off topic XD XD is this for soccer?

Sure, what else? You are from France, no?

We won 3:2, and qualified for knockout rounds as 1st in the group, ahead of France and the Netherlands - incredible, everyone is going crazy here!

Yes, I am French and I live in France, and I conclude that you must be in Austria? I am not interested in football… Otherwise, during tests on Destructor, I added 1 to the maximum depth of search for quiescence, and the time went up to 200 seconds (3 minutes) despite the transpositions tables… Do you have an explanation of why time believes so much (lol, Austria in English does not look at all like ‘Autriche’ in French, I thought it was a lost corner of America)

I have heard some ppl outside Europe mix us up with Australia, but France - hey, we have been fighting each other for centuries!

I took a quick look at Destructor 1.4.2 CompMode, please don't use listIndexOf() in inner loops / functions or on large lists, it is a performance killer with O(N) runtime complexity. Most of your CPU cycles are burnt there. There is always an alternative implementation. GoK only applies listIndexOf() at one place, not an inner loop, and the list is of size 4.

Frankly, I don’t even know where I used this or that… I would look at this, thank you for your advice! (Which I have no idea to apply), is there an alternative?

It's listIndexOf (list-indexOf or list-contains, e.g. In “Vérifier board 2 avec roi” for example), also indexOf (string-contains, e.g. in “Essayer de bouger et validation” and “”WEssayer de bouger 2“). If you are looking for the king index on your board, have a variable dedicated for that and update it when kings moves, but don't scan the board for the king each time. Scratch doesn't simply look for a value in a list (which would be bad enough), internally it does type conversions, tries to trail whitespaces, and more on each and every element comparison.

And the way to find these things is to use browser F12 dev tools on TurboWarp. E.g. on Chrome, press F12, click the record button in ”Performance" tab, then make move and let Destructor run a search, then stop recording, click Performance / Bottom Up, and sort the list of functions by total time. You can then click the .js source link and investigate each line of code on how much time is spent.

What I also noticed are busy loops, checking for user input. I suggest to turn all of that off, there are message handlers for what, e.g. when-clicked, when-key-pressed. Loops scanning for input will keep your CPU busy and slow down everything else.

Okay, I 'll see that

ArnoHu wrote:

ArnoHu wrote:

ArnoHu wrote:

GoK NNUE (10sec, black) vs. White Dove (P2), GoK wins: https://lichess.org/qanbjD8d#128

GoK NNUE (Medium, black) vs. Element (6+8), double-blunder by Element early on, loses in 14 moves: https://lichess.org/VF7AWBLw#28

White Dove (P3, white) with a strange queen blunder against GoK NNUE (Medium), loses in 39 moves: https://lichess.org/UFiV2iLs#78

ArnoHu wrote:

ababoin07 wrote:

ArnoH wrote:

ababoin07 wrote:

ArnoHu wrote:

ababoin07 wrote:

ArnoHu wrote:

ababoin07 wrote:

ArnoH wrote:

It is off-topic, but I have to say it. @Grijfland, Austria is 1:0 vs Netherlands, Possession 63% vs. 37% at minute 14

Just off topic XD XD is this for soccer?

Sure, what else? You are from France, no?

We won 3:2, and qualified for knockout rounds as 1st in the group, ahead of France and the Netherlands - incredible, everyone is going crazy here!

Yes, I am French and I live in France, and I conclude that you must be in Austria? I am not interested in football… Otherwise, during tests on Destructor, I added 1 to the maximum depth of search for quiescence, and the time went up to 200 seconds (3 minutes) despite the transpositions tables… Do you have an explanation of why time believes so much (lol, Austria in English does not look at all like ‘Autriche’ in French, I thought it was a lost corner of America)

I have heard some ppl outside Europe mix us up with Australia, but France - hey, we have been fighting each other for centuries!

I took a quick look at Destructor 1.4.2 CompMode, please don't use listIndexOf() in inner loops / functions or on large lists, it is a performance killer with O(N) runtime complexity. Most of your CPU cycles are burnt there. There is always an alternative implementation. GoK only applies listIndexOf() at one place, not an inner loop, and the list is of size 4.

Frankly, I don’t even know where I used this or that… I would look at this, thank you for your advice! (Which I have no idea to apply), is there an alternative?

It's listIndexOf (list-indexOf or list-contains, e.g. In “Vérifier board 2 avec roi” for example), also indexOf (string-contains, e.g. in “Essayer de bouger et validation” and “”WEssayer de bouger 2“). If you are looking for the king index on your board, have a variable dedicated for that and update it when kings moves, but don't scan the board for the king each time. Scratch doesn't simply look for a value in a list (which would be bad enough), internally it does type conversions, tries to trail whitespaces, and more on each and every element comparison.

And the way to find these things is to use browser F12 dev tools on TurboWarp. E.g. on Chrome, press F12, click the record button in ”Performance" tab, then make move and let Destructor run a search, then stop recording, click Performance / Bottom Up, and sort the list of functions by total time. You can then click the .js source link and investigate each line of code on how much time is spent.

What I also noticed are busy loops, checking for user input. I suggest to turn all of that off, there are message handlers for what, e.g. when-clicked, when-key-pressed. Loops scanning for input will keep your CPU busy and slow down everything else.

Okay, I 'll see that

OK, did you see the latest message at https://scratch.mit.edu/discuss/post/8029714/ ?

The question is also, why is a block invoked 50 million times within a 40sec search? I think you said it runs at 10k NPS? That would be 125 invocations per node…

ArnoHu wrote:

ababoin07 wrote:

ArnoH wrote:

ababoin07 wrote:

ArnoHu wrote:

ababoin07 wrote:

ArnoHu wrote:

ababoin07 wrote:

ArnoH wrote:

It is off-topic, but I have to say it. @Grijfland, Austria is 1:0 vs Netherlands, Possession 63% vs. 37% at minute 14

Just off topic XD XD is this for soccer?

Sure, what else? You are from France, no?

We won 3:2, and qualified for knockout rounds as 1st in the group, ahead of France and the Netherlands - incredible, everyone is going crazy here!

Yes, I am French and I live in France, and I conclude that you must be in Austria? I am not interested in football… Otherwise, during tests on Destructor, I added 1 to the maximum depth of search for quiescence, and the time went up to 200 seconds (3 minutes) despite the transpositions tables… Do you have an explanation of why time believes so much (lol, Austria in English does not look at all like ‘Autriche’ in French, I thought it was a lost corner of America)

I have heard some ppl outside Europe mix us up with Australia, but France - hey, we have been fighting each other for centuries!

I took a quick look at Destructor 1.4.2 CompMode, please don't use listIndexOf() in inner loops / functions or on large lists, it is a performance killer with O(N) runtime complexity. Most of your CPU cycles are burnt there. There is always an alternative implementation. GoK only applies listIndexOf() at one place, not an inner loop, and the list is of size 4.

Frankly, I don’t even know where I used this or that… I would look at this, thank you for your advice! (Which I have no idea to apply), is there an alternative?

It's listIndexOf (list-indexOf or list-contains, e.g. In “Vérifier board 2 avec roi” for example), also indexOf (string-contains, e.g. in “Essayer de bouger et validation” and “”WEssayer de bouger 2“). If you are looking for the king index on your board, have a variable dedicated for that and update it when kings moves, but don't scan the board for the king each time. Scratch doesn't simply look for a value in a list (which would be bad enough), internally it does type conversions, tries to trail whitespaces, and more on each and every element comparison.

And the way to find these things is to use browser F12 dev tools on TurboWarp. E.g. on Chrome, press F12, click the record button in ”Performance" tab, then make move and let Destructor run a search, then stop recording, click Performance / Bottom Up, and sort the list of functions by total time. You can then click the .js source link and investigate each line of code on how much time is spent.

What I also noticed are busy loops, checking for user input. I suggest to turn all of that off, there are message handlers for what, e.g. when-clicked, when-key-pressed. Loops scanning for input will keep your CPU busy and slow down everything else.

Okay, I 'll see that

OK, did you see the latest message at https://scratch.mit.edu/discuss/post/8029714/ ?

The question is also, why is a block invoked 50 million times within a 40sec search? I think you said it runs at 10k NPS? That would be 125 invocations per node…

AZURUS41 wrote:

https://lichess.org/hdBAo0nZ#0
Average game against GoK… Good opening, decide to sack a pawn for some reason cuz it thinks there is an attack, the attack is countered, and then the bot known for its offensive play lose to a sacrifice leading to checkmate.
Is there a way to modify the NNUE, even just a bit, so it stops giving away pawns ?
GoK NNUE TB Difficult btw

ArnoHu wrote:

ArnoHu wrote:

ArnoHu wrote:

GoK NNUE (10sec, black) vs. White Dove (P2), GoK wins: https://lichess.org/qanbjD8d#128

GoK NNUE (Medium, black) vs. Element (6+8), double-blunder by Element early on, loses in 14 moves: https://lichess.org/VF7AWBLw#28

White Dove (P3, white) with a strange queen blunder against GoK NNUE (Medium), loses in 39 moves: https://lichess.org/UFiV2iLs#78

birdracerthree wrote:

ArnoHu wrote:

ArnoHu wrote:

ArnoHu wrote:

GoK NNUE (10sec, black) vs. White Dove (P2), GoK wins: https://lichess.org/qanbjD8d#128

GoK NNUE (Medium, black) vs. Element (6+8), double-blunder by Element early on, loses in 14 moves: https://lichess.org/VF7AWBLw#28

White Dove (P3, white) with a strange queen blunder against GoK NNUE (Medium), loses in 39 moves: https://lichess.org/UFiV2iLs#78

Certified White Dove Moment©

Jokes aside, can you reproduce this move? I have seen White Dove play a poor move inexplicably in a tournament, so maybe it’s a type-1 Zobrist key collision

ababoin07 wrote:

ArnoHu wrote:

ababoin07 wrote:

ArnoH wrote:

ababoin07 wrote:

ArnoHu wrote:

ababoin07 wrote:

ArnoHu wrote:

ababoin07 wrote:

ArnoH wrote:

*snip*

Just off topic XD XD is this for soccer?

Sure, what else? You are from France, no?

We won 3:2, and qualified for knockout rounds as 1st in the group, ahead of France and the Netherlands - incredible, everyone is going crazy here!

*snip*

I have heard some ppl outside Europe mix us up with Australia, but France - hey, we have been fighting each other for centuries!

I took a quick look at Destructor 1.4.2 CompMode, please don't use listIndexOf() in inner loops / functions or on large lists, it is a performance killer with O(N) runtime complexity. Most of your CPU cycles are burnt there. There is always an alternative implementation. GoK only applies listIndexOf() at one place, not an inner loop, and the list is of size 4.

Frankly, I don’t even know where I used this or that… I would look at this, thank you for your advice! (Which I have no idea to apply), is there an alternative?

It's listIndexOf (list-indexOf or list-contains, e.g. In “Vérifier board 2 avec roi” for example), also indexOf (string-contains, e.g. in “Essayer de bouger et validation” and “”WEssayer de bouger 2“). If you are looking for the king index on your board, have a variable dedicated for that and update it when kings moves, but don't scan the board for the king each time. Scratch doesn't simply look for a value in a list (which would be bad enough), internally it does type conversions, tries to trail whitespaces, and more on each and every element comparison.

And the way to find these things is to use browser F12 dev tools on TurboWarp. E.g. on Chrome, press F12, click the record button in ”Performance" tab, then make move and let Destructor run a search, then stop recording, click Performance / Bottom Up, and sort the list of functions by total time. You can then click the .js source link and investigate each line of code on how much time is spent.

What I also noticed are busy loops, checking for user input. I suggest to turn all of that off, there are message handlers for what, e.g. when-clicked, when-key-pressed. Loops scanning for input will keep your CPU busy and slow down everything else.

Okay, I 'll see that

OK, did you see the latest message at https://scratch.mit.edu/discuss/post/8029714/ ?

The question is also, why is a block invoked 50 million times within a 40sec search? I think you said it runs at 10k NPS? That would be 125 invocations per node…

What block plz?

birdracerthree wrote:

ababoin07 wrote:

ArnoHu wrote:

ababoin07 wrote:

ArnoH wrote:

ababoin07 wrote:

ArnoHu wrote:

ababoin07 wrote:

ArnoHu wrote:

ababoin07 wrote:

ArnoH wrote:

*snip*

Just off topic XD XD is this for soccer?

Sure, what else? You are from France, no?

We won 3:2, and qualified for knockout rounds as 1st in the group, ahead of France and the Netherlands - incredible, everyone is going crazy here!

*snip*

I have heard some ppl outside Europe mix us up with Australia, but France - hey, we have been fighting each other for centuries!

I took a quick look at Destructor 1.4.2 CompMode, please don't use listIndexOf() in inner loops / functions or on large lists, it is a performance killer with O(N) runtime complexity. Most of your CPU cycles are burnt there. There is always an alternative implementation. GoK only applies listIndexOf() at one place, not an inner loop, and the list is of size 4.

Frankly, I don’t even know where I used this or that… I would look at this, thank you for your advice! (Which I have no idea to apply), is there an alternative?

It's listIndexOf (list-indexOf or list-contains, e.g. In “Vérifier board 2 avec roi” for example), also indexOf (string-contains, e.g. in “Essayer de bouger et validation” and “”WEssayer de bouger 2“). If you are looking for the king index on your board, have a variable dedicated for that and update it when kings moves, but don't scan the board for the king each time. Scratch doesn't simply look for a value in a list (which would be bad enough), internally it does type conversions, tries to trail whitespaces, and more on each and every element comparison.

And the way to find these things is to use browser F12 dev tools on TurboWarp. E.g. on Chrome, press F12, click the record button in ”Performance" tab, then make move and let Destructor run a search, then stop recording, click Performance / Bottom Up, and sort the list of functions by total time. You can then click the .js source link and investigate each line of code on how much time is spent.

What I also noticed are busy loops, checking for user input. I suggest to turn all of that off, there are message handlers for what, e.g. when-clicked, when-key-pressed. Loops scanning for input will keep your CPU busy and slow down everything else.

Okay, I 'll see that

OK, did you see the latest message at https://scratch.mit.edu/discuss/post/8029714/ ?

The question is also, why is a block invoked 50 million times within a 40sec search? I think you said it runs at 10k NPS? That would be 125 invocations per node…

What block plz?

Item#Of is the main block, list-contains is secondary. Although in the custom block Arno mentioned, it looks like the item#of block is only used for the king’s index on the board. Element does the same thing (although it checks if the king’s index variable isn’t already correct because the king tends to stay in the same place). I can’t profile right now because I’m on mobile, I can take a look when I can.

Arno, not everyone knows what “IndexOf” means. A while ago I was explaining why I needed it for piece-square tables, but I thought it meant the item-of block

S_P_A_R_T wrote:

birdracerthree wrote:

ArnoHu wrote:

ArnoHu wrote:

ArnoHu wrote:

GoK NNUE (10sec, black) vs. White Dove (P2), GoK wins: https://lichess.org/qanbjD8d#128

GoK NNUE (Medium, black) vs. Element (6+8), double-blunder by Element early on, loses in 14 moves: https://lichess.org/VF7AWBLw#28

White Dove (P3, white) with a strange queen blunder against GoK NNUE (Medium), loses in 39 moves: https://lichess.org/UFiV2iLs#78

Certified White Dove Moment©

Jokes aside, can you reproduce this move? I have seen White Dove play a poor move inexplicably in a tournament, so maybe it’s a type-1 Zobrist key collision

I personally cannot reproduce this. Maybe it is just a very very rare key collision, because on my system, the queen blunder is not considered even for a moment.