Andscacs nnue 0.1

[connor_mcmonigle]

If by "the evaluation function" you mean the neural net evaluation bit, you are wrong. He used the nn code from Daniel Shawul (scorpio) which allows such things. If you think he grabbed the SF evaluation in some way, you are completely, utterly, insanely incorrect.

The SF NN eval (by the way) was written by a brilliant fellow for calculations in Shogi. Perhaps you were referring to the SF team, who took that eval (with the author's permission and help) and pasted it on to SF. Because they most certainly did not write that code.

I am not utterly, insanely, incorrect and I do think he grabbed the SF evaluation...
- This "nn code" from Daniel Shawul is the NNUE evaluation code from CFish (a C rewrite of the inference code from Stockfish written by Ronald) packaged into an easy to use library
- The Andscacs author released this version with a copy of the weights directly from Stockfish

Therefore, the static eval of "Andscacs nnue 0.1" and Stockfish will exactly match. The author didn't take ideas, he took other people's code and fully admits to doing this, perhaps not realizing that doing so was in violation of GPL.

There's a big difference between this and a Shogi engine developer contributing to the Stockfish project from my perspective.

[Dann Corbit]

If by "the evaluation function" you mean the neural net evaluation bit, you are wrong. He used the nn code from Daniel Shawul (scorpio) which allows such things. If you think he grabbed the SF evaluation in some way, you are completely, utterly, insanely incorrect.

The SF NN eval (by the way) was written by a brilliant fellow for calculations in Shogi. Perhaps you were referring to the SF team, who took that eval (with the author's permission and help) and pasted it on to SF. Because they most certainly did not write that code.

I am not utterly, insanely, incorrect and I do think he grabbed the SF evaluation...
- This "nn code" from Daniel Shawul is the NNUE evaluation code from CFish (a C rewrite of the inference code from Stockfish written by Ronald) packaged into an easy to use library
- The Andscacs author released this version with a copy of the weights directly from Stockfish

Therefore, the static eval of "Andscacs nnue 0.1" and Stockfish will exactly match. The author didn't take ideas, he took other people's code and fully admits to doing this, perhaps not realizing that doing so was in violation of GPL.

There's a big difference between this and a Shogi engine developer contributing to the Stockfish project from my perspective.

This is a restatement of my entire argument in a nutshell.
Thank you for that.

[connor_mcmonigle]

This is a restatement of my entire argument in a nutshell.
Thank you for that.

Hmm. I'd be curious as to how you define "grabbing the evaluation function" in that case. If your engine's static eval exactly matches another preexisting engine's static eval, you've cloned the evaluation function...

[Dann Corbit]

You are right. Those SF dudes are thieving monsters
We should put them all in jail, after giving them a good flail.
You know, iron balls and chunks of bone in the whip.

[connor_mcmonigle]

You are right. Those SF dudes are thieving monsters
We should put them all in jail, after giving them a good flail.
You know, iron balls and chunks of bone in the whip.

strawman

I'm arguing two points:
- Engines with cloned evaluation functions (see my previous comment for a sufficient condition for this quality) shouldn't appear on rating lists.
- This Andscacs release is a GPL violation.

[Dann Corbit]

Well then, we should strike SF from all the rating lists.
The evaluation of SF is cloned from a Shogi program
Certainly in all the places that it matters (the quiet positions, which is the vast majority of chess)
Or do you deny this?

Yes, I agree, Let's strike SF from all the chess rating lists.
Dirty, nasty cloner

[connor_mcmonigle]

Well then, we should strike SF from all the rating lists.
The evaluation of SF is cloned from a Shogi program
Certainly in all the places that it matters (the quiet positions, which is the vast majority of chess)
Or do you deny this?

Yes, I agree, Let's strike SF from all the chess rating lists.
Dirty, nasty cloner

The static evaluation of Stockfish does not exactly match any Shogi program in existence (How could it?). Furthermore, the author of this NNUE code contributed it to the Stockfish project.

Taking the "NNUE idea" to use in an engine is perfectly acceptable in my opinion, but taking the idea is very different from replacing your engine's evaluation function with something functionally identical to that found in a preexisting engine.

(I'm certainly not claiming Andscac's author is a "dirty nasty cloner")

[Dann Corbit]

Well then, we should strike SF from all the rating lists.
The evaluation of SF is cloned from a Shogi program
Certainly in all the places that it matters (the quiet positions, which is the vast majority of chess)
Or do you deny this?

Yes, I agree, Let's strike SF from all the chess rating lists.
Dirty, nasty cloner

The static evaluation of Stockfish does not exactly match any Shogi program in existence (How could it?). Furthermore, the author of this NNUE code contributed it to the Stockfish project.
>>
Um,,, how could it?
The algorithm for evaluation is CUT AND PASTE.
<<

Taking the "NNUE idea" to use in an engine is perfectly acceptable in my opinion, but taking the idea is very different from replacing your engine's evaluation function with something functionally identical to that found in a preexisting engine.
>>
Certainly it is perfectly acceptable if SF copies something, but god forbid if someone else should do the same thing, which makes them criminals.
Once SF does it, it belongs to them. That's how it works now.
<<
(I'm certainly not claiming Andscac's author is a "dirty nasty cloner")

Of course you are.
So present your evidence and show that he did not use his GPL version of Andscacs

[connor_mcmonigle]

Well then, we should strike SF from all the rating lists.
The evaluation of SF is cloned from a Shogi program
Certainly in all the places that it matters (the quiet positions, which is the vast majority of chess)
Or do you deny this?

Yes, I agree, Let's strike SF from all the chess rating lists.
Dirty, nasty cloner

The static evaluation of Stockfish does not exactly match any Shogi program in existence (How could it?). Furthermore, the author of this NNUE code contributed it to the Stockfish project.
>>
Um,,, how could it?
The algorithm for evaluation is CUT AND PASTE.
<<

Taking the "NNUE idea" to use in an engine is perfectly acceptable in my opinion, but taking the idea is very different from replacing your engine's evaluation function with something functionally identical to that found in a preexisting engine.
>>
Certainly it is perfectly acceptable if SF copies something, but god forbid if someone else should do the same thing, which makes them criminals.
Once SF does it, it belongs to them. That's how it works now.
<<
(I'm certainly not claiming Andscac's author is a "dirty nasty cloner")

Of course you are.
So present your evidence and show that he did not use his GPL version of Andscacs

1.
I don't think you're understanding. This version of Andscac's evaluation function with the provided weights is functionally identical to Stockfish's evaluation function.

Here's what I mean by "functionally identical"

Let P be an arbitrary chess position.
If we were to call this version of Andscac's evaluation function on P we'd get some integer x.
If we were to call Stockfish's evaluation function on P we'd also get x.

We can't even use a Shogi NNUE network to evaluate a chess position...

2.
Andscacs is closed source. If the author wishes to keep it that way, he can't use GPL'd code.

[Dann Corbit]

1) Did you check with highly imbalanced positions?
Because when the position is not lopsided, both programs use NNUE.
2) If you feel that the program is a clone, you should ask for the source. Only after the author refuses do you have any cause for complaint.

The computer chess community is entirely peopled with hypocrites (no doubt, including me).