In what file I can find stockfish's evaluation?

[Uri Blass]

First candidate is of course evaluate.cpp but I find in the file the following:

Value nnue = smallNet ? networks.small.evaluate(pos, true, &nnueComplexity, psqtOnly)
: networks.big.evaluate(pos, true, &nnueComplexity, false);

I understand that basically I need to search networks.big.evaluate in case the net is not small but where can I find it?

[Guenther]

First candidate is of course evaluate.cpp but I find in the file the following:

Value nnue = smallNet ? networks.small.evaluate(pos, true, &nnueComplexity, psqtOnly)
: networks.big.evaluate(pos, true, &nnueComplexity, false);

I understand that basically I need to search networks.big.evaluate in case the net is not small but where can I find it?

I wonder how can it be so hard for you - well, actually I don't wonder anymore...

https://github.com/official-stockfish/S ... uate.h#L38 (and next line)

[Uri Blass]

First candidate is of course evaluate.cpp but I find in the file the following:

Value nnue = smallNet ? networks.small.evaluate(pos, true, &nnueComplexity, psqtOnly)
: networks.big.evaluate(pos, true, &nnueComplexity, false);

I understand that basically I need to search networks.big.evaluate in case the net is not small but where can I find it?

I wonder how can it be so hard for you - well, actually I don't wonder anymore...

https://github.com/official-stockfish/S ... uate.h#L38 (and next line)

Thanks
I understand that
nn-1ceb1ade0001.nnue should include the big network evaluation.
Where can I find this file.

I do not see it here
https://github.com/official-stockfish/S ... 9444d4/src

I looked for files with the end .nnue in the source and I do not see files with .nnue.

Even here I see no .nnue files

https://github.com/official-stockfish/S ... 4/src/nnue

going to layers or to features does not help me.

[Dann Corbit]

When you run the build script, it automatically downloads both the big and small network files.

[Guenther]

First candidate is of course evaluate.cpp but I find in the file the following:

Value nnue = smallNet ? networks.small.evaluate(pos, true, &nnueComplexity, psqtOnly)
: networks.big.evaluate(pos, true, &nnueComplexity, false);

I understand that basically I need to search networks.big.evaluate in case the net is not small but where can I find it?

I wonder how can it be so hard for you - well, actually I don't wonder anymore...

https://github.com/official-stockfish/S ... uate.h#L38 (and next line)

Thanks
I understand that
nn-1ceb1ade0001.nnue should include the big network evaluation.
Where can I find this file.

I do not see it here
https://github.com/official-stockfish/S ... 9444d4/src

I looked for files with the end .nnue in the source and I do not see files with .nnue.

Even here I see no .nnue files

https://github.com/official-stockfish/S ... 4/src/nnue

going to layers or to features does not help me.

Are these serious questions? Ever looked at a makefile? Those will be downloaded on the fly, while compiling.

https://tests.stockfishchess.org/nns
and/or
https://github.com/official-stockfish/networks

https://github.com/official-stockfish/S ... efile#L966

[Uri Blass]

First candidate is of course evaluate.cpp but I find in the file the following:

Value nnue = smallNet ? networks.small.evaluate(pos, true, &nnueComplexity, psqtOnly)
: networks.big.evaluate(pos, true, &nnueComplexity, false);

I understand that basically I need to search networks.big.evaluate in case the net is not small but where can I find it?

I wonder how can it be so hard for you - well, actually I don't wonder anymore...

https://github.com/official-stockfish/S ... uate.h#L38 (and next line)

Thanks
I understand that
nn-1ceb1ade0001.nnue should include the big network evaluation.
Where can I find this file.

I do not see it here
https://github.com/official-stockfish/S ... 9444d4/src

I looked for files with the end .nnue in the source and I do not see files with .nnue.

Even here I see no .nnue files

https://github.com/official-stockfish/S ... 4/src/nnue

going to layers or to features does not help me.

Are these serious questions? Ever looked at a makefile? Those will be downloaded on the fly, while compiling.

https://tests.stockfishchess.org/nns
and/or
https://github.com/official-stockfish/networks

https://github.com/official-stockfish/S ... efile#L966

I did not look at a makefile and did not need to do it in the past when I developed movei many years ago.

The questions are serious.
I think that people are allowed to ask questions that they do not know the answer for them and I do not see what is the problem with it.

It is not something that I could find by myself in a few minutes and I do not plan to spend many hours to try to find the answer by myself before asking a question.

[smatovic]

Idk if it helps, but Stockfish 16 removed the "classic, handcrafted evaluation" from source code, now 16.1 with small and big neural network.

--
Srdja

[Uri Blass]

Idk if it helps, but Stockfish 16 removed the "classic, handcrafted evaluation" from source code, now 16.1 with small and big neural network.

--
Srdja

Thanks but I know that stockfish removed the classic evaluation.

I wonder if every evaluation that humans can understand is called HCE (hand craft evaluation) even if people did optimizations to have better weights in the evaluation.


I read that Weiss is the best engine with HCE that still play in TCEC and I do not like the fact that somebody hopes that it is going to go down.
I basically prefer engines that humans can understand what they do and not engines that people tell me that humans cannot understand what they do.

[hgm]

My interest for computer chess has always been that I was curious what aspects of a position one would have to pay attention to, and how heavily, to make a good judgement on how well the players are doing. Computer programs seemed eminently suitable for that, as you can know and experiment with how they think, and then study teh consequences. But NNUE tell me nothing of the sort. NNUE is to HCE what alchemy is to chemistry. You might have a recipe that works, but no clue as to why. Which might be good if it was a recipe for making gold, but alas, it is a recipe for something that is completely useless in itself...

[smatovic]

IIRC team Deepmind used heatmaps to visualize stuff in AlphaGo's CNNs, and go manually through blunders in games, idk what people do in regard of NNUE as MLP neural network.

Acquisition of Chess Knowledge in AlphaZero
https://arxiv.org/pdf/2111.09259.pdf

--
Srdja