Stockfish engine, odd behaviour.

Are the moves that I entered manually correct? Move 22 was a large blunder. There was a checkmate.
UCI option name EvalFile type string default nn-ad9b42354671.nnue
Did you have the file nn-ad9b42354671.nnue in the same directory as the Stockfish engine?

@Toscani said in #21:
> Are the moves that I entered manually correct? Move 22 was a large blunder. There was a checkmate.
Do you mean 22...Kg2?

Lichess analysis board annotates this as a blunder, because the fishnet analysis identifies the forced checkmate after this move, but can't find the forced checkmate before this move. But there was already a forced checkmate before 22...Kg2, fishnet just didn't think deep enough to see it.

One way to verify that this is true is to check the tablebase (in lichess analysis, by clicking the little book icon). You can see that the tablebase announces that every possible move from Black on move 22 loses. Kg2 delays mate as long as possible (tied with Ng5). Looking at the tablebase, you can even see that the tablebase announces a forced win for White after 16...c2 (it doesn't announce it before this because the tablebase only goes up to seven pieces).

> Did you have the nn-ad9b42354671.nnue in the same directory of your stockfish?
I'm not sure what you mean by this because the neural net is included in the stockfish binary, it's not a separate file.

When I download the stockfish_15.1_win_x64_popcnt.exe from the website, I get the nnue from the test site. I don't recompile the one in the zipped folder that I had downloaded. I first type UCI to know the nnue and then download it from the tests stockfishchess nns
I used up my privileges for adding the link for today.

@Toscani said in #23:
> When I download the stockfish_15.1_win_x64_popcnt.exe from the website, I get the nnue from the test site. I don't recompile the one in the folder. I first type UCI to know the nnue and then download it from the tests stockfishchess nns

Are you trying to compile from source? I just downloaded the precompiled binary from stockfishchess.org, the AVX2 build for Windows. In the default UCI options, my neural net file is nn-ad9b42354671.nnue

Like i said, I don't often compile, but have done it in last year. I do the same, just press download to get stockfish, but the default nnue is not compiled when we download it and we unzip the files, I have never seen the nnue in the same directory as the program. It has it's own directory but it's not compiled so it's not usable until compiled. This is why I download it separately at
tests.stockfishchess.org/nns

@Toscani said in #25:
> I do the same, just press download to get stockfish, but the default nnue is not compiled when we download it and we unzip the files
When you download stockfish_15.1_win_x64_avx2.zip and unzip it, you get the stockfish binary, stockfish-windows-2022-x86-64-avx2.exe, plus sources. You don't need to do anything with the sources, you can simply point your UCI GUI to stockfish-windows-2022-x86-64-avx2.exe. The UCI Option "EvalFile" does not point to an actual file on your computer. You don't need to compile or generate any additional NN files.

If you don't have the NN file downloaded then you are using stockfish without the nnue file.
It can point to an actual nnue file if you tell it to. If not told, stockfish assumes the nnue file is in the default directory.
If it does not find it there, stockfish will still work without the nnue file.

@Toscani said in #27:
> If you don't have the NN file downloaded then you are using stockfish without the nnue file.
> It can point to an actual nnue file if you tell it to. If not told, stockfish assumes the nnue file is in the default directory.
> If it does not find it there, stockfish will still work without the nnue file.
That is not how it works. The NNUE weights are embedded in the .exe binary. There is no need for a separate external weights file.

So the default NNue is in the program and if we want to change the default then we need to do what I just said. Is that right?

@Toscani said in #29:
> So the default NNue is in the program and if we want to change the default then we need to do what I just said. Is that right?
Yes, if you want to use a different set of weights, you would want to point to a new weights file, but to just use the standard 15.1 production release weights, you don't need to do anything.