How to install Stockfish NNUE!

No. I was reffering to chess.com event. Leela vs Stockfish nnue.

Post #37

I am currently running an engine match between the latest version of Stockfish NNUE and S11. S11 is beating NNUE by quite a reasonable margin.

Then you just need to learn either one of

1.how to install engines properly

2. error bar in statistical sample

3. number of sample sizes etc.

But no one will take consideration of your results, as they had already tested 100,000+ game sample size. 

For example, if you make tournment between two identical clones of Stockfish 11 , you will see one engine will be beating another as well.

Maybe, maybe not. But no need to talk down to people without any relevant information to go on.

Not everyone uses it for bullet/blitz junk

What time controls are you using @neveraskmeforadraw ?

Really?

If stockfish developers' testing is junk, you will never see progress of Stockfish development. In fact they can test within 2 elo error bar.

40,000 games of what quality?

Quality of sample is as much, if not more, important as the size.

pretty sure you don't have it set up properly then...

well it depends on what you mean by setting it up properly!? I had the nnue function enabled in the settings. other than that it's basicaly default settings for both engines.

well, if one engine 100 elo points higher is losing to the other, i might suggest something's not right. pretty one-sided on mine...

You just need to know science, all stockfish patches are tested in ultra bullet pace ( so does the same as other majority of engines).

As majority  of patches had  +1 to +3 elo ( except from SF nnue patch that had +58 elo), developers need to test 20,000+ games minimum to know whether the new development is better or not.

Otherwise, if you test these patches in Long time control like 3 hour games for 10-20, you will never know the difference between 3600 rating vs  3602 rating.

If you test these games in 3h, for 40,000 games for 50,000 patches( SF had around 50,000 patches), it will probably take you 1 billion years .

Here is a link for you if you are intested in those statistics.

https://www.chessprogramming.org/Match_Statistics

• neveraskmeforadraw wrote:

  I am currently running an engine match between the latest version of Stockfish NNUE and S11. S11 is beating NNUE by quite a reasonable margin.

You are probably not getting NNUE installed properly or sample errors that I mentioned. Here is also official stockfish blog  how to use NNUE.

https://blog.stockfishchess.org/post/625828091343896577/introducing-nnue-evaluation

And the nerds are now partying!!!!

Nah, those are the hipsters - notice the straws in the coke cans!

I had tested SF NNUE vs SF 11, ! SF NNUE vs SF 11, +11 = 45 -4 ! (In 1+1 match) !

Nice topic, im really fan of chess Engines, and talking about NNs, why are SFNUE in different format??, could it be someday ported to Android??, also talking about Leela, i could say she is kinda bad in Android devices, maybe cuz of lack of good GPU, on my PC i got the DX12 version and it still loses to SF11 like 60% of games, anyone here has tested her on Fritz and maybe 2080 RTX??

• ---

  ---

  ElysiumKing wrote:

  Nice topic, im really fan of chess Engines, and talking about NNs, why are SFNUE in different format??, could it be someday ported to Android??, also talking about Leela, i could say she is kinda bad in Android devices, maybe cuz of lack of good GPU, on my PC i got the DX12 version and it still loses to SF11 like 60% of games, anyone here has tested her on Fritz and maybe 2080 RTX??

SF nnue become part of official stockfish  hybrid( no more split versions between traditional SF vs SFNNUE). So, when Stockfish 12 is released,  Peter Osterlund will do android version( Droidfish) as usual.

Regarding against Leela, SFnnue could perform well against Leela ( provided that you have hardware CPU vs GPU leela ratio= 1, or GPU speed = 1:1000 of CPU speed for 20x256 Lco net).

I dont how much better against Leela,  I havent follow up latest test results.

What if they tested the engines with the worst hardware? Would the better engine still win?

Thank you for your response!!, I'll be looking forward to new SF NNUE/Lc0 improvements. Also, I was not really aware the difference in strength due to different NNs in Lc0, I installed the new ones for Lc0 DX version on PC and she does way better vs SF10 , now I just want to find a way to use the same Nns on her in the "Acid Ape Chess app", cuz with the default Nns for android She doenst seem so strong lol, thank bro, have a nice day and let's have some gg's someday.