The next revolution in computer chess?

[Ovyron]

Yeah, all I can say is that the NNUE engines showcase a completely different playing style than Stockfish. I like it, while Stockfish remains at 0.00 and continues to play dully, NNUE will seek material imbalances that it likes.

Since NNUE started I have witnessed the most material imbalances I've seen in such a short time frame, most Stockfish-Stockfish games still have same material. Stockfish vs. NNUE games tend to reach positions where both disagree on what material imbalance is better (which is a subjective question because the position hasn't been played before!) and the reason NNUE outperforms Stockfish is that if NNUE is wrong it's just a draw, but if Stockfish is wrong, NNUE tends to win.

Why hasn't anybody asked noobpwnftw of ChessDB fame for his data? I think he has more than 10 billion positions analyzed with some 22 Depth or something, maybe training from that would be good.

Why hasn't anybody trained a NNUE based on Leela's eval? There's people without GPUs that haven't been able to use Leela fully, so it's great to use a CPU based NN, but we're basing it on eval we already have access to...

The revolution hasn't started yet, people are just getting dressed for it, things could get really wild soon, this could be the beginning of chess's killshot (remember my words when an entity that can't lose games appears.)

[cdani]

Why hasn't anybody asked noobpwnftw of ChessDB fame for his data? I think he has more than 10 billion positions analyzed with some 22 Depth or something, maybe training from that would be good.

Why hasn't anybody trained a NNUE based on Leela's eval? There's people without GPUs that haven't been able to use Leela fully, so it's great to use a CPU based NN, but we're basing it on eval we already have access to...

The two things are at least commented already or on work. More information on Discord.

[M ANSARI]

To be honest this all makes sense. I mean humans are just horrible at tactics and for sure will over look many moves as they are simply a tactical minefield. Just one misstep and it is an immediate loss. With engines complicated tactics are just simple parts of the game and so they can accept moves that make no sense to a human as a human needs things to crystalize within a few moves and can't comprehend the explosion of possible lines that need to be calculated. So it very reasonable that an evaluation function done by an engine will be better than a carefully tuned evaluation done by a human ... especially now that engines are dramatically stronger than the best humans.

This is most certainly an exciting time in computer chess history!

[corres]

I am inclined to believe so.
AB engines going NN.
I wrote a short page just to give the new technique more attention.
http://rebel13.nl/download/stockfish-nnue.html

So you think the most modern method to build a chess engine is the follows
1, Build a super strong AB-type chess engine like Stockfish is
2, Make the NN model of the evaluation of that super strong chess engine using for eg. NNUE method to get the most modern chess engine.
And what is the benefit of this too complicated, time and energy consuming course?
For a chess engine user it is more simpler to get a PC with 32 physical cores CPU and dual RTX 2080 Ti, I think.

[matejst]

Yeah, all I can say is that the NNUE engines showcase a completely different playing style than Stockfish. I like it, while Stockfish remains at 0.00 and continues to play dully, NNUE will seek material imbalances that it likes.

I noted the same thing in my testing at depth 18. The sacrifice of a quality, or a queen for compensation occurred very often. It could be just by chance, but Pleomati's 96 net was the net with the most aggressive game, although a bit weaker than the other nets.

For the rest of your points, Uly, you are simply wrong. This project is at its beginning. Creating good nets requires a lot of time and knowledge, and not many can do it now. Nonetheless, Henk (Raphexon), Pleomati, Dietrich Kappe, Sergio Vieri, Josh have already created good nets, and continue their interesting work. The engine has been cleaned of bugs and improved, and there are binaries for most CPUs now.

Finally, Ed is right: this project opens new paths to computerchess, and some of these paths can use and perpetuate all the work done years ago. We shall still wait a bit to see which paths.

[Rebel]

I am inclined to believe so.
AB engines going NN.
I wrote a short page just to give the new technique more attention.
http://rebel13.nl/download/stockfish-nnue.html

So you think the most modern method to build a chess engine is the follows
1, Build a super strong AB-type chess engine like Stockfish is
2, Make the NN model of the evaluation of that super strong chess engine using for eg. NNUE method to get the most modern chess engine.
And what is the benefit of this too complicated, time and energy consuming course?
For a chess engine user it is more simpler to get a PC with 32 physical cores CPU and dual RTX 2080 Ti, I think.

Stockfish NNUE on 32 physical cores will still beat Stockfish 11 on 32 physical cores, so I don't understand your point.

NNUE engines don't need expensive GPU cards.

[nabildanial]

As it seems the human programmer gets more and more pruned away.

Time to strike back.

I think the picture is more nuanced. Right now a hand tuned eval function at some search depth is approximated by the nnue. More data == better approximation. Up until now only stockfish’s eval has been used, most commonly at depth 8. But the training software will accept text data and convert it into its training format. So it’s possible to generate training data with other engines.

I’ve been working on generating nets with other engines and will be releasing some of them shortly. One thing that’s occurred to me is that old, slow engines with good evals might benefit quite a bit from conversion to nnue. The new competition might be finding better evals for training a nnue. So, dust off all the brilliant ideas you shelved because they were impractical and would reduce your nps by an order of magnitude. Their time may have come.

I believe Naum is one of the better chess engines out in terms of positional evaluation. The only thing that holds it back is it's speed. Can you please do a tutorial in how to generate an NNUE net out of other engines eval? I've been following all the tutorials on Discord but none of them mentioned about using other engines as the teacher engine.

[glennsamuel32]

Why hasn't anybody asked noobpwnftw of ChessDB fame for his data? I think he has more than 10 billion positions analyzed with some 22 Depth or something, maybe training from that would be good.

He already has made it available (depth 24 data)...

Posted by Aloril:
From TCEC Twtich chat earlier: <noobpwnftw>
A full database snapshot which contains about 10 billion unique positions as of 2020-07-08 is available at:

ftp://ftp.chessdb.cn/pub/chessdb/

[Patrice Duhamel]

As it seems the human programmer gets more and more pruned away.

Time to strike back.

Can we learn something from NN engines and improve AB engines with it ?

[corres]

...
NNUE engines don't need expensive GPU cards.

This is the only one real benefit.