this will be the merge of a lifetime : SF 80 Elo+

[JohnS]

"The idea is to use NNUE only on quite balanced material positions."
This was based on the fact that SF-NNUE searches much more slowly than classic SF on most hardware. But if one has fast hardware, so that SF-NNUE runs fast enough, then why not always use SF-NNUE?

How is stockfish defining 'quite balanced material positions', thanks.

https://github.com/official-stockfish/S ... 5aac9e322f

Thanks, does NNUEThreshold = Value(500) mean 5 pawns, 0.5 pawns or something else.

[Leo]

What does NNUE stand for?

[Leo]

This is the breakthrough I've been predicting for years! It finally happened, and I'm just glad chess looks very healthy after it (draw rates have gone down! Some lines are proving to be playable again!)

There's still so much room for experimentation. And I suspect that trying and training new approaches is quite a bit faster on a 20M NNUE net compared to a Leela one...

Dream situation for computer chess.

I agree, a Stockfish 12 release could be premature. These are new unexplored terrains, the code is nowhere near being optimized, the nets are nowhere near properly tested, imagine if SF 12 is released and one month later there's another 50 elo jump, we want a SF 12 that lasts for long!

This makes Alpha Zero look like peanuts! Bring back Deep Mind!

A lot of people considered the way Alpha Zero defeated SF flawed including myself. The one thing Alpha Zero did was to spur the computer chess society to develop its own learning network. I have always thought it sort of a waste to use all the power and resources having engines play and when the game was over the engine learned nothing.

[Leo]

. But just say there is a God, then atheist gets to spend his eternal years in hell and the rest of us are in the Glory of God for eternal time - what would a risk averse person do?

It's the well known : "Pascal's wager" https://en.wikipedia.org/wiki/Pascal%27s_wager

My variants are :
"If there are 1000 gods but you're not sure, what would a risk to believe in these 1000 gods ?"
"If there are 1 million gods but you're not sure, what would a risk to believe in these 1 million gods ?"
"If the real god is the Flying Spaghetti Monster, what would a risk to believe in the Flying Spaghetti Monster god ?"
God has always been a nonsense a will stay a nonsense forever.

You are spiritually dead. Psalm 14:1 The fool says in his heart, "There is no God."

[h1a8]

. But just say there is a God, then atheist gets to spend his eternal years in hell and the rest of us are in the Glory of God for eternal time - what would a risk averse person do?

It's the well known : "Pascal's wager" https://en.wikipedia.org/wiki/Pascal%27s_wager

My variants are :
"If there are 1000 gods but you're not sure, what would a risk to believe in these 1000 gods ?"
"If there are 1 million gods but you're not sure, what would a risk to believe in these 1 million gods ?"
"If the real god is the Flying Spaghetti Monster, what would a risk to believe in the Flying Spaghetti Monster god ?"
God has always been a nonsense a will stay a nonsense forever.

The only God where Hell is a possibility is the Christian God. Therefore, the pascal wager is all about either believing in that God vs believing in other God's/no God.

If there was another possible God that has Hell as an eternal punishment then chances are it is the Christian God that's is the true one (since one would have to make up a God for the sake of shaking the wager).

[Sven]

"The idea is to use NNUE only on quite balanced material positions."
This was based on the fact that SF-NNUE searches much more slowly than classic SF on most hardware. But if one has fast hardware, so that SF-NNUE runs fast enough, then why not always use SF-NNUE?

How is stockfish defining 'quite balanced material positions', thanks.

https://github.com/official-stockfish/S ... 5aac9e322f

Thanks, does NNUEThreshold = Value(500) mean 5 pawns, 0.5 pawns or something else.

Actually 2.5 pawns, since the formula also uses a fixed score of 200 for pawns (usually 256 in SF).

[zullil]

"The idea is to use NNUE only on quite balanced material positions."
This was based on the fact that SF-NNUE searches much more slowly than classic SF on most hardware. But if one has fast hardware, so that SF-NNUE runs fast enough, then why not always use SF-NNUE?

How is stockfish defining 'quite balanced material positions', thanks.

https://github.com/official-stockfish/S ... 5aac9e322f

Thanks, does NNUEThreshold = Value(500) mean 5 pawns, 0.5 pawns or something else.

Actually 2.5 pawns, since the formula also uses a fixed score of 200 for pawns (usually 256 in SF).

Yes, but things have already changed due to new patches. Wish someone would maintain a Stockfish+NNUEonly fork.

[JohnS]

"The idea is to use NNUE only on quite balanced material positions."
This was based on the fact that SF-NNUE searches much more slowly than classic SF on most hardware. But if one has fast hardware, so that SF-NNUE runs fast enough, then why not always use SF-NNUE?

How is stockfish defining 'quite balanced material positions', thanks.

https://github.com/official-stockfish/S ... 5aac9e322f

Thanks, does NNUEThreshold = Value(500) mean 5 pawns, 0.5 pawns or something else.

Actually 2.5 pawns, since the formula also uses a fixed score of 200 for pawns (usually 256 in SF).

Thanks for clearing that up Sven.

[Ovyron]

Yes, but things have already changed due to new patches. Wish someone would maintain a Stockfish+NNUEonly fork.

An "Use Classical Evaluation" tick box is still a possibility, the... huh... *air quotes* problem *air quotes* is that there's currently a big storm of good patches to improve Stockfish dev (it's expected to win another +20 elo by monday ) that nobody knows how the code will look like, and it's possible the hybrid patch works completely differently very soon. So many ideas and the one with the most elo will prevail.

No, we don't have a gold rush anymore, it's an elo rush!

If one week from now Stockfish-dev will find the best moves in all positions where speed is required (using classical eval) and will find the best moves in all positions where great eval is required (using NNUE eval) then the lack of the box will make sense, because there'll not be any scenario where NNUE is better than hybrid.

[zullil]

Yes, but things have already changed due to new patches. Wish someone would maintain a Stockfish+NNUEonly fork.

An "Use Classical Evaluation" tick box is still a possibility, the... huh... *air quotes* problem *air quotes* is that there's currently a big storm of good patches to improve Stockfish dev (it's expected to win another +20 elo by monday ) that nobody knows how the code will look like, and it's possible the hybrid patch works completely differently very soon. So many ideas and the one with the most elo will prevail.

No, we don't have a gold rush anymore, it's an elo rush!

If one week from now Stockfish-dev will find the best moves in all positions where speed is required (using classical eval) and will find the best moves in all positions where great eval is required (using NNUE eval) then the lack of the box will make sense, because there'll not be any scenario where NNUE is better than hybrid.

I don't care about maximizing Stockfish's self-play Elo at fishtest time controls. I am interested in seeing what Stockfish with pure NNUE evaluation can do. Yes, I can modify source code before compiling so that "Use NNUE" means "Use NNUE only". But that's going to get old very quickly.