Stockfish evaluation

Discussion of chess software programming and technical issues.

Moderators: bob, hgm, Harvey Williamson

Forum rules
This textbox is used to restore diagrams posted with the [d] tag before the upgrade.
Post Reply
Karlo Bala
Posts: 312
Joined: Wed Mar 22, 2006 9:17 am
Location: Novi Sad, Serbia

Stockfish evaluation

Post by Karlo Bala » Tue Jul 28, 2020 7:06 pm

After a while, I'm playing with Stockfish. The thing that is strange to me is that in equal positions Stockfish always think that stands better, even with reversed colors.

Does new Stockfish use asymmetric search or evaluation?
Best Regards,
Karlo Balla Jr.

Ras
Posts: 1403
Joined: Tue Aug 30, 2016 6:19 pm
Full name: Rasmus Althoff
Contact:

Re: Stockfish evaluation

Post by Ras » Tue Jul 28, 2020 7:13 pm

Karlo Bala wrote:
Tue Jul 28, 2020 7:06 pm
The thing that is strange to me is that in equal positions Stockfish always think that stands better, even with reversed colors.
Did you set the contempt parameter to 0?
Rasmus Althoff
https://www.ct800.net

Karlo Bala
Posts: 312
Joined: Wed Mar 22, 2006 9:17 am
Location: Novi Sad, Serbia

Re: Stockfish evaluation

Post by Karlo Bala » Tue Jul 28, 2020 7:47 pm

Ras wrote:
Tue Jul 28, 2020 7:13 pm
Karlo Bala wrote:
Tue Jul 28, 2020 7:06 pm
The thing that is strange to me is that in equal positions Stockfish always think that stands better, even with reversed colors.
Did you set the contempt parameter to 0?
You're right. At first sight, I simply didn't notice the contempt option. Somehow it is barely visible :)
Thank you!
Best Regards,
Karlo Balla Jr.

syzygy
Posts: 4601
Joined: Tue Feb 28, 2012 10:56 pm

Re: Stockfish evaluation

Post by syzygy » Thu Jul 30, 2020 2:41 pm

Karlo Bala wrote:
Tue Jul 28, 2020 7:06 pm
After a while, I'm playing with Stockfish. The thing that is strange to me is that in equal positions Stockfish always think that stands better, even with reversed colors.
To avoid this problem while analyising a position (Stockfish will think whoever has the move is in the lead), you can set "Analysis Contempt" to "Off". (The default "Both" setting makes little sense.)

User avatar
Ovyron
Posts: 4365
Joined: Tue Jul 03, 2007 2:30 am

Re: Stockfish evaluation

Post by Ovyron » Fri Jul 31, 2020 2:25 am

What I've found is that this is not a problem, advantage can be gained with a new paradigm where you keep two different evals, one for white and one for black, so both colors try to maximize their own scores, and there's positions to be avoided by both colors (like the Marshall-counter attack) or favorable to both colors (specially against weaker opposition, where you could beat them from both sides.)

NNUE broke all those paradigms, though (if both sides like their own position, one of them is wrong, says the net...)

syzygy
Posts: 4601
Joined: Tue Feb 28, 2012 10:56 pm

Re: Stockfish evaluation

Post by syzygy » Sat Aug 01, 2020 12:34 am

Ovyron wrote:
Fri Jul 31, 2020 2:25 am
What I've found is that this is not a problem, advantage can be gained with a new paradigm where you keep two different evals, one for white and one for black, so both colors try to maximize their own scores,
But alpha-beta does not work like that.
Both colors are always trying to maximise/minimise the same evaluations, which however suddenly change when the other gets to move in the root position (if you use "Both").

But if you're happy with the placebo effect, then never mind me.

User avatar
Ovyron
Posts: 4365
Joined: Tue Jul 03, 2007 2:30 am

Re: Stockfish evaluation

Post by Ovyron » Sat Aug 01, 2020 2:08 am

syzygy wrote:
Sat Aug 01, 2020 12:34 am
But alpha-beta does not work like that.
Both colors are always trying to maximise/minimise the same evaluations, which however suddenly change when the other gets to move in the root position (if you use "Both").

But if you're happy with the placebo effect, then never mind me.
I'm talking about analyzing positions for games against other people that are not using this paradigm, and getting advantageous positions and better analysis quality despite having worse hardware and not analyzing as many positions (or for much shorter depths; like analyzing at depth 32 all positions while opponent goes to depths of 60).

What I've said is that people that just analyze at home against themselves are fooling themselves into thinking they've found something good, and the only way to know is to go out there and play with someone with a better a analysis method and get beaten badly, discovering the analysis method is garbage. Like the unassisted alpha-beta analysis of engines which makes it waste more time with more analysis (the time wasted grows exponentially.) I'll not even say that this new paradigm is good, it's just that it's very easy to improve over unassisted alpha-beta.

Question: what's the highest Stockfish depth that you can beat? (this means that you get all the time in the world and takebacks and whatever you require to beat this depth) - the higher the depth you can beat, the better your analysis method, because that depth will rely on alphabeta.

Post Reply