StockFish : best move not found due to MultiPV ?

Discussion of chess software programming and technical issues.

Moderators: hgm, Harvey Williamson, bob

Forum rules
This textbox is used to restore diagrams posted with the [d] tag before the upgrade.
Sesse
Posts: 168
Joined: Mon Apr 30, 2018 9:51 pm
Contact:

Re: StockFish : best move not found due to MultiPV ?

Post by Sesse » Wed Jul 18, 2018 11:02 pm

tissatussa wrote:
Wed Jul 18, 2018 5:45 pm
Clear, but HOW do all engines evaluate?
There's a ton of information on the chess programming wiki.
What's your comment on the screenshots in the linked ZIP?
Probably that if you want anyone to care about a screenshot, you should not pack it into a zip file. Second, that you are way off topic for the thread as far as I can see.

User avatar
tissatussa
Posts: 24
Joined: Sat Sep 24, 2016 2:13 am
Location: Netherlands
Full name: Roelof Berkepeis
Contact:

Re: StockFish : best move not found due to MultiPV ?

Post by tissatussa » Wed Jul 18, 2018 11:07 pm

this one is with classical music : https://www.youtube.com/watch?v=h2JcvR0tb2I
patterns and timing
-simple is not always best but best is always simple-

User avatar
tissatussa
Posts: 24
Joined: Sat Sep 24, 2016 2:13 am
Location: Netherlands
Full name: Roelof Berkepeis
Contact:

Re: StockFish : best move not found due to MultiPV ?

Post by tissatussa » Thu Jul 19, 2018 7:36 am

@Sesse - Why keep pointing to commonly used depth-search methods and evalution functions when those do not seem to be able to find the proper "best move" of my diagram? It's rather obvious for a human being this move has to be considered. So, better look for other approaches to evaluate a position.

I'd hoped someone comes up with a fresh evaluation idea. Don't be rigid by "here is how evaluations are done", because that doesn't seem to work in this position. Try to think out of the box and replace the existing methods, like Leela does in her way. I just put the video's here to get inspiration for that.

About the screenshots in the ZIP file : why does TalkChess limit the total size of uploaded images to 2 Mb? The screenshots are full screen. When you're intersted you could easily open the ZIP file, which i used as an alternative.
-simple is not always best but best is always simple-

User avatar
tissatussa
Posts: 24
Joined: Sat Sep 24, 2016 2:13 am
Location: Netherlands
Full name: Roelof Berkepeis
Contact:

Re: StockFish : best move not found due to MultiPV ?

Post by tissatussa » Thu Jul 19, 2018 11:47 am

so, regarding my FEN position, where i state b4! should be the best move, it seems that most engines judge at first not to consider this move to be in the top 20 of MultiPV. However a human can see the Bc5 is overstretched : when it captures the b4 pawn, the square g1 is no longer covered and White can play Rg1! and reach a good (?!) attacking position, although a pawn is lost .. How does the engine evaluate ? Does it only count material ? I guess modern engines do not. It's all about options, therefor all pieces should be on "good squares" .. i guess you know what i mean, as a chess player .. so, what is a "good square" ? As i stated, the asmFish engine finds b4! within 20 MultiPV, and is rather fast. It's calculation / evaluation / judging which move will be in the top # MultiPV / seems different.

How does an engine evaluate any position ? You can point me to the existing tree search methods and "the way evaluation works", but can we try to create alternative evaluation scores ? Eg.

amount of pawns
amount of double pawns
amount of squares a piece can reach in a concerning position

amount of squares a piece can reach in a concerning position, as a percentage of the max (according to our chess rules), for example:
max_square( piece=Knight, square=e5) is 8
max_square( piece=Knight, square=a1) is 2

regarding pawn structure : how many rows for each pawn to promote
regarding the pieces Bishop, Rook and Queen, which can all move a long range, we should consider all the squares in between the from-square and the to-square : some kind of "pressure" factor for each square should exist (occupied by some piece, the line is blocked).
etc.


Some engines introduce some King Safety factor .. is this alike ?
I think i understand the general coding of an engine, with (recursive) depth etc., but i never build one.

Many test FEN positions exist, but how can a programmer use the output to tune and code new parameters to create a (more) suitable evaluation ? Now, you can tell me about min-max etc. but what about alternative evaluation scores in the first place ..

Who can elaborate on that ?

[ some new patterns .. https://www.youtube.com/watch?v=crgEYMsg7sY ]
-simple is not always best but best is always simple-

brianr
Posts: 320
Joined: Thu Mar 09, 2006 2:01 pm

Re: StockFish : best move not found due to MultiPV ?

Post by brianr » Thu Jul 19, 2018 11:51 am


Sesse
Posts: 168
Joined: Mon Apr 30, 2018 9:51 pm
Contact:

Re: StockFish : best move not found due to MultiPV ?

Post by Sesse » Thu Jul 19, 2018 1:05 pm

tissatussa wrote:
Thu Jul 19, 2018 11:47 am
so, regarding my FEN position, where i state b4! should be the best move
You state that without any proof. It does not seem to be correct.
it seems that most engines judge at first not to consider this move to be in the top 20 of MultiPV.
I don't know why you think it seems so.
How does the engine evaluate ? Does it only count material ?
Why do you bother asking such questions if you don't like the answers?
How does an engine evaluate any position ? You can point me to the existing tree search methods and "the way evaluation works", but can we try to create alternative evaluation scores ? Eg.

amount of pawns
amount of double pawns
amount of squares a piece can reach in a concerning position
Engines already do this and much more.

You should read more and write less.

User avatar
tissatussa
Posts: 24
Joined: Sat Sep 24, 2016 2:13 am
Location: Netherlands
Full name: Roelof Berkepeis
Contact:

Re: StockFish : best move not found due to MultiPV ?

Post by tissatussa » Thu Jul 19, 2018 3:33 pm

@Sesse:
i wrote:

it seems that most engines judge at first not to consider this move to be in the top 20 of MultiPV.

you wrote:

I don't know why you think it seems so.

And I don't know why you are still not aware b4! is only evaluated when this move is in the top 20+ MultiPV (and depth is high enough). Try it, especially with my FEN position. When existing evaluation methods obviously fail, we should tune parameters to get a better evaluation, that's obvious i think. Never mind.
-simple is not always best but best is always simple-

Sesse
Posts: 168
Joined: Mon Apr 30, 2018 9:51 pm
Contact:

Re: StockFish : best move not found due to MultiPV ?

Post by Sesse » Thu Jul 19, 2018 8:20 pm

tissatussa wrote:
Thu Jul 19, 2018 3:33 pm
And I don't know why you are still not aware b4! is only evaluated when this move is in the top 20+ MultiPV (and depth is high enough).
I'm “not aware” of it because it's obviously not true. At MultiPV 5, it shows up already at d25… as the fourth best move.
When existing evaluation methods obviously fail
They don't “obviously fail”, because b4 is not the best move in the position. You can write it with an exclamation mark all you want, but there's no good reason to believe why it should be the best one. (Furthermore, if it really were the best move, you could just as well argue it's a search failure, not an evaluation failure.)

User avatar
tissatussa
Posts: 24
Joined: Sat Sep 24, 2016 2:13 am
Location: Netherlands
Full name: Roelof Berkepeis
Contact:

Re: StockFish : best move not found due to MultiPV ?

Post by tissatussa » Thu Jul 19, 2018 11:36 pm

At MultiPV 5, it shows up already at d25… as the fourth best move.
i think b4! should soon (after a small depth) be in the top # MultiPV list when alternative evaluation values are used / developed. Do all engines show b4! at d25 when MultiPV is 5 ? I guess their evaluation top # will be different. But the move b4! should be considered at a possible best move and so given a high priority for deeper search. Are you a chess player ? What is your ELO ? Mine is 1870 (in Holland). Did you see the videos ? They show that positional play gives certain (and most of the times, a very few) moves to continue the idea of harmony in the piece team - it's a war game of options, the execution comes later ..
They don't “obviously fail”, because b4 is not the best move in the position.
It should be in top # MultiPV from start.

I hope you get my point now. I think you're too rigid in following the existing, known brute force methods.
-simple is not always best but best is always simple-

Dann Corbit
Posts: 9314
Joined: Wed Mar 08, 2006 7:57 pm
Location: Redmond, WA USA
Contact:

Re: StockFish : best move not found due to MultiPV ?

Post by Dann Corbit » Fri Jul 20, 2018 12:43 am

I think it is unclear what the best move is. It is not certain that b4 is the best move. It is not certain that b4 is not the best move.

I guess that with optimal play, this position is drawn, but it is a rather complex position, so it is hard to be sure.
Taking ideas is not a vice, it is a virtue. We have another word for this. It is called learning.
But sharing ideas is an even greater virtue. We have another word for this. It is called teaching.

Post Reply