Komodo Dragon vs. GM Perelshteyn, knight odds match.

Discussion of anything and everything relating to chess playing software and machines.

Moderator: Ras

Chessqueen
Posts: 5685
Joined: Wed Sep 05, 2018 2:16 am
Location: Moving
Full name: Jorge Picado

Re: Komodo Dragon vs. GM Perelshteyn, knight odds match.

Post by Chessqueen »

lkaufman wrote: Mon Oct 25, 2021 5:32 pm
Chessqueen wrote: Mon Oct 25, 2021 5:03 pm
Chessqueen wrote: Mon Oct 25, 2021 4:22 pm
Chessqueen wrote: Sun Oct 24, 2021 10:47 am I have never seen any GM playing like Komodo a complete open position without castling its King and going after its opponent


You don't state the time controls, was this with the simulated match conditions (3' + 2" for Dragon, 15' + 10" for Gaviota)? If so, then I agree, Gaviota 0.84 32 bit is stronger in Rapid than a 2500 human GM. My latest studies suggest the following conversion formula for estimating human FIDE elo in Rapid (15' + 10") play from CCRL Rapid ratings: human FIDE = 2/3 (CCRL) + 1000. So with Gaviota at 2570 CCRL Rapid, this gives 2713 as the estimated human FIDE rating that should be needed to score 50% vs. Gaviota 0.84 32 bit on one core of modern i7 at Rapid. That's why I suggested having both engines play with 3' + 2"; that would reduce the estimated strength of Gaviota to a level much closer to human 2500 (still above it, but not way above it). Regarding hardware, my latest simulations suggest that using more than 16 threads for this type of match is actually counterproductive, so I'll probably limit Dragon to 16 threads unless I get new data. Regarding opening, it is well known that with b1 missing, 1e4 is dubious due to 1...d5! It's very difficult for White to get any opening advantage (ignoring the missing piece) when giving b1 odds. Human GMs will at least work out the best initial 3 or 4 moves against reasonable opening tries by White at knight odds, that's one reason they have a better chance than similar strength engines with no book.

Komodo rules!

The time control was 3'+2" for the last 4 games. Mr. Kaufman, I have a question, is a Knight Plus a pawn a bigger handicap than a Rook Odds, With Komodo Dragon playing with White having the first move. ?
The "f" pawn almost compensates for the Exchange (rook for knight), but you would have to flip the colors to make it about even. Giving White to the side with the rook gives him a clear advantage, as shown by the evals, but it is less than a winning advantage, so in a long match White will win way more than Black wins, but Black will draw more than White wins, I expect. We know from past matches that Exchange Odds (Ra1 for Nb8) is more than f2 odds but slightly less than f7 odds.

This is what I meant, I did NOT let it finish, but it seems to be that they are equal handicaps, But we probably need 1000 games to prove that this is not an equal Odds :roll:

Chessqueen
Posts: 5685
Joined: Wed Sep 05, 2018 2:16 am
Location: Moving
Full name: Jorge Picado

Re: Komodo Dragon vs. GM Perelshteyn, knight odds match.

Post by Chessqueen »

I see what you meant I just finished playing 10 games and it ended 5 to 5, so for two players that do not want to memorized openings this setup would be perfect, even better than Non castling :roll:

Chessqueen
Posts: 5685
Joined: Wed Sep 05, 2018 2:16 am
Location: Moving
Full name: Jorge Picado

Re: Komodo Dragon vs. GM Perelshteyn, knight odds match.

Post by Chessqueen »

Here is a game against Stockfish 14

Chessqueen
Posts: 5685
Joined: Wed Sep 05, 2018 2:16 am
Location: Moving
Full name: Jorge Picado

Re: Komodo Dragon vs. GM Perelshteyn, knight odds match.

Post by Chessqueen »

[quote=Chessqueen post_id=909790 time=1635193127 user_id=10732]
Here is a game against Stockfish 14

Sorry, I had the Contempt preset to 100 the previous game against Stockfish14 does not count, here is a replay

Chessqueen
Posts: 5685
Joined: Wed Sep 05, 2018 2:16 am
Location: Moving
Full name: Jorge Picado

Re: Komodo Dragon vs. GM Perelshteyn, knight odds match.

Post by Chessqueen »

For people who do NOT want to memorize chess Opening this could be the perfect setup :roll:

lkaufman
Posts: 6231
Joined: Sun Jan 10, 2010 6:15 am
Location: Maryland USA
Full name: Larry Kaufman

Re: Komodo Dragon vs. GM Perelshteyn, knight odds match.

Post by lkaufman »

Chessqueen wrote: Tue Oct 26, 2021 12:00 am
Chessqueen wrote: Mon Oct 25, 2021 10:18 pm Here is a game against Stockfish 14

Sorry, I had the Contempt preset to 100 the previous game against Stockfish14 does not count, here is a replay

Yes, this starting position is probably pretty balanced. But it won't solve either of the two big problems of computer chess: 1: too many draws -- if the starting position is roughly balanced, the draw percentage at top level on good hardware will approach 100% 2. No variety: if we specify some suitably unbalanced opening position, like Black can't castle short or White is missing c2 pawn, we'll get roughly equal wins and draws at this level, but no variety, no way to test directly and engines will learn the best moves even if we randomize play a bit. We need a solution that combines the variety of FRC with enough advantage to be near the win/draw line, in a way that still feels like normal chess as much as possible. I don't have the answer yet.
Komodo rules!
User avatar
AdminX
Posts: 6363
Joined: Mon Mar 13, 2006 2:34 pm
Location: Acworth, GA

Re: Komodo Dragon vs. GM Perelshteyn, knight odds match.

Post by AdminX »

Revisiting my thoughts on this match, my opinion has now changed given that the match does not start until Nov 6th and that the GM does have a version of Komodo to practice with, I now think he should have good chances to win the match. I don't believe I gave enough thought to this little detail.
"Good decisions come from experience, and experience comes from bad decisions."
__________________________________________________________________
Ted Summers
Chessqueen
Posts: 5685
Joined: Wed Sep 05, 2018 2:16 am
Location: Moving
Full name: Jorge Picado

Re: Komodo Dragon vs. GM Perelshteyn, knight odds match.

Post by Chessqueen »

AdminX wrote: Tue Oct 26, 2021 11:59 am Revisiting my thoughts on this match, my opinion has now changed given that the match does not start until Nov 6th and that the GM does have a version of Komodo to practice with, I now think he should have good chances to win the match. I don't believe I gave enough thought to this little detail.
Since he is practicing with Komodo Dragon 2.5.1, he only need ChrisW opening book and Mr. Kaufman 32 core Threadripper, but I strongly recommend Mr. Kaufman to set his Komodo Dragon xxx.x to play at (5'+2") instead of (3' + 2") :roll:
lkaufman
Posts: 6231
Joined: Sun Jan 10, 2010 6:15 am
Location: Maryland USA
Full name: Larry Kaufman

Re: Komodo Dragon vs. GM Perelshteyn, knight odds match.

Post by lkaufman »

Chessqueen wrote: Tue Oct 26, 2021 3:15 pm
AdminX wrote: Tue Oct 26, 2021 11:59 am Revisiting my thoughts on this match, my opinion has now changed given that the match does not start until Nov 6th and that the GM does have a version of Komodo to practice with, I now think he should have good chances to win the match. I don't believe I gave enough thought to this little detail.
Since he is practicing with Komodo Dragon 2.5.1, he only need ChrisW opening book and Mr. Kaufman 32 core Threadripper, but I strongly recommend Mr. Kaufman to set his Komodo Dragon xxx.x to play at (5'+2") instead of (3' + 2") :roll:
I've run enough simulations to say that using more than 16 threads or more than 3' +2" (with 16 threads) is totally useless for knight odds. Imagine if we had a 32 man tablebase (from a million years in the future!). How would that help us in a knight odds match? Sure, if we got down to less than a pawn deficit in a game it might see a weird draw that normal search can't find, but against a knight-odds human opponent there is no point in assuming that he will play perfectly, it's better not to see every little detail in the far future that he will never even think of. If we get within drawing range we will surely not lose even if we play imperfectly, and if we reach a position where a win is possible we will probably win even if we miss the first win. Knight odds chess requires a different mental model than normal chess. Also, the ChrisW opening book is of no use for a human preparing, it is only for getting variety for computer vs computer matches. The positions (mostly) won't actually occur in real knight odds games.
Komodo rules!
lkaufman
Posts: 6231
Joined: Sun Jan 10, 2010 6:15 am
Location: Maryland USA
Full name: Larry Kaufman

Re: Komodo Dragon vs. GM Perelshteyn, knight odds match.

Post by lkaufman »

AdminX wrote: Tue Oct 26, 2021 11:59 am Revisiting my thoughts on this match, my opinion has now changed given that the match does not start until Nov 6th and that the GM does have a version of Komodo to practice with, I now think he should have good chances to win the match. I don't believe I gave enough thought to this little detail.
Yes, he has dragon 2.5.1. Gm Wirig only trained with K12, pre-NNUE and much weaker. So this might offset progress from dragon 2 to now. If so, we might lose by 4 to 2, same percentage as Wirig match.
Komodo rules!