Return Match for Komodo

[M ANSARI]

I am sorry to be blunt, but it really looks like you were cheating by either taking moves back or by using engine output or take move back to choose your moves. I mean sacrificing a pawn to improve activity and allow your Knights and queen to create an attack that allows the win of the opponents Queen by discovered check while hanging a bishop .... brilliant calculation and balls made out of titanium. I would like to see this type of play under controlled conditions, similar to what the GM in this match had to endure. So before trying to make the GM look like a patzer, why not show those skills in similar conditions.

I have not been on these forums for a while ... but are you the same guy that insisted a while back that Borislav Ivanov was not cheating when he was slaying GM after GM with spectacularly tactical chess that was above the level of even the best super GM's?

[Guenther]

Rigth, Kai,

I have been looking at the engine output, this gives considerable edge to the human.
The reason though is not that I wanted to do so, but simply that it stayed like that after I had been play-analysing some positions.

I really do not think though that this is a decisive factor with such a huge handicap.

Completely wrong again! It is deadly decisive exactly because of the handicap White already had and you should know it, if you had any clue about what you are talking.

And on a side note for your other post, I really don't care about your 'Fritz' settings, nor do I ever use Fritz or similar for what I want to do.
I also don't care in which ways you botch up your hardware/software or what else, because it is clear that nothing works correct on your side
and that you have no idea of a scientific approach to anything.

This is my last post ever, which is directly written towards you.

[yurikvelo]

Code: Select all

D=16, 47MN, K9.3
-1,37	16.h4 b6 17.Qe2 a6 18.a4 a5
-1,87	16.Nh4 b6 17.Qb3 g5 18.Nf5 Nxf5
-1,87	16.Bc4 g5 17.Nxe5 dxe5 18.d6+ Kh8
-1,92	16.Nh3 Bxh3 17.gxh3 b6 18.Kh1 Qd7
-1,93	16.Be2 g5 17.h3 g4 18.Nd2 gxh3
-1,93	16.h3 g5 17.Nd2 g4 18.Be2 gxh3
-2,00	16.Qb3 g5 17.Nxe5 dxe5 18.d6+ Kh8        <---- K8 -1.18 @ D=16
-2,05	16.c4 g5 17.c5 g4 18.cxd6 cxd6
-2,07	16.Nd2 g5 17.h3 g4 18.Be2 gxh3
-2,08	16.a4 g5 17.a5 b6 18.Nd2 g4
-2,09	16.Bc2 g5 17.h3 b6 18.Nh2 Bd7
-2,10	16.Kh1 g5 17.g3 Ng6 18.gxf4 exf4

D=16, 7.2 MN
-1,54	20.Nxg5 Bxg5 21.Be2 Rg8 22.c4 Bf6
-2,13	20.Nh4 Rg8 21.Kh1 Nh7 22.Nf3 b6            <---- K8 -1.04 @ D=16

D=16, 18.6 MN
-2,37	23.Bc1 Nxf3+ 24.Qxf3 Bxc1 25.Qxh5+ Kg7
-2,62	23.g3 Nxf3+ 24.Qxf3 Qh4 25.gxh4 Nxh4+
-2,70	23.Nh2 Nf4 24.Kh1 Nh7 25.Rg1 Qh4
-2,82	23.Ne1 Nf4 24.Kh2 Nf7 25.Rh1 Qh4+         <---- K8 -2.00 @ D=16
-4,06	23.Bb5 Nxf3+ 24.Qxf3 Nf4 25.g3 Qh4
-4,28	23.Nxe5 Nxe5 24.Qxh5 Ngf3+ 25.Kh1 Qh4+

[lkaufman]

Aside from the time and other settings, the game shows why Contempt is so vital in these matches (which wasn't available on Komodo 8 or even 9). A high setting not only avoids piece trades, it avoids BLOCKING the POSITION, which was the key to your victory. I don't think that Komodo with a setting like Contempt - 80 or more would ever allow such a blocked King's Indian structure. It strives to open the game, which is of course the way to play against humans. I don't think we need a special engine for these matches as has been suggested; contempt does just what is needed to beat humans. Keep pieces on the board but trade enough pawns for an open game.

[Jesse Gersenson]

Code: Select all

47MN

I don't understand the comparison you're making but, if it's relevant, Komodo averaged 2566 million nodes per move in game 4. If anyone wants the raw engine output, or arena.debug files, I'd be happy to post them.

[mjlef]

Contempt, the type which helps Komodo play odds matches better, was introduced publically in Komodo 9.2. It is not in Komodo 8 which Lyudmil Tsvetkov is using.

Average search depth reached in Neuman match was 31, and just 17 in Tsvetkov game.

The starting position is nearly correct -- b2+f2 is very close. Try again with the actual handicap used, b2+e2.

I can confirm that. The older "Contempt" we even called "Drawscore" since it was what Komodo considered a draw to be worth (so the default -7 just tied to avoid draws). The new Contempt does much more. A positive Contempt discourages accepting early draws, avoids trading material to keep the game complex, and tries to keep a more open pawn structure. These make a huge difference in these games. If Lyudmil privately send us his email I would be glad to get him a more recent copy of Komodo to see how it goes.

Mark

[clumma]

So who won?

I don't see the result in this thread, nor anything about the match on chess.com, nor anything about the match on "our site" (komodochess.com?).

-Carl

[lkaufman]

So who won?

I don't see the result in this thread, nor anything about the match on chess.com, nor anything about the match on "our site" (komodochess.com?).

-Carl

The result and games were posted in this thread. Result 3.5 to 0.5 in favor of Komodo. Look for post by Jesse Gersenson for the games. Game 4 was spectacular. I suppose he'll have our site updated in the next few days. Chess.com has the results and games, not sure if they plan a story.

[Jesse Gersenson]

Added results here:
https://komodochess.com/store/pages.php?cmsid=17

[Lyudmil Tsvetkov]

Contempt, the type which helps Komodo play odds matches better, was introduced publically in Komodo 9.2. It is not in Komodo 8 which Lyudmil Tsvetkov is using.

Average search depth reached in Neuman match was 31, and just 17 in Tsvetkov game.

The starting position is nearly correct -- b2+f2 is very close. Try again with the actual handicap used, b2+e2.

I can confirm that. The older "Contempt" we even called "Drawscore" since it was what Komodo considered a draw to be worth (so the default -7 just tied to avoid draws). The new Contempt does much more. A positive Contempt discourages accepting early draws, avoids trading material to keep the game complex, and tries to keep a more open pawn structure. These make a huge difference in these games. If Lyudmil privately send us his email I would be glad to get him a more recent copy of Komodo to see how it goes.

Mark

Thanks for the offer, Mark.

If it would not considerably hurt your budget, I will take advantage of the opportunity to thrash newest Komodo in some games.

My mail is ltsvet((at))yahoo((dot))com

Not that it will make significant difference.
I have been following SF and Komodo development closely for the past 3-4 years, and, if some things really change for the better, others simply remain the same.

I also have just a 4-core, so any successful performance migth be attributed to that.

I would also shun posting games on this forum, as the backlash is oversharpish.

Btw., congratulations for the match win, and the new contempt( I know Larry is furious at me for disrespecting contempt, I am doing it out of general considerations, as any contempt simply produces weaker moves, but of course, from a purely psychological point of view, this migth be an excellent option vs humans indeed).

PS. An interesting thing I discovered during trying-out these handicap positions, no matter if the engine plays without a knigth, rook, 2 pawns or just a pawn, is that, with very small specificities for the particular positions, almost always the best 1st move for black is a kingside fianchettoe (g6,Bg7,Nf6), followed by 1...c5, then 1...e5, and 1...d5 comes only in 4th.

I did extensive testing with those positions, and indeed, kingside fianchettoe is always the best line of play, while d5 seems like a low-ranging choice. (Fischer and Kasparov should have known that)

woops, got too long...