Komodo vs GM Martin Petr (2515)

[lkaufman]

Well, Milov won the handicap games (some f7, some Exchange) by 4 to 2, a 120 elo margin. Komodo is much more than 120 above Rybka 3 (it leads Rybka 4 by 73 wins to 1 (with 26 draws) in CEGT testing at 40/20), and the computer has three times as many cores now. So on paper Komodo should be a big favorite at these handicaps against Milov (or Movsessian now, they are of similar strength I think). But the problem is that handicap results depend much more on the strength of the weaker player than they do on the strength of the stronger player. If Rybka already outsearched Milov typically by say ten ply, does it really help to look another two ply deeper in a lost position, where there is nothing good to find? Probably it helps a bit, but not by nearly as much as it helps in normal chess. So I would say that these handicaps are pretty fair for Komodo against players of the level of Milov or Movsessian. I don't know which way I would bet.

52 points elo difference , although 79 places below milov in world ranking as he was then.


http://www.chess.com/news/shock-result- ... bka?page=2

In a recent 8-game match with the silicon monster, Milov (rated a mighty 2705 Elo and ranked 28th among earth-dwelling carbon based life forms) ended the winner by a narrow margin of 4.5-3.5.



https://ratings.fide.com/card.phtml?event=310204

rate std. 2653

World (all players): 107


https://ratings.fide.com/card.phtml?event=1304364

Milov was at or near his peak rating then; normally he was a bit lower. Movsessian had a higher peak rating but is relatively low now. So probably they are of comparable strength.

[Jesse Gersenson]

There are any number of factors which contribute to lowering a players rating as they age but positional understanding isn't the first to go. Movsessian was at both the Petr match and the Neuman match to see how the matches were being run and, while he was there the players discussed the starting positions. It was very clear Sergie's understanding of chess is on a totally different level.

Here's an article about the match in Czech:
http://www.praguechess.cz/index.php?mod ... &langue=cs

Here's my rushed translation, skipping over the Neuman section:
The best chess program of our time, Komodo 9, brutally beat grandmaster Martin Petr, in loo of giving two pawns odds each game. The final score 4,5 : 1,5 (3 wins for Komodo and 3 draws) was cruel.

On sunday and monday, 23rd and 24th of august in the chess cafe Little Vaclav in the hotel Julis they played three games per day at a slow rapid (45m+15s) and the silicon monster in turn in the starting position was missing a+b, a+d a a+e. GM Petr prepared the openings well and out of the openings he had decent positions in every game. In his prepareation he found the most difficult position was what to play after 1.e4 in the a+d games. He decided on a solid structure of d6, e6 and g6 and it wasn't apparent where he planned to beat the machine. However Komodo 9 played calmly, defending it's position with exchanges, improved the position of it's pieces and didn't make mistakes.

During the course of the games and with time running low, the verdict was clear why there is a rating difference of 800 points between these players.

All games can be downloaded from...
Also interesting was the big difference between these pawn configurations and the c2f2 configurations of the Neuman match....
The third Komodo 9 match in Prague will be 9-10 september, again at the Chess Cafe Vaclavek, starting at 3pm (9am ET). it will be played between the super program running on a 24 core computer battling against Sergei Movsesjan. In the match the armeanian grandmaster will always have the white pieces and komodo will also be missing the f7 pawn. Come and watch!

[duncan]

Well, Milov won the handicap games (some f7, some Exchange) by 4 to 2, a 120 elo margin. Komodo is much more than 120 above Rybka 3 (it leads Rybka 4 by 73 wins to 1 (with 26 draws) in CEGT testing at 40/20), and the computer has three times as many cores now. So on paper Komodo should be a big favorite at these handicaps against Milov (or Movsessian now, they are of similar strength I think). But the problem is that handicap results depend much more on the strength of the weaker player than they do on the strength of the stronger player. If Rybka already outsearched Milov typically by say ten ply, does it really help to look another two ply deeper in a lost position, where there is nothing good to find? Probably it helps a bit, but not by nearly as much as it helps in normal chess. So I would say that these handicaps are pretty fair for Komodo against players of the level of Milov or Movsessian. I don't know which way I would bet.

52 points elo difference , although 79 places below milov in world ranking as he was then.


http://www.chess.com/news/shock-result- ... bka?page=2

In a recent 8-game match with the silicon monster, Milov (rated a mighty 2705 Elo and ranked 28th among earth-dwelling carbon based life forms) ended the winner by a narrow margin of 4.5-3.5.



https://ratings.fide.com/card.phtml?event=310204

rate std. 2653

World (all players): 107


https://ratings.fide.com/card.phtml?event=1304364

Milov was at or near his peak rating then; normally he was a bit lower. Movsessian had a higher peak rating but is relatively low now. So probably they are of comparable strength.

http://www.chessgames.com/perl/chessplayer?pid=19920

Movsesian's peak rating was 2751 in January 2009 when he was ranked #10 in the world, that being his peak ranking to date.

[Jesse Gersenson]

The exact odds of the Movsessian match will be agreed this week. The rough sketch is 4 games at pawn and move odds and 2 games at exchange odds. (The article I translated got it wrong, saying all games would be pawn and move).

[Vinvin]

If Rybka already outsearched Milov typically by say ten ply, does it really help to look another two ply deeper in a lost position, where there is nothing good to find?

You pointed the problem.
The idea is not to search deep but to search well and wide.
The program should guess the moves the human player is more likely to make and find good replies for those moves.
The eval should take in account all the parameters that make a position difficult to play for a human.

[Jhoravi]

The exact odds of the Movsessian match will be agreed this week. The rough sketch is 4 games at pawn and move odds and 2 games at exchange odds. (The article I translated got it wrong, saying all games would be pawn and move).

I believe the Martin Pert Odds is more interesting because we have a much stronger Grandmaster this time. The extra pawn in pawn odds has a potential to promote to queen for Mankind whereas Exchange Odds doesn't have that advantage. The engine can easily balance its Bishop against Rook as these are already drawn in the endgame.

[lkaufman]

The exact odds of the Movsessian match will be agreed this week. The rough sketch is 4 games at pawn and move odds and 2 games at exchange odds. (The article I translated got it wrong, saying all games would be pawn and move).

I believe the Martin Pert Odds is more interesting because we have a much stronger Grandmaster this time. The extra pawn in pawn odds has a potential to promote to queen for Mankind whereas Exchange Odds doesn't have that advantage. The engine can easily balance its Bishop against Rook as these are already drawn in the endgame.

It's not rook for bishop, it's rook for knight, and most endgames with equal pawns are won for the rook. I suppose you are talking about endings with no pawns, but they are rare. Engine vs. engine tests show this handicap and the f7 pawn are about equally large, and the Rybka vs. Milov match also had Milov one game ahead at each handicap.

[lkaufman]

If Rybka already outsearched Milov typically by say ten ply, does it really help to look another two ply deeper in a lost position, where there is nothing good to find?

You pointed the problem.
The idea is not to search deep but to search well and wide.
The program should guess the moves the human player is more likely to make and find good replies for those moves.
The eval should take in account all the parameters that make a position difficult to play for a human.

This would require quite a different program than normal Komodo. Perhaps we'll try it some day.

[duncan]

If Rybka already outsearched Milov typically by say ten ply, does it really help to look another two ply deeper in a lost position, where there is nothing good to find?

You pointed the problem.
The idea is not to search deep but to search well and wide.
The program should guess the moves the human player is more likely to make and find good replies for those moves.
The eval should take in account all the parameters that make a position difficult to play for a human.

This would require quite a different program than normal Komodo. Perhaps we'll try it some day.

to do this well may be tricky, but would it be hard to do a simple version ?

'The eval should take in account all the parameters that make a position difficult to play for a human'

[Jesse Gersenson]

Pictures taken by Anezka Kruzikova during the match. Thanks Anezka!