Sunday, June 17, 2007

Big Fishes and Little Fish Prologue


Blue Devil (links: left), looks onto Wormwood (center), as Temposchlucker (right) asks them: "How can we stop this planned series of essays by dk-transform? What if he mentions us? What if he goes to print as soon as Tuesday?"

While I round out the second essay on bloggers, and prepare to write the third, so that, as it were I can get a head start on all readers, and in so doing as I publish essay one, write the fourth essay, then publish essay two the next day or two days latter, then write the fifth as I publish the second, etc, and so stay 'ahead of everyone'. Fifteen essays, fifteen bloggers.

I have been thinking about Rybka a lot lately. Do I really need it? No. But so cool!

Opening ICC tonight and, to amazement, finding the results from round 8 for The World Computor Chess Championship (WCCC), Amsterdam, Netherlands June 11-18, 2007, only flamed my sense of enchantment all the more... sporting a sharp game between Deep Sjeng and our subject program. Talk about brilliance and creativity! [other (bad) link provided by ICC and citations at Google alike inoperable at 17:45 EST, 22:45 BST Sat, so directly thus: ]

I am inches from buying Rybka Version UCI for multiprocessor, I assume, 32 bit 2.3 for dual core processors. Without further ado, one of the many articles linked at their handsome site, this one from

Rybka’s immortal game
Sunday 10 December 2006, 14.20 CET, by

Kramnik’s 4-2 loss of last week against the computer program Deep Fritz was a big blow. Not as big of course as Kasparov’s defeat in 1997 against Deep Blue, which is universally considered as the end of the human supremacy over chess engines, but the international media did mention it, for example the BBC and CBS. The news even inspired a computer programmer to try and make a program that will be the best in the poker world as well! Especially for the company Chessbase the victory of their darling pet Fritz was nice of course, and they did not dread in calling it the world’s leading chess computer program, already during the match. However, Deep Fritz is not the best. Not at all.

Chess connoisseurs (like the readers of this site) know already that between the chess programs we have to speak of an absolute monarch that listens to the name Rybka. But how strong is Rybka? And then, how strong is (Deep) Fritz? A lot of chess players are asking such questions these days and therefore today we pay attention to computer chess.

The strength of human beings is represented quite clearly in the FIDE rating list, but the situation in the world of chess engines is a bit more complicated since several lists are beging used there. There are different lists for 32-bit or 64-bit engines that run on one or more processors. But there’s also the difference between the 40/20 and 40/120 lists: based on 40 moves in 20 minutes or 40 moves in 120 minutes. We’re talking about test games, because computer engines rating lists are (because of a lack of enough man vs machine tournaments) based on test games against each other.

If the above made you dazzle, what follows will bring both of your feet back onto the ground: Rybka tops every computer rating list there is. In the next list, by
Computer Schach und Spiele, only 32-bit versions of engines for one processor appear. In other words: programs that can be bought and used by you and I for playing or analysing.CSS Rankings:

1. Rybka 2.2 2972
2. Loop 10.32 2853
3. Shredder 10 UCI 2838
4. Toga II 2835
5. Fritz 10 2832
6. Fruit 2.2.1 2820
7. Hiarcs X54 UCI 2810
8. Spike 1.2 Turin 2782
9. Junior 10 2772
10. Ktulu 8 2752

There’s a huge hole between the number 1 (the latest version of Rybka, just released) and the rest. Compared to version 2.1, Rybka’s programmer, Vasik Rajlich, managed to make his engine stronger especially in tactics. The list does show this. Version 10 of the world famous engine Fritz was released very recently as well. Compared to predecessor Fritz 9 this program got stronger about 30 to 50 points but between the computer engines it occupies a modest fifth place.

So what about Deep Fritz, the program that beat Kramnik? According to Jeroen Noomen, a journalist for Computerschaak and responsible for Rybka’s opening repertoire, Deep Fritz should be compared to Rybka 2.2 mp, the multi-processor version: “Both play on two or more processors. At the moment it’s clear that Fritz 10 is about a hundred rating points weaker than Rybka 2.2, and the situation will probably be about the same for Deep Fritz 10 and Rybka 2.2 mp. So far I ran about 650 test games between Rybka 2.2 and Fritz 10 and the score is 65% for 2.2. Fritz 10 will probably fight for the second place with Loop, Shredder 10 and Hiarcs 11. The distance to Rybka 2.2 is quite big.” And this is quite unique, also in the world of chess engines. Noomen: “Never before in the history of computer chess there was such a big difference between the number 1 and 2!”

What is it that makes Rybka so strong? The answer is: her (the programmer considers Rybka to be female) evaluation. It’s far better than with other programs. One of the reasons must be that programmer Rajlich is an IM. Jeroen Noomen is a member of the Rybka team and cannot be really objective here, but still, I’ll let him speak once more. In the September issue of Computerschaak he writes: “The big news must be brought now. Despite their sky-high ratings, the evalution of most of the engines I’ve seen, is crap! This sounds harsh, but it is what I’m observing.”

In May this year, at the Turin Olympiad, I spoke to Vasik Rajlich myself. Then I already asked him what makes Rybka so strong The programmer also mentioned her evalution first. As an example he gave this position:

And to explain how engines can differ in evaluating positions, he added this list:

Rybka 1.2 0.57
Fritz9 0.86
Rybka 1.0 1.30
Fritz 5.32 1.00

An early version of Fritz only sees the extra pawn. Endings was Rybka 1.0’s weak spot and so here she’s far too optimistic. But this can also be said for Fritz 9, because naturally White’s winning chances are not bigger than, say, half a pawn.
Rajlich told me he doesn’t really believe in brute force, but considers selective search more important. Intelligence that is. “My motivation is to create a chess program that understands chess.”

Vasik Rajlich at the Turin Olympiad, May 2006 [photo, dk]
In Chess Today’s edition of last Friday, I ran into the next, fantastic game by Rybka. If you thought that games between computers are never interesting, forget about that. Chess Today’s annotator, IM Andrei Deviatkin, is right when he compares the game with Anderssen-Dufresne and Anderssen-Kieseritzky. This is Rybka’s immortal game.

Shredder 10 - Rybka 2.1WBEC13 Premier Division, 2006
1. e4 e5 2. Bc4 Nf6 3. d3 Nc6 4. Nf3 Bc5 5. O-O d6 6. c3 O-O 7. Nbd2 a6 8. Re1 Ba7 9. Bb3 Ng4! 10. Re2 Kh8 11. h3 Nh6 12. g4?Better is 12. Nf1 f5 13. d4 Qf6 like in Kosteniuk-Mamedyarov, Lausanne 2003. Then Black is doing fine as well.

12… Nxg4!The start of a beautiful attack on the king, in the style of the old masters.13. hxg4 Bxg4 14. Nh2 Qh4! 15. Nxg4 Qxg4+ 16. Kf1 f5 17. Rb1 (17. exf5 Rxf5 18. Ne4 Rh5! (18… Raf8 19. Ke1 Bxf2+ 20. Kd2 Rf3 21. Kc2 ended in a draw in Izmukhambetov-Aronian, Moscow 1996) 19. Ng3 Rh2 and Black wins (Deviatkin).17… Rad8 18. Qc2 f4! 19. Ke1 f3 20. Nxf3 Rxf3 21. Be3 Rdf8 22. Kd2

“Black’s position is obviously won. But, instead of winning technically, Black preferred to play very… romantic and impressive chess! It may seem that Black’s pieces belonged, for example, to Paul Morphy or Alexander Alekhine. However, in reality Black is only a computer program, which chose these “romantic” moves only for the reason of the better evaluation after it!” - Deviatkin.

22… Bxe3+ 23. fxe3 Rxe3! 24. Qd1 Qxe2+ Deviatkin.
23. fxe3 Bxe3+! 24. Rxe3 Nd4! 25. cxd4 Rf2+ 26. Kc3
26. Kc1 Qf4 27. Qxf2 Qxf2 28. dxe5 Qxe3+ Deviatkin.
26… exd4+ 27. Kxd4

27… Qg5!!
“A fantastic move! Instead of taking the queen (which would win too), Black leaves his own rook under the queen’s threat. But actually, White is helpless.” - Deviatkin.

28. e5
28. Qxf2 Qc5#; 28. Qxc7 Qe5+ 29. Kc4 d5+ 30. Kb4 Qxc7; 28. Rf3 c5+ 29. Kc4 b5+ 30. Kc3 Qe5+ 31. d4 Qxd4#; 28. Qc4 Qe5#; 28. Qc3 c5+ 29. Kc4 b5# Deviatkin.

28… Rxc2 29. Bxc2 c5+ 30. Kd5 Qxe3 31. Kxd6 Qd4+ 32. Kc7 Qxe5+ 33. Kxb7 g5 34. Bb3 g4 35. Kxa6 g3 36. Kb5 g2 37. Kc6 Qd4 38. Bd5 g1D 39. Rxg1 Qxg1 40. a4 h5 41. Be4 0-1


Post a Comment

<< Home