The World Computer Chess Championship in 2006 was held in Torino Italy. It was my second chance to attend a WCCC; I also went to the 2005 version in Reykjavik. Unlike that report, this one will not be quite as much fun to write, but despite not winning this year I managed to have a good time. Being half Italian, I was curious to see my native country, and I was also eager to see the old gang from Iceland again.
One of the cool things about this event was being next to the Chess Olympiad. Almost every chess star except for Kasparov was there, including Anand, Kramnik, Aronian, and Svidler. It was somehow disappointing, though, as they all looked exactly the same as in their photographs on chessbase. Unlike movie stars, it seems chess players don't really try to look good for the camera. As computer people, our consensus was that the team from Silica (Zappa, Rybka, Shredder, and Junior) would have little trouble taking first place, although I suspect the GMs would disagree. I even got to play against a member of the admittedly rather lowly Thai team (score 2-2. I thought he was a bit better than me but that he would have trouble with Anand).
Italy lived up to its reputation for great food, especially once I realized not to order the low carb antipasto. Unlike Iceland, where it cost an arm and a leg for a coke, the prices were reasonable and the taste excellent. Shay in particular took us to a superb restaurant near the Junior Hotel where I had a carbonara that can only be called heavenly, despite the fact that it was served at 3 AM.
The drawback of Italy was the low English level. I remember having to turn on the tap to say I wanted water (aqua, fizzante if you want bubbles) and spending 10 minutes figuring out that the man selling me lasagna was asking if I had 1.50 in change. I was actually somewhat forewarned about this. On the plane I sat next to a cute girl from Denver who was now teaching English in Italy. She claimed that they learned by reading Shakespeare and in general had almost no practice with conversational English. It was a big change from Iceland where even the security guard was able to chat with us comfortably. I don't want to come across as an American tourist annoyed that the world doesn't revolve around him, but English really has become the de facto business language of the world.After the closing ceremony, the ICGA treated us to dinner. Computer chess is one of those obscure sports where instead of million dollar prize funds and endorsement deals we have to settle for free food. Near the end, Amir, one of the programmers of Junior, told me that he had failed to win the world championship three times. He said that the first time and second time were about equally bad, but the third time was a little better. My response was to reach for another wine bottle. But in all seriousness, congratulations to Amir Ban and Shay Bushinsky. This is their third world championship - they also have a few microcomputer world titles - and I suspect this will not be their last.
As I am now a graduate student in Computer Science at the University of Illinois at Urbana-Champaign, I managed to get some time on NCSA's Cobalt supercomputer, an SGI Altix 3700 with 1024 1.6GHZ Itanium2 processors. The machine ranks 60th in the world on the Top500.org list. Sadly, the machine is split into two parts and Zappa was only able to use 512 processors at once.
From a technical perspective, the supercomputer did very well. Writing code to run well on a supercomputer is rather challenging. There just isn't enough work to do, sort of like 512 people trying to make a sandwich. In most middlegame positions Zappa managed to examine 100 million positions per second, with a peak of 300 million. This is somewhat faster than Deep Blue and by far the fastest software ever. It also managed to reach the fairly respectable depth of 20 ply on most searches. Unlike Shredder and Rybka, Zappa does not use much dubious forward pruning, which makes 20 ply much harder to reach (Vasik would say that I am bragging about my engine being stupid). My only disappointment was the performance in the late endgame, which was not quite up to par. In general I was quite pleased.
From a chess perspective things did not go so well. Over the last few years I have been developing an appreciation for how variable life is. This was capped off by reading Fooled by Randomness last month. Chess imitates life: your program will play some positions well and others poorly. If the right positions come up your program will look like a genius, if the wrong positions come up it will look like a patzer. I think that if the tournament had been played 100 times, this would have been among the worst finishes by Zappa. We had bad luck in the pairings, played the strongest set of opponents, and in general almost everything that could go wrong did. Perhaps it was some sort of arcane karma deficit from Reykjavik, where everything that could go right did.
1. e4 c5 2. Nf3 d6 3. d4 cxd4 4. Nxd4 Nf6 5. Nc3 g6 6. Be3 Bg7 7. f3 O-O 8. Qd2 Nc6 9. O-O-O Nxd4 10. Bxd4 Be6 11. Kb1 Qc7 12. h4 Rfc8 13. h5 Qa5 14. h6 Bh8 15. a3 Rab8 16. Bxf6 Bxf6 17. Nd5 Qxd2 18. Nxf6+ exf6 19. Rxd2 Rc6 20. a4
Erdo, the author of Zappa's opening book, has played a version of his favourite Dragon opening. In test games he had found that often White will allow many pieces to be traded, while Black will play f5, fxe4, Kf8-e7-f6-g5 and munch the "Erdo Pawn" on h6. Unfortunately, Black must do something about White's threat of Bb5+Rhd1, winning the d pawn.
The only moves that stop this threat are Rb6 and a6, and Zappa gleefully chooses Rb6??. The problem is that its Rook is trapped after the next few moves as White plays b4-b5. Most programs don't do a computation of a trapped rook; it takes valuable CPU time and happens only every 50 games or so. The reason it happens here is that White's h1 rook will also be stuck at home for quite a while. The difference is that Black's rook is permanently trapped, while White's rook can be activated after another 20 moves.
Around this point, and after hearing my complaints about the cruelty of life, Sandro Necchi, the book author for Shredder, bought Erdo and I some consolation grape juice.
Rb6 21. Rd4 Kf8 22. b4 a6 23. b5 axb5 24. axb5 f5 25. exf5 Bxf5 26. Kb2 Re8 27. g4 Bd7 28. c4 Re1 29. Kc3 Ke7 30. Bg2
The results of silicon stupidity. After Re2 White can remove the only active Black piece at the cost of a pawn with Re4! and Black's trapped rook is in dire straights, forcing him to shed material. I was not happy, but surprisingly Shredder missed this shot. After this Zappa sacrifices a pawn on d5 to activate its rook, and the draw was never really in question. Somehow Zappa has managed to trap its rooks at a far higher rate in world championship games (10%) than in test games (2%), but amazingly it hasn't lost either of them!
1... Re1-e2 2. Rd4-e4 Re2xe4 3. f3xe4 Bd7xg4 4. Rh1-f1 Bg4-d7 5. Kc3-b4 d6-d5 6. e4xd5 g6-g5 7. Kb4-a5 Rb6xh6 8. c4-c5 b7-b6 9. c5xb6 Rh6-g6 10. Bg2-e4 Rg6-g8 11. Be4xh7 Rg8-a8 12. Ka5-b4 g5-g4Re2 31. Rg1 d5 32. Rxd5 Be6 33. Rd3 Ra2 34. Rdd1 Rd6 35. Ra1 Rf2 36. Ra7 Rdd2 37. Rxb7+ Kd6 38. Rb6+ Ke7 39. Bh1 Rc2+ 40. Kd3 Rfd2+ 41. Ke3 Re2+ 42. Kf4 Rxc4+ 43. Kg3 Rb2 44. Rb8 Rb3 45. Kh2 Kf6 46. Rh8 Rxb5 47. Rxh7 Rb8 48. Bg2 g5 49. Rg7 Rh8 50. Kg3 Rxh6 1/2-1/2
1. d4 Nf6 2. c4 e6 3. Nf3 b6 4. g3 Ba6 5. b3 Bb4+ 6. Bd2 Be7 7. Bg2 Bb7 8. Nc3 O-O 9. O-O Na6 10. Bc1 d5 11. Ne5 c5 12. Bb2 Nc7 13. Rc1 Bd6 14. cxd5 exd5 15. Nc4 Be7 16. dxc5 Bxc5 17. Na4 Be7 18. Ne3 Re8 19. Nf5 Ne6 20. Rc2 Qd7 21. Qd3 Rad8 22. Rd1 h6 23. Rcc1 Kh8 24. h4 d4 25. Nxd4 Bxg2 26. Kxg2 Nxd4 27. Bxd4 Qg4 28. f3 Qh5 29. Qc2 Bd6 30. Bf2
Junior has sacrificed a pawn to arrive at this rather superficially active position. Black's pieces look good: the minors and the rooks are centralized, White has weakened his pawn shield, and the Black queen sits menacingly near the White King. Despite this, it is hard to see how Black can prevent White from regrouping.
Rd7 31. Nb2 Rc7 32. Qd3 Rxc1 33. Rxc1 Ba3 34. Rc2 Bxb2 35. Rxb2 Qe5 36. Rd2 Kg8 37. e4 Rc8 38. g4 Nh7
And so Zappa has more or less consolidated, enjoying more activity, a good bishop versus a bad knight, and an extra pawn. I smelled victory - all Zappa had to do was trade the queens and it would all be over but the kicking and the screaming. It was at this point that Zappa popped out with Qa6?? The problem with this move is that it wins two more pawns that White simply doesn't need while giving the Black pieces a new lease on life. Munching the first pawn was healthy greed; the second two will prove to be a case of indigestion.
39.Qa6 Qc3 40. Rd5 Nf8 41. Qxa7 Ne6 42. Kg3 Nc7
It may be hard to believe, but Black is already close to a draw. White's best is probably Rf5, for example Rf5 g6 (Ne6 Qxf7, anything else Qxb6 and Rc5) Qb7 Rd8 Bxc7 gxf5 Qxc7 Qxc7 Bxc7 where he can pressure Black in the endgame for a while, although Black should manage a draw. After Rd1, Black has enough activity to force the draw fairly quickly.
43. Rd1 Ne6 44. Qxb6 Qe5+ 45. Kg2 Rc2 46. a4 Nf4+ 47. Kh1 Nh3 48. Rf1 h5 49. gxh5 Re2 50. Qd8+ Kh7 51. Qb6 Qf4 52. Kg2 Qxh4 53. Qc5 Nf4+ 54. Kg1 Nh3+ 55. Kg2 Nf4+ 1/2-1/2
If you are the type of person who likes to keep large log files of computer chess programs on your hard drive, you have come to the right place. The log file is almost completely undoctored; I removed some debug information and some time between games when Erdo and I were analyzing to keep the size down and make things more readable. When reading the log file, please note that Zappa's scores are for itself, not white. You will also have to use a text editor that handles Unix linefeeds.