Author: Serge Desmarais
Date: 17:11:11 09/13/98
Here is an interesting game played on the Internet Chess Club. Crafty 15.18 was K6-Krafty (running on a AMD-K6 250 MHz - for generally over 100,000 nps) Fritz 5.03 was Braincan (running on a Cyrix P150+ with 32 Megs hash - default book customized by adding games). The game time was 15 minutes for the game with a 15 seconds incrementation after every move. Crafty got into trouble for being too materialistic. [Event "15 15"] [Site "Internet Chess Club"] [Date "1998.09.13"] [Round "?"] [Result "0-1"] [White "Crafty 15.18"] [Black "Fritz 5.03"] 1. e4 Nf6 2. Nc3 d5 3. exd5 Nxd5 4. Bc4 Nb6 5. Bb3 Nc6 6. Nf3 Bf5 7. d4 e6 8. O-O Be7 9. Be3 O-O 10. d5 exd5 11. Bxd5 Nxd5 12. Nxd5 Be4 13. Nxe7+ Qxe7 14. Re1 Rad8 15. Bd2 h6 16. Re3 f5 17. Qe2 Qf7 18. Bc3 Rfe8 19. Nh4 Ne7 20. Rg3 g5 21. f4 Nd5 22. fxg5 f4 23. Rh3 Nxc3 24. Rxc3 hxg5 25. Nf3 Bc6 26. Qf2 Qg7 27. Qxa7 g4 28. Nh4 Rd2 29. Qc5 Qf6 30. Rd3 Rxd3 31. cxd3 Qxh4 32. Qc4+ Kg7 33. Qxf4 Re2 34. Qxc7+ Kg6 35. Qd6+ Kh5 36. Qc5+ Qg5 37. Qxg5+ 37... Kxg5 38. g3 Rg2+ 39. Kf1 Rxb2 40. Re1 Bg2+ 41. Kg1 Rxa2 42. Re8 Bh3 43. Rg8+ Kf5 44. Re8 b5 45. Rb8 Rb2 46. Re8 b4 47. d4 b3 48. Re5+ Kf6 49. Rb5 Rg2+ 50. Kf1 Rxh2+ 51. Ke1 b2 52. Kd1 Bf1 53. Rb6+ Kf5 54. Ke1 Bd3 55. Kd1 Ke4 56. Rxb2 Rxb2 57. Kc1 Rf2 58. d5 Kd4 59. Kd1 Kc3 60. d6 Rf1# The bad mistake Crafty did is 27.Qxa7?? Strangely enough, K6-Krafty was kibitzing its evaluation to its opponent after every move. Here is what I noted in Crafty's eval. 27.Qxa7 +0.05 28.Nh4 -0.56 29.Qc5 -1.94 30.Rd3 -3.86 31.cxd3 -3.89 32.Qc4+ -4.56 33.Qxf4 -4.65 34.Qxc7+ -4.83 and so on... Now, it is interesting to note that Fritz never seriously considered 27.Qxa7??, but instead 27.Rc5 (attacking the g5 pawn) was the move expected to which it could/would have replied 27...Bd5 with an approximate equality. Immediately after the pawn capture, Fritz saw a big advantage for Black. From Crafty's eval, above, Crafty thought the position was about equal after the capture (e.g. Black has compensations (?) sufficient for the pawn, but not more). Now, immediately after the pawn capture, Crafty sees that Black is at least slightly better, while 2 moves after it says "I am losing". I noted this typical computer behaviour quite often on different programs. For example, at 13 ply, a program thinks that after move X the position is equal. Suddenly, after making that move, it says it is losing while at ply 7-9. Who is it possible to say so, when the depth ways at least 3 ply greater when the move was played??? So, what it sees after the move is played has necessarily been taken into account before playing the move? If it was okay then, why is it not okay after the move is played? The alpha-beta is supposed to mean that ALL the other moves, apart from the one played, are inferior to it. Then it realize that maybe MOST of the other moves were better? I have had Fritz play several automated Crafty on the ICC and I noticed this Crafty behaviour quite often. This way Fritz/Cyrix P150+ was able to blast Crafty PII-450 Mhz quite a few times. Here is the "secret" : you let Crafty capture unimportant pawns far away from the main scene of the action, while you bring your piece closer to Crafty's king. Then, a sacrifice (temporary) opens Crafty's king's surroundings. Then, a king chase occurs and a mating net is created. Maybe something could be implemented as to make Crafty more cautious about capturing free pawns that are far from the main scene? Of course, sometimes Fritz attack could not break and it would lose a long and painful ending. But it seems that Fritz has generally a strange ease at beating the Crafty clones. Maybe it is caused by their respective "style"? Serge Desmarais
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.