Author: David Eppstein
Date: 20:13:51 09/27/98
Go up one level in this thread
On September 27, 1998 at 09:38:30, Vincent Lejeune wrote: >[Event "ICS Rated blitz match"] >[Site "chessclub.com"] >[Date "1998.08.09"] >[Round "-"] >[White "Good-Boy"] >[Black "singacrafty"] >[Result "*"] >[WhiteElo "2914"] >[BlackElo "2855"] >[TimeControl "120+8"] > >1. e4 c5 2. c3 e6 3. d4 d5 4. e5 cxd4 5. cxd4 Nc6 6. Nf3 Nge7 7. Na3 Nf5 8. >Nc2 Be7 9. Bd3 Qb6 10. g4 Nh4 11. Nxh4 Bxh4 12. O-O f6 13. g5 Bxg5 14. Qh5+ >Kf8 15. f4 Bh6 16. exf6 Qc7 17. f5 Bxc1 18. fxe6 Bxe6 19. fxg7+ Kxg7 20. >Raxc1 Bf7 21. Qg5+ Bg6 22. Qf6+ Kg8 23. Bf5 Nd8 24. Bc8 Nf7 25. Nb4 Qd8 26. >Nxd5 Qxf6 27. Nxf6+ Kg7 28. Bxb7 Rad8 29. Nd5 Nd6 30. Rc7+ Kh6 31. Bc6 Nf5 >32. Rxa7 Nxd4 33. Bb7 Rd7 34. Rd1 Nf3+ 35. Kg2 Rg7 36. Kf2 Ne5 37. Ke3 Rb8 >38. b4 Rgxb7 39. Rxb7 Rxb7 40. a4 Bf7 41. Ke4 Bxd5+ 42. Rxd5 Nc4 43. b5 Ra7 >44. b6 Rxa4 45. Kd3 Na5 46. Rd6+ Kg5 47. Rd7 Nc6 48. Rxh7 Rb4 49. Rb7 Rh4 >50. Rc7 Nb4+ 51. Ke2 Rxh2+ 52. Kf3 Rh3+ 53. Kg2 Rb3 54. b7 Na6 55. Rc6 Nb8 This is one of those give-away only-a-computer-could-be-this-stupid moves, right? The best black can hope for by playing Nb8 is to eventually win a pawn and be in a drawn KRN-KR endgame, but the downside is (as happens below) losing because the pawn eventually queens. Why not play 55...Rb7 and draw immediately? I do understand why a computer might play this: if it follows the orthodox material-count plus positional terms eval, a knight might be worth more than an isolated passed pawn, maybe even one as far advanced as this. The quick fix is to add a little code to somehow tweak the eval of these KRN-KRP positions to be nonpositive, so the draw looks good. I think a more principled approach would be to have an eval that gives you explicit and separate probabilities of various game-ending events: What are the probabilities that black or white wins by a quick checkmate or decisive combination? That black or white can force a repetition draw? That (if the game reaches an endgame) white or black will win there? The overall eval would then be a simple nonlinear combination of these probabilities (just the expected value of the overall game). I don't see why this approach should be any slower than the orthodox one, since it is again just a simple numerical combination of the same sorts of eval terms. But it should have two big advantages: you can measure how often the events actually occur, and tune your eval accordingly, and you can correctly judge sacrifices or positions like this where material-count is just wrong. A third less important advantage is that by changing the combination formula, you could make the program play appropriately in must-win or must-draw tournament situations. By the way, who is good-boy? >56. Rc7 Rb2+ 57. Kf3 Kf6 58. Ke4 Rb4+ 59. Kd5 Na6 60. Rh7 Rb5+ 61. Kc6 Rb2 >62. Rh6+ Kg7 63. Rd6 Kf7 64. Rd7+ Ke6 65. Rg7 Rc2+ 66. Kb5 Nb8 67. Rg4 Kf7 >68. Rc4 Rb2+ 69. Rb4 Rd2 70. Kc4 Kf6 71. Rb5 Rc2+ 72. Kd3 Rc1 73. Ke4 Re1+ >74. Kf4 Rf1+ 75. Kg4 Rc1 76. Kf3 Ke6 77. Ke4 Re1+ 78. Kf4 Kd6 79. Rb4 Kd5 >80. Rb6 Re4+ 81. Kf5 Re2 82. Kf6 Re3 83. Rb1 Rd3 84. Kf5 Rd2 85. Kf4 Kd4 >86. Rb4+ Kd3 87. Ke5 Kc3 88. Rb5 Kc4 89. Rb1 Rc2 90. Kd6 Kd3 91. Kd5 Rd2 >92. Rb3+ Kc2+ 93. Kc4 Rd8 94. Rb5 Rd6 95. Kc5 Rf6 96. Kd4 Rf4+ 97. Ke5 Rc4 >98. Kd5 Ra4 99. Kd6 Rg4 100. Kc7 Rg8 101. Ra5 Kc3 102. Ra8 Kd4 103. Rxb8 >Rg7+ 104. Kb6 Rg6+ 105. Ka7 Rg7 106. Rd8+ Kc5 107. Ka8 Rg1 108. b8=Q Ra1+ >109. Kb7 Rb1+ 110. Kc7 Rxb8 111. Kxb8 Kc4 112. Kb7 Kc5 113. Rd7 Kb4 114. >Rc7 Kb5 115. Rc1 Kb4 116. Kb6 Kb3 >*
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.