Author: Steven Edwards
Date: 18:33:06 04/10/05
Go up one level in this thread
On April 10, 2005 at 19:54:14, Roger D Davis wrote: >On April 10, 2005 at 19:06:00, Steven Edwards wrote: >>Symbolic: Status report 2005.04.10 >I love this, because you're kind of thinking outloud as you're keeping us >informed of your progress, and it looks like really futuristic stuff. > >A question...I assume the toolkit itself can play a complete game, yes? If so, >then the toolkit's performance becomes an important reference point for the >success of the project. Let's say that the toolkit obtains an average rating of >2350 at some time control. Let's say that the whole project obtains an average >rating of 2450. That means that the LISP code, architecture, etc., is yielding >about 100 rating points over what the toolkit alone could achieve. What I'm >saying is that the value of the LISP code, architecture, etc., can be judged by >how much better the total project plays relative to the toolkit alone, right? Yes, the toolkit can play a complete game by itself. It has played quite a few on ICC and has achieved a 2240 Elo rating at Standard time controls over some 450 games hosted on a modest computer. Here's a recent game: [Event "Xboard game"] [Site "cynthia.local"] [Date "2005.04.10"] [Round "31"] [White "Symbolic v2005.04.10"] [Black "SlowBo"] [Result "1-0"] [Termination "Black is checkmated"] [TimeControl "G/1500+10"] [UTC "2005.04.10 23:37:54"] [WhiteElo "2240"] [BlackElo "1899"] [ICS "chessclub.com"] 1. e4 c5 2. Nc3 Nc6 3. g3 g6 4. Bg2 Bg7 5. d3 e6 6. Nge2 Nge7 7. O-O O-O 8. f4 d6 9. Be3 b6 10. Qd2 Bb7 11. Kh1 Rc8 12. g4 d5 13. exd5 Nxd5 14. Nxd5 exd5 15. Rab1 Qh4 16. g5 d4 17. Bf2 Qg4 18. Rbe1 Nd8 19. Bxb7 Nxb7 20. Ng3 f6 21. Re7 fxg5 22. fxg5 Rfe8 23. Rfe1 Qf3+ 24. Kg1 Rxe7 25. Rxe7 Bf8 26. Qe2 Qc6 27. Qe6+ Qxe6 28. Rxe6 Kf7 29. Rf6+ Kg7 30. Ne4 Rb8 31. Bg3 Re8 32. Rf1 Re6 33. Bb8 Be7 34. Bxa7 h5 35. gxh6+ Kh7 36. Rf7+ Kxh6 37. b3 Nd8 38. Rxe7 Rxe7 39. Bxb6 Nb7 40. Kf2 Re5 41. a3 Rh5 42. Kg3 Re5 43. Kf3 Kh5 44. a4 Rf5+ 45. Kg3 g5 46. a5 Nxa5 47. Bxa5 c4 48. Bb6 cxb3 49. cxb3 g4 50. Nf2 Rf3+ 51. Kg2 Rf8 52. b4 Rb8 53. Bc5 Kh4 54. Ne4 Kh5 55. Kg3 Rb7 56. Nf6+ Kg6 57. Nxg4 Kf5 58. h3 Rb8 59. Nh2 Rg8+ 60. Kf2 Rh8 61. Kg2 Ra8 62. Nf3 Ra2+ 63. Kg3 Ra3 64. Nxd4+ Kg6 65. b5 Rxd3+ 66. Kh4 Rc3 67. b6 Rxc5 68. b7 Rh5+ 69. Kg3 Rh8 70. Nc6 Kg5 71. b8=Q Rxb8 72. Nxb8 Kf5 73. h4 Kg6 74. Kg4 Kg7 75. Kg5 Kh7 76. h5 Kg8 77. Kg6 Kh8 78. Nd7 Kg8 79. h6 Kh8 80. Ne5 Kg8 81. h7+ Kf8 82. Kf6 Ke8 83. h8=Q# 1-0 It will likely be a rather long time before the Lisp based portion of the system can play as well as does the toolkit's A/B searcher. This is because there is a large amount of chess specific knowledge in the form of patterns that must be added into the program's knowledge base, and this will take many months. As that knowledge is added, the role of the toolkit's A/B searcher will be gradually reduced until the point when it is called only when all planning and cognitive plan exploration fails to find a move. When such failure occurs, it means that I have to add yet more patterns to fix the hole. As you suggest, it can be useful to play the cognitive search against the toolbox A/B searcher as a way to measure relative strength. In fact, there is already a command, "set search selection strategy", in the toolbox's interactive command processor to do just this: ssss <White search selection strategy> <Black search selection strategy> Example: ssss iter cogn In the above, "iter" means iterative A/B and "cogn" means cognitive planner. I sometimes think of them as "Flash" and "Batman", but that's a topic for another post. And while the program has this little schizophrenia option, I'd rather have it play a variety of opponents as that will better demonstrate any weaknesses caused by missing knowledge. That such competition also establishes a rating is a side effect. You mention that some of the ideas appear futuristic. Perhaps, but talk of plan formation and exploration date back to the earliest days of computer chess. If you study the papers on the topic from the 1950s and 1960s, you'll find that nearly each one of them had one or more truly new ideas and that the thought of mimicking human behavior in move selection was always a motivating concept. I want a return to the glory days of computer chess research where real discoveries can be made and where the results aren't measured in Elo points but rather in the number of new research projects in the field and in other fields.
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.