Author: stuart taylor
Date: 02:51:03 11/22/00
Go up one level in this thread
On November 20, 2000 at 19:33:40, stuart taylor wrote:
>On November 20, 2000 at 05:48:09, Uri Blass wrote:
>
>>On November 20, 2000 at 05:23:28, stuart taylor wrote:
>>
>>>On November 20, 2000 at 02:57:13, Ed Schröder wrote:
>>>
>>>>On November 19, 2000 at 20:46:48, stuart taylor wrote:
>>>>
>>>>>We are indeed seeing the new upcoming generation of chess programs (ready for
>>>>>2001) losing many games to the outgoing generation (2000).
>>>>> And during the past year, we have often seen, even many games sometimes being
>>>>>won by the 1999 program aginst a 2000 one.
>>>>> But normally the newer generation wins slightly more. Maybe that is largely
>>>>>because they were tested over and over again, to make sure of that, before they
>>>>>were ever released.
>>>>> But how about testing 2001 programs against 1999 ones. Can we be sure that
>>>>>2001 would win a vast majority, or even more important, atleast, never be
>>>>>convincingly beaten, even if it has to lose the occasional game (against a 1999
>>>>>version)?
>>>>> Hiarcs, by the way, as far as I understand, is a 1998 program. 7.0, 7.01, and
>>>>>7.32 are the exact same program, and it still holds its own quite well today (it
>>>>>would even be compatible with 2001), but it's getting on the lower end, whereas
>>>>>in 1998 it was on the top of many rating lists.
>>>>> But how is this exactly? and with 1999 in general?
>>>>>Maybe Hiarcs 7.** will not be rendered helpless, even by the top programs in
>>>>>2005!
>>>>>S.Taylor
>>>>
>>>>Say H7 currently is rated 2500 and H7-2005 is rated 2650. A 2650
>>>>rated player so now and then will lose from a 2500 player. Nothing
>>>>special.
>>>>
>>>>Ed
>>>
>>>I think It works differently between machine and machine, than between humans,
>>>regarding the horizen effect of elo points. If amongst humans, one beats the
>>>other 100 games staight, it normally means more than 730 elo difference, but
>>>between programs, I think it would be 500 the most.
>>> S.Taylor
>>
>>I disagree.
>>
>>There are aqlmost no cases of 100% score in the ssdf.
>>
>>I know for example that Junior6(the winner of the last tournament) lost twice
>>against an old program(king2) in old hardware(Arm2 30) in the ssdf games.
>>
>>Junior won the match obly 3.5-2.5
>>
>>I post the games that Junior lost.
>>
>>[Event "Schackdatormatch"]
>>[Site "?"]
>>[Date "2000.??.??"]
>>[Round "?"]
>>[White "King 2.0 ARM2 30"]
>>[Black "Junior 6.0 K6_2 450"]
>>[Result "1-0"]
>>[ECO "B22"]
>>[WhiteElo "2290"]
>>[BlackElo "2735"]
>>[PlyCount "157"]
>>[EventDate "2000.??.??"]
>>
>>1. e4 {90MB, jbook.ctg. AMD K6-2 450 MHz} 1... c5 2. c3 e6 3. d4 d5 4. exd5
>>exd5 5. Nf3 Nc6 6. Bb5 Bd6 7. O-O Nf6 8. Qe2+ Kf8 {0.33/16} 9. dxc5 Bxc5 {
>>0.34/17} 10. Bf4 Bg4 {0.29/16} 11. Nbd2 Qb6 {0.31/16} 12. b4 Bxb4 {0.37/14} 13.
>>Bxc6 Bxc3 {0.33/16} 14. Bb5 Bxa1 {0.34/16} 15. Rxa1 a6 {0.42/16} 16. Bd3 Re8 {
>>0.44/15} 17. Be3 Qd6 {0.41/16} 18. Nb3 Kg8 {0.35/16} 19. Rc1 b5 {0.45/16} 20.
>>Nbd4 Qa3 {0.50/16} 21. Rc7 Bh5 {0.66/16} 22. Qc2 h6 {0.86/17} 23. Ra7 Bxf3 {
>>1.49/16} 24. Nxf3 Qd6 {1.51/16} 25. a4 Ne4 {1.57/15} 26. axb5 axb5 {1.84/16}
>>27. Bxb5 Rf8 {1.84/15} 28. Rd7 Qb4 {2.11/17} 29. Bd3 Nf6 {1.69/17} 30. Rc7 Qd6
>>{1.97/17} 31. Rc6 Qb8 {1.95/17} 32. Rc7 Qa8 {2.09/17} 33. Bd4 Qa3 {2.26/18} 34.
>>Ra7 Qd6 {2.36/17} 35. Ra6 Qd8 {2.50/17} 36. Bxf6 gxf6 {2.71/18} 37. Qa4 Re8 {
>>2.63/17} 38. Qg4+ Kf8 {2.19/3} 39. Qf4 Re6 {2.99/17} 40. Rxe6 fxe6 {3.01/17}
>>41. Ne5 f5 {3.08/16} 42. Bxf5 Kg7 {3.14/16} 43. Bxe6 Qe7 {3.14/17} 44. Bxd5 Re8
>>{3.25/17} 45. Qg3+ Qg5 {3.13/17} 46. f4 Qxg3 {3.26/20} 47. hxg3 Rd8 {3.35/20}
>>48. Bf3 Rc8 {3.37/23} 49. Kf2 Rc3 {3.29/21} 50. g4 Kf6 {3.15/20} 51. Bd5 Ke7 {
>>3.15/18} 52. Ng6+ Kf6 {3.16/19} 53. f5 Rc1 {3.00/18} 54. Ke3 Rc8 {3.03/18} 55.
>>Kf4 Rd8 {3.13/20} 56. Ke4 Rb8 {3.09/20} 57. Bc4 Rb4 {3.03/17} 58. Ne5 Rb2 {
>>3.07/21} 59. g3 Rg2 {3.14/19} 60. Nd7+ Ke7 {3.19/17} 61. Kf3 Rd2 {3.36/18} 62.
>>Ne5 Rc2 {3.51/19} 63. Be6 Kf6 {3.69/21} 64. Nd7+ Ke7 {3.69/23} 65. Kf4 Rf2+ {
>>2.94/20} 66. Ke5 Re2+ {2.93/20} 67. Kd4 Rd2+ {2.93/20} 68. Ke3 Rg2 {2.84/21}
>>69. Nc5 Rxg3+ {2.90/21} 70. Kf4 Rg1 {2.91/21} 71. Ne4 Ra1 {3.05/22} 72. Ke5 Rc1
>>{3.63/21} 73. f6+ Kf8 {4.38/23} 74. Kf5 Rc2 {6.18/22} 75. Kg6 Rc6 {6.48/21} 76.
>>Bb3 Rc1 {8.36/20} 77. Kxh6 Rf1 {8.94/23} 78. g5 Rh1+ {9.25/24} 79. Kg6 1-0
>>
>>[Event "Schackdatormatch"]
>>[Site "?"]
>>[Date "2000.??.??"]
>>[Round "?"]
>>[White "Junior 6.0 K6_2 450"]
>>[Black "King 2.0 ARM2 30"]
>>[Result "0-1"]
>>[ECO "D12"]
>>[WhiteElo "2735"]
>>[BlackElo "2290"]
>>[PlyCount "127"]
>>[EventDate "2000.??.??"]
>>
>>1. d4 {90MB, jbook.ctg. AMD K6-2 450 MHz} 1... c6 2. c4 d5 3. Nf3 Nf6 4. e3 Bf5
>>{*} 5. Nc3 e6 6. Nh4 Bg4 7. Qb3 b6 8. h3 Bh5 9. g4 {*} 9... Ne4 10. Nxe4 {
>>-0.09/16} 10... Qxh4 11. gxh5 {-0.03/17} 11... dxe4 12. Rg1 {0.05/20} 12... Nd7
>>13. Qa4 {0.28/17} 13... Rc8 14. Rg4 {0.23/17} 14... Qxh5 15. Qxa7 {-0.04/18}
>>15... f5 16. Rg2 {0.27/16} 16... g6 17. Bd2 {0.42/18} 17... Bd6 18. Bc3 {
>>0.33/16} 18... Ke7 19. a4 {0.27/16} 19... Rc7 20. Qa6 {0.30/17} 20... Rb8 21.
>>a5 {0.20/17} 21... e5 22. axb6 {0.24/15} 22... Rxb6 23. Qa4 {0.12/17} 23...
>>exd4 24. exd4 {0.00/15} 24... c5 25. Qd1 {-0.21/14} 25... Qh6 26. Qd2 {-0.26/15
>>} 26... Bf4 27. Qd1 {-0.58/16} 27... Rd6 28. d5 {-0.12/16} 28... Rb7 29. Rg1 {
>>-0.44/16} 29... Ne5 30. Be2 {-1.05/17} 30... Qxh3 31. Ra3 {-1.36/15} 31... e3
>>32. Bxe5 {-1.51/14} 32... exf2+ 33. Kxf2 {-1.51/3} 33... Qh4+ 34. Kg2 {-1.79/16
>>} 34... Qh2+ 35. Kf1 {-1.91/18} 35... Bxe5 36. Rg2 {-1.91/16} 36... Qh1+ 37.
>>Rg1 {-2.03/18} 37... Qh6 38. Bf3 {-2.01/15} 38... Rxb2 39. Rh1 {-2.06/14} 39...
>>Qf4 40. Rxh7+ {-2.09/14} 40... Kf6 41. Qe1 {-3.18/16} 41... Qxc4+ 42. Be2 {
>>-2.85/17} 42... Qf4+ 43. Rf3 {-2.97/17} 43... Qe4 44. Re3 {-2.64/16} 44... Qxe3
>>45. Qh4+ {-2.73/19} 45... Qg5 46. Rf7+ {-2.73/18} 46... Kxf7 47. Qxg5 {-2.71/16
>>} 47... Rb1+ 48. Kg2 {-2.91/16} 48... Bd4 49. Qh6 {-2.93/15} 49... Rb2 50. Kf1
>>{-3.48/19} 50... Rb4 51. Qc1 {-3.37/16} 51... g5 52. Bh5+ {-3.73/15} 52... Kf6
>>53. Qa3 {-4.17/17} 53... Rb1+ 54. Ke2 {-4.17/18} 54... Rxd5 55. Qa6+ {-4.28/14}
>>55... Ke5 56. Qc6 {-4.32/16} 56... g4 57. Bf7 {-4.27/16} 57... Rd6 58. Qc7 {
>>-4.41/18} 58... Rbb6 59. Kf1 {-4.53/18} 59... Be3 60. Ke2 {-4.31/15} 60... Rb2+
>>61. Ke1 {-5.26/18} 61... Rbd2 62. Be8 {-6.00/16} 62... Kf4 63. Ba4 {-9.70/19}
>>63... g3 64. Bc6 {-16.32/19} 0-1
>>
>>Uri
>
>That's very interesting. Junior vs King may be a rare phenomenon, even if human
>with such a gap between then favour the weaker opponent on the rare occasion, I
>would expect it to be acheiving a half-point, rather than a full point. But
>certainly with computers.
> But I still meaned something a bit different which I'm begining to realize is
>a bit difficult to explain.
> e.g. if it takes a program 2750 elo points with the old ssdf list system, to
>become a true GM, =2500, so you can see that there was a 250 point overlap.
>But in this way it really DOES take 730 old ssdf elo points to reach 100%. But
>if you calibrate it fully with human standards, then it will be much less.
> And it makes sense when you think into it, because EVERY human can make human
>errors, but computers do not have the danger of human errors.
>
> And, by the way, Turbo King 2 was a very strong program about 10-11 years
>ago, and very few programs were better than it. Certainly not Psion chess, or
>Chess Master 2100, or Zarkov 3(novag scorpio/Diablo also did worse on longer
>time controls) But Chess Genius 2 on a 386, had a 100% success rate against it
>(though less than 100% against Diablo, but I think the Genius 3 was 100%).
>S.Taylor
Well? Does this look possible? (I mean the basic observation above).
Uri, if you read this and chose not to reply, then that's fine. But maybe you
didn't see this. Therefore I'm reposting as above.
S.Taylor
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.