Glass 1.5 Gauntlets

[Aser Huerga]

Glass 1.5 Gauntlets

Hardware: Intel(R) Core(TM) i7 CPU 950 @ 3.07GHz8x 3074 MHz with 4.096 MB Memory
Operating system: Microsoft Windows XP 64 Bit Professional Service Pack 2 (Build 3790)
All engines 1 CPU
Arena 2.0.1 GUI
128 Hash Tables
3-4-5 EGTBs
Ponder off
Book GrandPQRS PGN up to 8 moves (151.320 starting positions from years 1981-2009, games selected by Norm Pollock), each pair plays the same position with switched colours.
40 moves in 23 minutes repeating (adapted to the CCRL 40/40 standard)

Gauntlet I:

Code: Select all

1   Glass 1.5          -2438- 41.0/120
2   Xpdnt 091007        2603  17.5/24
3   SpiderChess 070603  2582  17.0/24
4   Muse 0.899b         2530  15.5/24
5   Gosu 0.16           2500  15.0/24
6   Kiwi 0.6d           2553  14.0/24

(All ELOs referred to the last Chess Computer Rating List when tournament starts, except -XXXX- one, wich is referred to performance).

Statistics:

Code: Select all

Glass 1.5                     : 120 (+ 26,= 30,- 64), 34.2 %

Xpdnt 091007                  :  24 (+  3,=  7,- 14), 27.1 %
SpiderChess 070603            :  24 (+  3,=  8,- 13), 29.2 %
Kiwi 0.6d                     :  24 (+  8,=  4,- 12), 41.7 %
Muse 0.899b                   :  24 (+  5,=  7,- 12), 35.4 %
Gosu 0.16                     :  24 (+  7,=  4,- 13), 37.5 %

Games: http://kirill-kryukov.com/chess/discuss ... f=7&t=5040

[Graham Banks]

Looking forward to seeing the results.

[Aser Huerga]

By the moment it seems Glass 1.5 is not playing better than the earlier version, but there are few games by now ... let's see how Glass 1.5 performs in an easier gauntlet.

[Aser Huerga]

In my modest opinion, maybe Glass 1.5 prunning is too much agressive ...

Gauntlet II:

Code: Select all

1   Glass 1.5     -2451- 61.0/120
2   Tytan 9.3      2461  15.0/24
3   Aice 0.99.2    2443  13.5/24
4   Phalanx XXII   2492  13.0/24
5   Ares 1.003a    2435  10.5/24
6   AliChess 4.20  2403  7.0/24

(All ELOs referred to the last Chess Computer Rating List when tournament starts, except -XXXX- one, wich is referred to performance).

Statistics:

Code: Select all

Glass 1.5                     : 120 (+ 48,= 26,- 46), 50.8 %

Phalanx XXII                  :  24 (+  8,=  6,- 10), 45.8 %
Tytan 9.3                     :  24 (+  6,=  6,- 12), 37.5 %
Aice 0.99.2                   :  24 (+  8,=  5,- 11), 43.8 %
Ares 1.003a                   :  24 (+ 11,=  5,-  8), 56.2 %
AliChess 4.20                 :  24 (+ 15,=  4,-  5), 70.8 %

Games: http://kirill-kryukov.com/chess/discuss ... 725#p52725

[Edmund]

Aser, thanks for your tests and advice. Currently we indeed have some promising preliminary results limiting the aggressivenes of Late Move Reduction.

I also had a quick look at the games and noticed some strange behaviour in Round 2 Ares vs Glass. In the drawn endgame Glass takes very long to get to a reasonable depth and returns very high scores. On my machine it takes hardly any time to get to the maximum depth of 100 with a score of 0 from depth 25 onwards eg. in the position before move 69. Could it be that there are initialization problems of the egtbs?

regards,
Edmund

[Event "Glass 1.5 Gauntlet IIa"]
[Site "PC-8A284AD7BDD5"]
[Date "2010.04.05"]
[Round "2"]
[White "Ares 1.003a"]
[Black "Glass 1.5"]
[Result "1/2-1/2"]
[BlackElo "1"]
[ECO "B49"]
[Opening "Sizilianisch"]
[Variation "Paulsen-Variante, 6.Be3 a6 7.Be2 Nf6 8.O-O b5"]
[TimeControl "40/1380:40/1380:40/1380"]
[Termination "normal"]
[PlyCount "151"]
[WhiteType "human"]
[BlackType "human"]

1. e4 c5 2. Nf3 e6 3. c4 a6 4. Be2 Qc7 5. Nc3 Nf6 6. O-O Nc6 7. d4 cxd4 8.
Nxd4 Bb4 9. Bg5 {+0.44/14 67} Bxc3 {-0.05/16 21} 10. Bxf6 {+0.31/16 44}
Bxb2 {-0.02/17 29} 11. Nxe6 {+0.43/16 31} dxe6 {+0.01/17 31} 12. Bxb2
{+0.28/17 57} O-O {+0.01/16 24} 13. Qb3 {+0.20/15 35} e5 {+0.03/15 35} 14.
Rfd1 {+0.17/13 27} Be6 {+0.06/15 77} 15. Qe3 {+0.15/13 36} Na5 {+0.06/15
32} 16. Ba3 {+0.05/14 26} Rfd8 {+0.14/16 32} 17. Rxd8+ {+0.05/15 49} Qxd8
{+0.12/15 24} 18. Rd1 {-0.14/13 24} Qc8 {+0.29/16 24} 19. Qg3 {-0.01/15 29}
Bxc4 {+0.52/17 46} 20. Rc1 {-0.29/14 23} b5 {+0.07/16 49} 21. Qxe5
{-0.03/15 23} Qe8 {+0.04/17 64} 22. Qb2 {+0.11/14 22} Bxe2 {+0.05/16 29}
23. Qxe2 {+0.21/11} Nc4 {+0.01/16 23} 24. Bb4 {+0.21/14 41} Qe6 {-0.05/15
42} 25. Bc3 {+0.25/14 33} Re8 {-0.04/16 54} 26. f3 {+0.31/12 20} Rd8
{0.00/16 54} 27. Rd1 {+0.31/14 39} Qe7 {0.00/16 38} 28. Rxd8+ {+0.36/12 17}
Qxd8 {+0.10/16 33} 29. f4 {+0.31/14 19} h5 {+0.12/16 35} 30. e5 {+0.25/12
18} Qb6+ {+0.04/16 84} 31. Qf2 {+0.14/13 8} Qc6 {+0.11/16 27} 32. Qd4
{+0.27/14 18} Qg6 {0.00/16 26} 33. a4 {+0.23/13 20} Kh8 {+0.30/16 78} 34.
h3 {+0.33/12 15} Qg3 {+0.22/17 33} 35. Bb4 {0.00/13 23} Qe3+ {0.00/16 45}
36. Qxe3 {+0.16/18 14} Nxe3 {+0.20/22 47} 37. axb5 {+0.45/17 15} axb5
{+0.21/21 49} 38. Bd2 {+0.26/17 15} Nd5 {0.00/20 25} 39. Kf1 {+0.42/17 25}
Kh7 {0.00/21 54} 40. Ke2 {+0.47/16 12} Kg6 {-0.06/21 100} 41. g4 {+0.53/17
46} hxg4 {0.00/21 31} 42. hxg4 {+0.36/19 81} f5 {0.00/20 25} 43. g5
{+0.65/17 20} b4 {0.00/25 28} 44. Kd1 {+0.39/21 43} Kf7 {0.00/27 18} 45.
Kc2 {+1.44/21 65} Ke7 {0.00/29 68} 46. Kb3 {+1.47/20 59} Kd7 {-0.43/25 19}
47. Kc4 {+1.42/20 71} Kc6 {-1.18/25 20} 48. Bc1 {+0.31/20 42} g6 {-1.23/25
18} 49. Bd2 {+1.53/21 37} b3 {-1.60/27 22} 50. Kxb3 {+1.56/21 42} Kc5
{-1.84/27 27} 51. Bc1 {+1.55/21 34} Nc7 {-1.82/26 25} 52. Kc3 {+1.55/22 38}
Nd5+ {-1.80/27 27} 53. Kd3 {+1.52/22 45} Nb4+ {-1.80/26 71} 54. Ke3
{+1.54/21 32} Nd5+ {-1.72/27 22} 55. Ke2 {+1.52/20 30} Nc3+ {-1.82/26 48}
56. Kd3 {+1.54/20 28} Nb5 {-1.82/26 31} 57. Be3+ {+1.53/22 29} Kd5
{-1.82/28 31} 58. Bb6 {+1.53/22 32} Na3 {-1.82/25 22} 59. Kc3 {+1.55/21 28}
Nc4 {-1.78/26 26} 60. Bf2 {+1.54/21 27} Nxe5 {-1.45/28 38} 61. fxe5
{+1.86/25 30} Kxe5 {-1.39/35 22} 62. Be3 {+1.94/25 23} f4 {-0.46/36 19} 63.
Bd4+ {+1.86/21 49} Kf5 {-0.46/37 21} 64. Bf6 {+1.72/26 25} f3 {-0.46/40 26}
65. Kd2 {+1.80/27 25} Kf4 {0.00/36 28} 66. Ke1 {+0.59/26 23} Ke3 {0.00/40
24} 67. Bb2 {+1.70/25 27} f2+ {0.00/46 26} 68. Kf1 {+3.42/11} Kf4 {0.00/46
27} 69. Bf6 {+0.62/22 41} Kf3 {0.00/48 53} 70. Be7 {+0.56/26 19} Ke3
{0.00/46 36} 71. Bd6 {+0.62/26 38} Kf3 {0.00/48 65} 72. Bb4 {+1.80/20 32}
Kf4 {0.00/45 28} 73. Be7 {+0.56/25 18} Ke3 {0.00/100 3} 74. Bd6 {+0.60/26
18} Kf3 {0.00/100 3} 75. Be5 {+1.64/25 33} Ke3 {0.00/47 33} 76. Bd6
{+1.77/27 27 3-fold repetition} 1/2-1/2

[Aser Huerga]

Hi Edmund,

I suppose the difference could be that I'm running three different matches (using 3 cores) at the same time (in my i7 950 testings), so resources related with EGTB access has to be shared, and it's not as fast as if one single core is running.
However, I'm using high speed USB flash drive to alocate EGTBs and 6GB of RAM, and in CCRL we think this way, the impact on results is insignificant.

Congratulations for your program, and go ahead with its improve!

[Edmund]

I am sorry, my mistake. I mixed up the two sides. Not Glass but Ares was displaying the inappropriate scores in the quoted game. The egtb seem to be working alright.