[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [computer-go] GnuGo self testing results
> According to autoplay tests I did a month ago on GnuGo 3.5.3 with komi=6.5,
> size 9x9, level 10:
> Number of played games: 3000 (restarting GNUGo each time with a different
> seed)
> Number of different games: 584 (big surprise; many seeds give exactly the
> same game)
> Black won 287 of these 584 games (49%).
> If anyone is interested, the Java GTP autoplay class and SGF game records
> (with territory marked) are available.
That's interesting. When I test my own program, a komi of 3.5 gets
closest to even results on 9x9. That may be because my program is
weaker which I can only guess would reduce the advantage of black.
My gnugo tests are with 19x19. I'm trying to test the uniquness of the
games by looking at limited sequences of moves which is the easiest
way I can do this based on how I report the results of the game.
Since I am testing 10 different levels, it's hard to say how unique
the games are since I would expect a lot of variety playing so many
levels. Some of the games stayed the same for many moves before
changing. This could be because 1 level played another level and they
are similar enough to play the same for many moves.
But in general, once the first 25 moves for each side were played,
there were no duplicate move sequences and I didn't impose any seeds.
However, since I'm using so many players (10) most of them have met
the same opoonents less than 15 times with the same color, so it's
easy to imagine that I will need a lot more games to see a lot of
duplicates. I might try a quick test with just 2 players to get a
feel of how often I will get duplicate games.
I am planning to get the latest version of gnugo and play both
versions against each other. Two versions with 10 levels each would
be a pretty big round robin so I might limit the levels to even levels
to get 10 unique players. I'll bump komi up to 6.5 for the next test.
- Don
drd@garfield:~/Games/mygo/CHAMPIONSHIP$ grep '^1\.' gg.pgn | wc
991 14865 50515
drd@garfield:~/Games/mygo/CHAMPIONSHIP$ grep '^1\.' gg.pgn | sort | uniq | wc
786 11790 40034
drd@garfield:~/Games/mygo/CHAMPIONSHIP$ grep '^6\.' gg.pgn | sort | uniq | wc
944 14160 49209
drd@garfield:~/Games/mygo/CHAMPIONSHIP$ grep '^11\.' gg.pgn | sort | uniq | wc
977 14655 55105
drd@garfield:~/Games/mygo/CHAMPIONSHIP$ grep '^16\.' gg.pgn | sort | uniq | wc
987 14805 55590
drd@garfield:~/Games/mygo/CHAMPIONSHIP$ grep '^21\.' gg.pgn | sort | uniq | wc
990 14850 55677
drd@garfield:~/Games/mygo/CHAMPIONSHIP$ grep '^26\.' gg.pgn | sort | uniq | wc
991 14865 55641
[Event "Autotest Sequence"]
[Site "Local Site"]
[Date "2004.02.01"]
[Round "1"]
[Black "g10"]
[BlackTime "00:05:27"]
[White "g3"]
[WhiteTime "00:01:58"]
[Boardsize "19x19"]
[Komi "5.5"]
[Rules "Tromp/Taylor"]
[Final "201-160"]
[Result "1-0"]
1. q16 d4 2. q4 e17 3. c10 o17 4. f3 o3 5. c17 d15
6. b15 l17 7. r14 q18 8. r17 h3 9. c3 c4 10. d3 e4
11. f2 g5 12. l4 o5 13. q6 q2 14. r3 l2 15. r10 l3
16. c13 h17 17. o15 r2 18. c7 e13 19. m15 o7 20. q8 p8
21. o11 l15 22. l14 s3 23. s4 s2 24. q9 k15 25. j13 h14
26. r18 r5 27. r4 q5 28. s5 p6 29. r6 m8 30. p18 o18
31. l10 p17 32. k4 j5 33. j9 m5 34. d18 h2 35. g19 m10
36. m11 n10 37. n11 g18 38. h19 h13 39. j7 j18 40. m16 g11
41. j11 h12 42. p5 j12 43. o4 n4 44. p4 l11 45. k11 l9
46. l12 b6 47. b7 c5 48. k9 k8 49. j8 k7 50. b4 j19
51. f18 f17 52. e18 q19 53. j14 j15 54. r19 b5 55. b3 a7
56. a8 a6 57. c8 j6 58. o10 o9 59. p9 d12 60. n9 c12
61. f7 b13 62. m9 c15 63. b14 c14 64. a13 b12 65. c16 b9
66. b8 b10 67. c9 o8 68. f10 e6 69. n8 l8 70. m7 k6
71. n6 n5 72. o6 a12 73. m6 l5 74. g10 g7 75. g8 f6
76. h7 e7 77. g6 a14 78. e8 b16 79. b17 a15 80. a17 h6
81. g7 f4 82. m17 m18 83. l16 k17 84. g2 e3 85. e2 k13
86. e11 k14 87. l13 e12 88. h11 f11 89. d11 n17 90. q17 p19
91. a16 b15 92. k16 j16 93. c11 a10 94. b11 a11 95. d7 d6
96. h1 j1 97. g1 g3 98. t3 t2 99. t4 p3 100. p16 d16
101. n16 a4 102. a3 a5 103. p7 o16 104. d17 h18 105. f19 k12
106. q3 c6 107. l6 a9 108. l7 d13 109. q7 f12 110. p15 e16
111. m14 g17 112. q10 j17 113. m12 l18 114. e10 n18 115. h10 o19
116. g9 j4 117. m10 j3 118. e9 k3 119. n15 g4 120. p14 k5
121. q14 m4 122. b2 n3 123. c1 p2 124. d2 j2 125. b18 pass
126. c19 pass 127. b19 pass 128. b1 pass 129. n7 pass 130. pass
_______________________________________________
computer-go mailing list
computer-go@xxxxxxxxxxxxxxxxx
http://computer-go.org/mailman/listinfo/computer-go