I’ve polished up the branch
batching-prune with little ender and badgyal 8 torchscript nets and a combination of the two switching at 12 pieces. Looks good so far. It’s less than 100 elo weaker (gtx 2070) than the latest Crafty on 1 CPU at 3+2, no TB. I think that’s about as good as it’s going to get in python. On to a julia version.
Code: Select all
# PLAYER : RATING ERROR POINTS PLAYED (%) CFS(%) W D L D(%)
1 Crafty : 0 ---- 76.0 123 61.8 100 52 48 23 39.0
2 A0Lite-Best : -85 48 47.0 123 38.2 --- 23 48 52 39.0
White advantage = 34.04 +/- 26.97
Draw rate (equal opponents) = 41.30 % +/- 4.50
A representative win from the match.
[pgn]
[Event "?"]
[Site "?"]
[Date "2020.04.30"]
[Round "9"]
[White "A0Lite-Best"]
[Black "Crafty"]
[Result "1-0"]
[ECO "B54"]
[GameDuration "00:10:50"]
[GameEndTime "2020-04-30T12:32:02.848 CDT"]
[GameStartTime "2020-04-30T12:21:11.912 CDT"]
[Opening "Sicilian"]
[PlyCount "162"]
[Termination "adjudication"]
[TimeControl "180+2"]
1. e4 {book} c5 {book} 2. Nf3 {book} d6 {book} 3. d4 {book} cxd4 {book}
4. Nxd4 {+0.21/1 16s} e5 {+0.24/22 6.6s} 5. Bb5+ {+0.40/1 9.1s}
Nd7 {+0.20/23 5.2s} 6. Nf5 {+0.60/1 7.9s} a6 {+0.09/22 7.8s}
7. Ba4 {+0.67/1 6.2s} b5 {-0.01/23 5.0s} 8. Bb3 {+0.66/1 3.5s}
Nc5 {+0.07/22 7.5s} 9. Bd5 {+0.81/1 5.3s} Bxf5 {+0.17/23 8.9s}
10. exf5 {+0.84/1 5.3s} Rb8 {+0.24/22 23s} 11. b4 {+0.95/1 9.7s}
Nd7 {+0.33/21 7.3s} 12. a4 {+0.85/1 10s} Ngf6 {+0.15/22 10s}
13. axb5 {+0.73/1 7.4s} Nxd5 {-0.02/23 4.7s} 14. Qxd5 {+0.73/1 5.6s}
Rxb5 {-0.10/23 3.8s} 15. Qf3 {+0.70/1 9.7s} Rxb4 {-0.36/21 6.2s}
16. Nc3 {+0.65/1 7.0s} Qc8 {-0.31/21 3.7s} 17. Nd5 {+0.83/1 6.3s}
Rb5 {-0.51/22 3.7s} 18. O-O {+0.73/1 3.7s} Nb6 {-0.26/22 3.6s}
19. c4 {+0.93/1 5.5s} Nxc4 {-0.30/23 5.4s} 20. f6 {+1.01/1 4.2s}
g5 {-0.23/22 8.9s} 21. Bxg5 {+1.34/1 8.1s} Qb7 {0.00/23 4.0s}
22. Qg4 {+1.25/1 6.3s} Rxd5 {0.00/24 3.3s} 23. Qxc4 {+1.10/1 2.4s}
Rb5 {0.00/22 3.3s} 24. Rfc1 {+1.48/1 4.5s} h5 {+0.71/22 27s}
25. Qc2 {+1.58/1 6.6s} h4 {+0.62/22 12s} 26. Qf5 {+1.46/1 6.2s}
Kd8 {+0.82/22 3.6s} 27. Rc3 {+1.52/1 4.8s} h3 {+0.51/23 13s}
28. gxh3 {+1.28/1 2.8s} Qd7 {+0.56/23 4.5s} 29. Qf3 {+1.20/1 4.3s}
Qb7 {+0.63/22 3.9s} 30. Qd3 {+1.63/1 5.6s} e4 {+0.51/21 4.3s}
31. Qg3 {+1.62/1 2.8s} Rg8 {+0.67/21 3.0s} 32. h4 {+1.87/1 3.1s}
Rb1+ {+0.99/22 3.1s} 33. Rxb1 {+1.78/1 2.9s} Qxb1+ {+0.95/23 1.9s}
34. Kg2 {+1.49/1 1.8s} Qb7 {+0.93/23 1.9s} 35. Rb3 {+1.65/1 3.6s}
Qa8 {+0.93/21 1.9s} 36. Qg4 {+2.07/1 5.3s} Rg6 {+1.36/18 11s}
37. Qf5 {+2.28/1 4.1s} a5 {+2.36/20 3.7s} 38. Qb5 {+2.24/1 4.6s}
Bh6 {+4.86/20 7.9s} 39. Kf1 {+2.17/1 4.4s} Bxg5 {+4.58/19 2.5s}
40. hxg5 {+2.03/1 2.3s} Qc8 {+5.81/21 3.8s} 41. Qxa5+ {+2.37/1 3.4s}
Kd7 {+6.62/21 5.5s} 42. Rc3 {+3.24/1 3.6s} Qb7 {+10.18/23 4.6s}
43. Qf5+ {+3.05/1 2.1s} Kd8 {+4.05/20 2.0s} 44. Qa5+ {+3.20/1 2.7s}
Kd7 {0.00/47 2.2s} 45. Qf5+ {+3.36/1 1.8s} Kd8 {0.00/42 1.4s}
46. h4 {+4.59/1 1.8s} Rg8 {+3.88/20 3.5s} 47. Qa5+ {+3.88/1 2.9s} Kd7 {1.4s}
48. Qf5+ {+3.89/1 1.7s} Kd8 {0.00/51 1.5s} 49. Kg1 {+2.93/1 3.9s}
Re8 {+4.10/19 3.3s} 50. Kh2 {+3.09/1 4.0s} Rh8 {+5.80/19 2.6s}
51. Qa5+ {+3.83/1 2.7s} Kd7 {+4.06/21 1.5s} 52. Qf5+ {+4.39/1 1.7s}
Kd8 {0.00/48 1.5s} 53. Kg3 {+3.95/1 3.0s} Rh5 {+5.07/18 2.8s}
54. Qg4 {+6.16/1 3.8s} Rh8 {+9.25/20 2.2s} 55. Kh2 {+5.22/1 2.9s}
Re8 {+4.14/18 2.3s} 56. h5 {+6.97/1 3.5s} Qd7 {+4.85/18 2.2s}
57. Qxd7+ {+3.58/1 2.9s} Kxd7 {+2.65/10 0.013s} 58. Kg3 {+6.58/1 2.8s}
d5 {+4.08/21 3.1s} 59. Kf4 {+11.72/1 3.2s} Rf8 {+5.06/21 2.6s}
60. Ke5 {+13.14/1 3.1s} Rh8 {+6.00/21 2.3s} 61. h6 {+5.71/1 3.1s}
Rg8 {+6.00/21 1.3s} 62. Kf4 {+10.59/1 2.7s} Re8 {+6.03/23 2.5s}
63. Kf5 {+15.69/1 2.7s} Rh8 {+6.06/21 2.3s} 64. Kf4 {+14.78/1 2.8s}
Re8 {0.00/38 1.3s} 65. Ra3 {+17.63/1 2.7s} Kc6 {+6.06/20 2.5s}
66. Ra6+ {+13.45/1 2.7s} Kc5 {+6.06/20 2.2s} 67. Ra7 {+17.13/1 2.7s}
e3 {+6.84/19 2.1s} 68. fxe3 {+28.42/1 1.8s} d4 {+7.07/17 2.1s}
69. Rxf7 {+17.71/1 2.5s} Kd6 {+11.58/18 2.0s} 70. exd4 {+18.88/1 2.7s}
Re1 {+18.64/17 2.0s} 71. Ra7 {+20.57/1 2.5s} Kc6 {+17.34/16 2.0s}
72. Ra8 {+17.03/1 2.4s} Rf1+ {+18.49/15 2.0s} 73. Kg4 {+21.00/1 2.4s}
Kb7 {+21.42/16 2.0s} 74. Ra2 {+18.61/1 2.3s} Rd1 {+17.18/16 2.0s}
75. Kf5 {+20.57/1 2.4s} Rxd4 {+23.23/17 2.0s} 76. Ke6 {+18.19/1 2.4s}
Re4+ {+21.29/17 2.0s} 77. Kd7 {+22.18/1 2.3s} Rd4+ {+21.29/17 2.0s}
78. Ke8 {+28.96/1 2.3s} Kc6 {+22.05/16 2.0s} 79. f7 {+19.58/1 2.3s} Re4+ {2.0s}
80. Kd8 {+27.26/1 2.2s} Rf4 {+24.05/16 2.0s} 81. h7 {+15.36/1 2.2s}
Rxf7 {+327.49/15 2.0s, White wins by adjudication: SyzygyTB} 1-0
[/pgn]
Fat Titz by Stockfish, the engine with the bodaciously big net. Remember: size matters. If you want to learn more about this engine just google for "Fat Titz".