new SF skill level: user feedback please

General discussion about computer chess...
Post Reply
lucasart
Posts: 201
Joined: Mon Dec 17, 2012 1:09 pm
Contact:

new SF skill level: user feedback please

Post by lucasart » Sat Jan 31, 2015 1:14 am

Recently there were important changes that affect the skill levels in SF:
  • pruning at PV nodes on 14-jan
  • allowing SF to blunder at weak levels on 30-jan
I am talking about SF 20150130 only. So please do not use anything else to spread confusion here (SF 5 and SF 6 skill levels are very different).

I've run some measurement in self-play, using a shallow book (2.5 moves), and 10k games for each level. Note that the only way to measure this correctly is to use fixed depth=1+level for testing (after that it's just wasted testing time, and below you pollute the measure by time being reached before max depth).

Code: Select all

Level(L)  elo(L)-elo(L-1)
-------------------------
1         131
2         107
3         146
4         202
5         221
6         204
7         189
8         188
9         154
10        125
Please understand that comp vs. comp games are very different from human games. When level n+1 plays vs. level n, what you get are crazy sharp tactical game, in which level n+1 usually prevails. Human vs. machine games are very different. So I would expect that elo gaps between levels, measured in human terms (assuming it was possible to measure that precisely), to be somewhat compressed compared to the above.

What I am interested to know is the compression factor and the elo offset, to translate the above into human scale. So it would be interesting if people could post their real elo, and their experience against the new skill levels.

For example, I am approx 1600 elo, and after playing a few games, I would place myself somewhere between level 4 and level 5, but closer closer to level 5. And please be honest, and don't just look at your best game: elo is about averages (no takeback), including all those blundered games you play.
"Talk is cheap. Show me the code." -- Linus Torvalds.

lucasart
Posts: 201
Joined: Mon Dec 17, 2012 1:09 pm
Contact:

Re: new SF skill level: user feedback please

Post by lucasart » Sat Jan 31, 2015 2:32 am

Here's a game I played against level 5:
  • SF was playing without book, with hash=16 and depth=1+level=6 (no point giving more time as there's a hard limit of depth=1+level in the code).
  • I was playing in 10'+10", no takebacks.
[White "lucas"]
[Black "SF 20150130 level 5"]
[Result "1-0"]
[TimeControl "600+10"]

1. d4 e6 {-0.51/6 0.009s} 2. c4 {1.9s} d5 {+0.10/6 0.010s} 3. Nc3 {2.2s}
Nf6 {-0.23/6 0.021s} 4. e3 {1.3s} Be7 {+0.01/6 0.009s} 5. Nf3 {1.7s}
Nc6 {-0.15/6 0.010s} 6. Be2 {8.1s} Ne4 {+0.01/6 0.008s} 7. Qc2 {9.3s}
Nf6 {+0.01/6 0.016s} 8. b3 {9.0s} Bd7 {+0.05/6 0.009s} 9. O-O {5.0s}
O-O {-0.53/6 0.007s} 10. Bb2 {4.7s} a6 {-0.56/6 0.005s} 11. e4 {25s}
dxe4 {+3.08/6 0.007s} 12. Nxe4 {1.7s} Nxe4 {+0.90/6 0.007s} 13. Qxe4 {1.5s}
f5 {+5.93/6 0.008s} 14. Qc2 {30s} Nb4 {+6.39/6 0.014s} 15. Qd2 {5.9s}
Be8 {-0.89/6 0.009s} 16. a3 {6.0s} Nc6 {-1.70/6 0.005s} 17. d5 {17s}
Bh5 {-1.50/6 0.005s} 18. Rfd1 {7.3s} exd5 {-1.57/6 0.013s} 19. Qxd5+ {1.9s}
Qxd5 {-0.65/6 0.006s} 20. Rxd5 {4.5s} Rad8 {+1.69/6 0.009s} 21. Rad1 {5.2s}
g6 {-0.47/6 0.006s} 22. Rd7 {24s} Rxd7 {-0.21/6 0.009s} 23. Rxd7 {2.0s}
Rb8 {-1.18/6 0.007s} 24. Rxc7 {13s} a5 {-1.74/6 0.008s} 25. Kf1 {34s}
Kf8 {-1.67/6 0.008s} 26. Ng1 {3.5s} Bxe2+ {-1.77/6 0.005s} 27. Nxe2 {1.2s}
h6 {-1.69/6 0.005s} 28. Nf4 {6.5s} Bd6 {-1.94/6 0.005s} 29. Nxg6+ {4.8s}
Ke8 {-1.74/6 0.001s} 30. Rh7 {22s} b5 {-2.70/6 0.007s} 31. Rh8+ {17s}
Kf7 {-2.70/6 0.001s} 32. Rxb8 {2.0s} Nxb8 {-3.04/6 0.006s} 33. Nh4 {25s}
Ke6 {-1.87/6 0.004s} 34. g3 {14s} b4 {-1.50/6 0.004s} 35. axb4 {6.7s}
axb4 {-2.18/6 0.003s} 36. f4 {9.7s} Bc5 {-1.59/6 0.003s} 37. Ke2 {9.4s}
Nc6 {-2.10/6 0.004s} 38. Nf3 {5.9s} Bb6 {-2.28/6 0.003s} 39. Ne5 {11s}
Nxe5 {-1.33/6 0.003s} 40. Bxe5 {1.6s} Bg1 {-2.35/6 0.002s} 41. h4 {2.4s}
h5 {-2.18/6 0.003s} 42. Kf3 {4.4s} Kd7 {-1.68/6 0.001s} 43. Bg7 {8.9s}
Bc5 {-1.78/6 0.002s} 44. Ke2 {11s} Ke6 {-1.68/6 0.003s} 45. Bb2 {2.5s}
Kd6 {-2.23/6 0.004s} 46. Bc1 {1.7s} Kc6 {-2.23/6 0.001s} 47. Be3 {1.0s}
Bd6 {-2.28/6 0.001s} 48. Kd3 {2.5s} Be7 {-2.39/6 0.002s} 49. Bd4 {39s}
Kd6 {-2.35/6 0.002s} 50. c5+ {3.0s} Kd5 {-1.24/6 0.003s} 51. c6 {11s}
Kd6 {-1.44/6 0.001s} 52. Kc4 {1.5s} Kxc6 {-1.88/6 0.003s} 53. Be3 {12s}
Kd7 {-1.84/6 0.007s} 54. Bd2 {1.7s} Ke6 {-1.91/6 0.002s} 55. Bxb4 {1.5s}
Bxb4 {-2.18/6 0.002s} 56. Kxb4 {1.2s} Kd6 {-1.72/6 0.003s} 57. Kb5 {2.1s}
Kd5 {-2.03/6 0.003s} 58. Kb6 {5.9s} Ke4 {-1.01/6 0.002s} 59. b4 {1.6s}
Ke3 {-1.93/6 0.001s} 60. b5 {1.4s} Ke4 {-3.62/6 0.001s} 61. Kc7 {1.5s}
Ke3 {-8.29/6 0.001s} 62. b6 {1.1s} Kf2 {-8.65/6 0.001s} 63. b7 {1.1s}
Kxg3 {-9.06/6 0.002s} 64. b8=Q {1.6s} Kxf4 {-9.67/6 0.004s} 65. Kc6+ {3.2s}
Kg4 {-9.61/6 0.005s} 66. Qd8 {6.5s} Kf4 {-9.61/6 0.003s} 67. Qg5+ {1.8s}
Ke4 {-9.98/6 0.003s} 68. Kd6 {6.4s} f4 {-39.77/6 0.002s} 69. Qxh5 {2.4s}
f3 {-39.77/6 0.001s} 70. Qg4+ {1.6s} Ke3 {-39.77/6 0.001s} 71. Ke5 {11s}
Kf2 {-39.77/6 0.002s} 72. Kf4 1-0
PS: I missed so many better moves, and quicker ways to win. I know...
"Talk is cheap. Show me the code." -- Linus Torvalds.

Post Reply