Code: Select all
56053 game(s) loaded, 0 game(s) with unknown result ignored.
Rank Name Elo + - games score oppo. draws
1 Glaurung 2.1 180 6 6 11212 78% -40 17%
2 Fruit 2.1 64 5 6 11203 64% -40 23%
3 opponent-21.7 23 5 6 11210 59% -40 34%
4 Glaurung 1.1 SMP 8 6 6 11207 56% -40 20%
5 Crafty-22.2 -35 5 5 17144 43% 15 22%
6 Crafty-22.2R1 -42 4 4 38909 42% 15 23%
7 Arasan 10.0 -197 6 6 11221 29% -40 19%
About 1/2 way through, there is a 7 Elo difference, but when you compare the ranges with the error bar (-40 to -30) and (-47 to -37) there is some overlap. The remaining 20K games will reduce the error bars for both to +/-4 roughly, I will report the results when they finish probably tomorrow morning (still waiting on A/C repair so we are at 1/2 cluster powered down). I also have one more run with plain null-move R=3, since the adaptive null move (R=2~3) tapered R off near the leaves due to the danger of overlooking a serious threat, and the q-search checks are partially intended to help with that.
So far, +500 lines of code, minimal improvement if it holds up.
More as it becomes available. This version is a pretty clean implementation. that is efficient in terms of speed, and pretty cute in the code that generates direct/discovered checks without any loops.