Calculated variance and proportioning by tree size are more or less similar with
many number of games. But using the standard deviation instead gave a better result (I think) atleast in estimating the mean..
Code: Select all
Mean 6.285437E+16
StdDev 5.212337E+12
Error 5.3E+11
with the actual mean is magnitudes smaller. It had very good intermediate results for both perft values but
got a little bit worse in the end.
Another variant I tried is to have the same standard deviation for all moves! (see the 1.55e12 stderror for all moves below). This weird looking constraint has revealed a possible problem.
It has done an irreparable damage to the first perft estimate which is used for move selection.
Even after 100 million iterations the first estimate is at 6.282e16. Intermediate estimates are much worse.
So I think not using the first estimate all in all even when having a split depth constraint is something to consider.
Code: Select all
Mean 6.283614E+16
StdDev 6.916737E+12
Error 1.87625E+13
Code: Select all
Equal std
0 h2h3 1.81E+15 +- 1.55E+12 1.82E+15 1425865 380 2.39E+24
0 h2h4 2.62E+15 +- 1.55E+12 2.62E+15 2884306 420 2.39E+24
0 g2g3 2.50E+15 +- 1.55E+12 2.49E+15 2502763 420 2.39E+24
0 g2g4 2.22E+15 +- 1.55E+12 2.22E+15 1705969 421 2.39E+24
0 f2f3 1.55E+15 +- 1.55E+12 1.55E+15 846748 380 2.39E+24
0 f2f4 2.05E+15 +- 1.55E+12 2.05E+15 1796446 401 2.39E+24
0 e2e3 7.16E+15 +- 1.55E+12 7.16E+15 20995846 599 2.39E+24
0 e2e4 7.26E+15 +- 1.55E+12 7.26E+15 20142018 600 2.39E+24
0 d2d3 4.59E+15 +- 1.55E+12 4.59E+15 6366435 539 2.39E+24
0 d2d4 6.33E+15 +- 1.55E+12 6.33E+15 15512703 560 2.39E+24
0 c2c3 2.75E+15 +- 1.55E+12 2.75E+15 3068007 420 2.39E+24
0 c2c4 3.12E+15 +- 1.55E+12 3.12E+15 3859430 441 2.39E+24
0 b2b3 2.40E+15 +- 1.55E+12 2.41E+15 2417406 420 2.39E+24
0 b2b4 2.41E+15 +- 1.55E+12 2.41E+15 2320316 421 2.39E+24
0 a2a3 1.82E+15 +- 1.55E+12 1.82E+15 1497705 380 2.39E+24
0 a2a4 2.57E+15 +- 1.55E+12 2.57E+15 2756916 420 2.39E+24
0 g1h3 2.13E+15 +- 1.55E+12 2.13E+15 1827200 400 2.39E+24
0 g1f3 2.71E+15 +- 1.55E+12 2.71E+15 2547848 440 2.39E+24
0 b1c3 2.73E+15 +- 1.55E+12 2.73E+15 3390280 440 2.39E+24
0 b1a3 2.10E+15 +- 1.55E+12 2.10E+15 1925793 400 2.39E+24
6.282292E+16 6.284936E+16 99790000 8902 4.784125E+25
Mean 6.283614E+16
StdDev 6.916737E+12
Error 1.87625E+13
Stddev
0 h2h3 1.81E+15 +- 8.03E+11 1.81E+15 3623153 380 6.44E+23
0 h2h4 2.62E+15 +- 1.02E+12 2.62E+15 4594931 420 1.04E+24
0 g2g3 2.50E+15 +- 9.76E+11 2.50E+15 4405621 420 9.52E+23
0 g2g4 2.22E+15 +- 8.73E+11 2.22E+15 3938972 421 7.61E+23
0 f2f3 1.55E+15 +- 6.94E+11 1.55E+15 3132723 380 4.82E+23
0 f2f4 2.05E+15 +- 8.73E+11 2.05E+15 3942728 401 7.63E+23
0 e2e3 7.16E+15 +- 1.98E+12 7.16E+15 8958211 599 3.94E+24
0 e2e4 7.27E+15 +- 1.98E+12 7.26E+15 8928458 600 3.91E+24
0 d2d3 4.59E+15 +- 1.37E+12 4.59E+15 6183808 539 1.88E+24
0 d2d4 6.33E+15 +- 1.80E+12 6.33E+15 8129509 560 3.24E+24
0 c2c3 2.75E+15 +- 1.05E+12 2.75E+15 4728075 420 1.10E+24
0 c2c4 3.12E+15 +- 1.13E+12 3.12E+15 5119144 441 1.29E+24
0 b2b3 2.41E+15 +- 9.57E+11 2.41E+15 4319738 420 9.16E+23
0 b2b4 2.41E+15 +- 9.50E+11 2.41E+15 4287109 421 9.02E+23
0 a2a3 1.83E+15 +- 8.11E+11 1.82E+15 3660697 380 6.57E+23
0 a2a4 2.57E+15 +- 1.00E+12 2.57E+15 4516322 420 1.00E+24
0 g1h3 2.13E+15 +- 8.78E+11 2.13E+15 3964579 400 7.71E+23
0 g1f3 2.71E+15 +- 9.98E+11 2.70E+15 4503755 440 9.95E+23
0 b1c3 2.73E+15 +- 1.07E+12 2.73E+15 4831800 440 1.15E+24
0 b1a3 2.10E+15 +- 8.91E+11 2.10E+15 4020667 400 7.93E+23
6.285870E+16 6.285004E+16 99790000 8902 2.716845E+25
Mean 6.285437E+16
StdDev 5.212337E+12
Error 5.3E+11