|
BGonline.org Forums
Stick vs Merriman: XG++ Data Makes No Sense
Posted By: Daniel Murphy In Response To: Stick vs Merriman: XG++ Data Makes No Sense (Paul Weaver)
Date: Monday, 30 July 2012, at 6:27 p.m.
I don't what exactly the problem is, but isn't it likely to have something to do with bad evaluations (that is, estimates of "cubeless" games results) at 7-ply truncation? I agree with Stick's comment that the cube is very unlikely to be used, so evaluations at point of double/pass should not be the problem.
XG Roller++ is: 360 Games using VR, truncated after 7 moves using 3-ply, stop when reaching 0.010 Confidence (minimum 180 games), first 2 decisions in 3-ply then 2-ply checker and 3-ply cube.
Neil's 4-ply/4-ply rollout and Stick's 3-ply/XG Roller rollout correctly rank 10/5*/2 first. So will, I believe, a Gnubg 2-ply/2-ply rollout.
But if I duplicate the XG Roller++ settings in Gnubg, the result is similarly screwy, in that 10/5*/2 wins more "cubeless" games and gammons but 15/10 6/3 is ranked first:
White 0, Blue 8, match to 11 points
White 174
Blue 72 Position ID: zB5wQHDbzUMQAA Match ID: UYluAQAAQAAE Alert: doubtful move ( -0.012)
# Ply Move Equity 1 R 15/10 6/3 +1.289163
0.946143 0.885774 0.041681 - 0.053857 0.000716 0.000003 +1.362201 +1.289163 0.001056 0.001679 0.001585 - 0.001056 0.000099 0.000002 0.002733 0.003994 Truncated cubeful rollout (depth 7) with var.redn. 360 games, Mersenne Twister dice gen. with seed 691526212 and quasi-random dice Play: world class 2-ply cubeful prune [world class] keep the first 0 0-ply moves and up to 8 more moves within equity 0.16 Skip pruning for 1-ply moves. Cube: 2-ply cubeful prune [world class] Different evaluations after 2 plies: Player 0: Play: 1-ply cubeful, noise 0.015 (d) keep the first 5995833 0-ply moves Cube: 2-ply cubeful, noise 0.015 (d) Player 1: Play: 1-ply cubeful prune keep the first 0 0-ply moves and up to 16 more moves within equity 0.32 Cube: 2-ply cubeful prune [world class] 2 R 10/5*/2 +1.283965 ( -0.005198)
0.949127 0.894044 0.069789 - 0.050873 0.000637 0.000003 +1.372664 +1.283965 0.001064 0.001848 0.001795 - 0.001064 0.000108 0.000005 0.002838 0.004001 Truncated cubeful rollout (depth 7) with var.redn. 360 games, Mersenne Twister dice gen. with seed 691526212 and quasi-random dice Play: world class 2-ply cubeful prune [world class] keep the first 0 0-ply moves and up to 8 more moves within equity 0.16 Skip pruning for 1-ply moves. Cube: 2-ply cubeful prune [world class] Different evaluations after 2 plies: Player 0: Play: 1-ply cubeful, noise 0.015 (d) keep the first 5995833 0-ply moves Cube: 2-ply cubeful, noise 0.015 (d) Player 1: Play: 1-ply cubeful prune keep the first 0 0-ply moves and up to 16 more moves within equity 0.32 Cube: 2-ply cubeful prune [world class] • 3 R 15/7 +1.277112 ( -0.012051)
0.942208 0.888034 0.043620 - 0.057792 0.002039 0.000025 +1.353855 +1.277112 0.001149 0.001636 0.001803 - 0.001149 0.000378 0.000007 0.002976 0.004339 Truncated cubeful rollout (depth 7) with var.redn. 360 games, Mersenne Twister dice gen. with seed 691526212 and quasi-random dice Play: world class 2-ply cubeful prune [world class] keep the first 0 0-ply moves and up to 8 more moves within equity 0.16 Skip pruning for 1-ply moves. Cube: 2-ply cubeful prune [world class] Different evaluations after 2 plies: Player 0: Play: 1-ply cubeful, noise 0.015 (d) keep the first 5995833 0-ply moves Cube: 2-ply cubeful, noise 0.015 (d) Player 1: Play: 1-ply cubeful prune keep the first 0 0-ply moves and up to 16 more moves within equity 0.32 Cube: 2-ply cubeful prune [world class]
|
BGonline.org Forums is maintained by Stick with WebBBS 5.12.