Here is an example of the rollout with only three trials, obviously this bugs goes beyond the std. errors (but perhaps only std errors are effected in a long trial). One more command, if I read the code right the cube rollout is set up through CommandRollout, while the moves rollouts have their own setup in the gtk code. Perhaps that is why the bug only effects the cube rollouts.
Christian. Cube analysis Rollout cubeless MWC 68.79% (Money: +1.000) Cubeful equities: 1. No double 103.59% 2. Double, pass 69.50% (-34.09%) 3. Double, take 69.12% (-34.47%) Proper cube action: Too good to double, take Rollout details: Centered 1-cube: 1.000 0.013 0.000 - 0.000 0.013 0.000 CL 68.79% CF 103.59% [0.398 0.009 0.000 - 0.398 0.000 0.000 CL 1.14% CF 34.99%] Player gnubg owns 2-cube: 0.770 0.005 0.000 - 0.230 0.007 0.000 CL 69.96% CF 69.12% [0.019 0.007 0.000 - 0.019 0.005 0.000 CL 0.39% CF 0.64%] Full cubeful rollout with var.redn. 3 games, Mersenne Twister dice gen. with seed 1835638889 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] _______________________________________________ Bug-gnubg mailing list [email protected] http://lists.gnu.org/mailman/listinfo/bug-gnubg
