>There are other areas where it is much much worse. For example >evaluating cube situations in containment games, I have positions >(several), where the 2-ply might say No Double/take, and a rollout >will say Too Good to Double/Pass.
That's well known, for some positions classes the neural net is not as good as we'd like it to be. But the positions reported by Mochy were wrong due to a recently introduced (and more recently fixed) bug in the code, not in the nets. Fixing this take snothing (actually it's already done in the very latest source code), while solving the issue of your positions is a much tougher task (not even started at the time). MaX. _______________________________________________ Bug-gnubg mailing list [email protected] http://lists.gnu.org/mailman/listinfo/bug-gnubg
