>This explanation looks inexact. >The move matching will always be better than 1/branching factor.
I think you missed the point of the explanation. If you ponder taking the opponent's point of view, and you have a match on his move, then you will have a tree for *your* move that is 1/BF of the size of a full tree. So your gain from pondering is a factor of 1 + 1/BF. But that only applies when you have a match, which happens 50% of the time, so the net gain is 1 + 1/2*BF. (Note: BF here is the effective branching factor == growth rate of the search process, not the nominal branching factor, which equals the growth rate of the search space. So in chess, BF ~2.5, not 36.) If you speculatively ponder, and you have a match, then you gain a factor of 2. This is scaled back to a factor of 1.5 because you match about 50% of the time. It is clear that speculative pondering is much better. One of the key points is that the move matching rate does not depend much on the search depth. This is true enough in chess. In Go, I suspect that longer searches have a higher probability of matching, but I haven't verified that. Brian _______________________________________________ Computer-go mailing list [email protected] http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
