Hi, Eduardo Sabbatella wrote: > I'm not sure at all. But, in your example, you are > assuming that "XX move worths 10 points". > > Thats impossible to said without error %. > That error could lead the alfa-beta algorithm to prune > an important move on a 'bad' board configuration for > our evaluation function. This statement is totally correct.
However, as Arthur said, it is not a problem of the alpha-beta algorithm. Rather, it is fundamental problem with minimax search with limit in search depth and imperfect evaluation function. The only thing that could be said is that, as long as we use the same evaluation function, if we are going to search for the same deapth, the alpha-beta algorithm will: give exactly the same best move and, search at most the same number of nodes compared to a full width minimax search. Note that, if the move orderings are in the worst case, the alpha-beta algorithm will search exactly the same number of nodes compared to a full width minimax search. -- Shunsuke SOEDA _______________________________________________ computer-go mailing list [email protected] http://www.computer-go.org/mailman/listinfo/computer-go/
