>It is very interesting to me that you use the clump correction rule. I >could never get that to work in Fuego, either.
It is my impression (with no analysis whatsoever, so draw your own conclusions) that Fuego's use of an evaluation function helps it to overcome problems in the playouts. Pebbles has no evaluation function. This is a deliberate choice because I want to lose as many games as possible. Seriously. I learn more from losing. Regarding clumps: Pebbles will fail to create eyes quite often if there is no rule that moves clumpy stones to empty adjacent points in the playouts. Fuego will evaluate clumps as probably bad things within the UCT search, so it creates eyes despite the fact that the underlying playouts allow clumps to form. I am not certain of this diagnosis. If true, it would illustate how design objectives can be met in different ways, and why heuristics that succeed in one program can fail in another. _______________________________________________ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
