Hideki Kato wrote:
> What is "correct" move? It has sense only for some artificial
> problems or very limited positions, and so, it cannot evaluate total
> performance of a program.
There are many positions where winning the game requires one or a small set of
moves.
Suppose the board is equally divided, and all groups are settled but one. The
last group
will live or die, depending on whether the program makes the single correct
move or not.
Many life-and-death program can be converted to this form.
An alternative approach would be to take a traditional life-and-death problem,
and assign
a komi such that the problem must be solved to win the game.
A third approach might be to devise a UCT framework which can be given specific
goals,
such as "kill this group" or "live", or "select a move from this subset", which
otherwise uses
the same algorithms for whole-board play.
Given a large set of life-and-death problems which have been converted into
whole-board problems,
I believe that any UCT program which handles the life-and-death problems
correctly would probably
do quite well in whole-board play also. The knowledge should generalize well.
____________________________________________________________________________________
Never miss a thing. Make Yahoo your home page.
http://www.yahoo.com/r/hs_______________________________________________
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/
