On Sat, Jun 16, 2007 at 02:29:51PM -0400, [EMAIL PROTECTED] wrote: > My program tends to amplify noise fluctuations. If it's presented with > several equally good moves, it will pick winners and losers early on.
Actually, the more I think about this, the more sense it makes to me. If a program finds one good move, it should not spend too much time getting a precise estimate for another move that seems not to be much better. I still think it a bit strange that on an empty board, a program can prefer a 3-3 point in one corner, and in another corner find it quite unplayable. I guess my experiment would only make sense for pure MC evaluations, or other systems where each move is evaluated equally often (say, UCT after the first move). Probably more hazzle than it is worth. But I would still like to know how many MC evaluations it would take until all corners look at least somewhat similar... I bet that is many more than we see used currently - if it ever gets there. - Heikki -- Heikki Levanto "In Murphy We Turst" heikki (at) lsd (dot) dk _______________________________________________ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
