On Thu, 2007-07-26 at 05:21 -0700, steve uurtamo wrote: > > The way to think about a play-out policy is to ask, "how good would it > > be given an infinite number of simulations?" The answer for uniform > > random is, "not very." > > really?
Again it depends on your definition of "good." My main point is that after a few thousand simulations it doesn't improve very much and it never gets remotely close to perfect. However, that doesn't mean it's not "good", that depends on what your standard of reference is. I suspect an infinite number of simulations as an evaluation function is a pretty reasonable evaluation function. Not close to perfect by any means but certainly but what evaluation function is? If you could actually compute this quickly, it might be a very good practical evaluation function for a highly selective alpha beta searcher. Please note that you do not have to play an infinite number of play-outs to compute what the expected score of such an evaluation function should be. There is probably no fast way to compute it, but it could be calculated recursively by counting the number of legal moves at each level down to the end of the game and doing some simple math. Of course this is a hypothetical calculation since this would require more computing power than we can muster. - Don > s. > > > > > > > > ____________________________________________________________________________________ > Park yourself in front of a world of choices in alternative vehicles. Visit > the Yahoo! Auto Green Center. > http://autos.yahoo.com/green_center/ _______________________________________________ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/