On Wed, Jun 29, 2011 at 10:26 PM, Peter Drake <[email protected]> wrote: > On Jun 29, 2011, at 12:14 PM, Erik van der Werf wrote: > >> I hope you are aware that some strong MCTS programs use (at least) a >> factor hundred less playouts to break even with gnugo. In fact, to get >> to 50% they don't even need a tree at all... (so UCT is perhaps not >> really that relevant at these levels) > > Yes -- as stated, this new method is not (yet?) competitive with > cutting-edge MCTS (e.g., RAVE and fancy domain-specific playouts). Our claim > is merely that it beats vanilla UCT.
That claim can't be true in general. Sure you can beat a weak implementation at some fixed playout level but in general (e.g., as the number of simulations grows much larger) its simply not possible because the linear classifier alone doesn't have the capacity to represent a big tree (which eventually it must be able to do to infer deep tactics). Erik _______________________________________________ Computer-go mailing list [email protected] http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
