I think the issue with MC or deterministic is following. With MC your value of the node can get more accurate with larger number of playouts. With a deterministic you don'thave such a mechanism here (there are other?mechanisms in UCT).
DL -----Original Message----- From: Don Dailey <[EMAIL PROTECTED]> To: computer-go <computer-go@computer-go.org> Sent: Wed, 9 Jan 2008 3:46 pm Subject: Re: [computer-go] How to design the stronger playout policy? Mark Boon wrote: > > On 8-jan-08, at 17:04, Don Dailey wrote: > >> And yes, it slows down the play-outs. Still, the play-outs seem to >> >> require a good bit of randomness - certainly they cannot be >> >> deterministic and it seems difficult to find the general principles that >> >> are important to the play-out policy. >> > > I was thinking about this and wanted to make sure I understand what > you mean by 'cannot be deterministic'. You could build a playing strategy that would always play the same exact game from a particular position and this would be deterministic but then the "Monte Carlo" component is completely missing from the program. You would still get some chaos from the tree itself since you always expand a node once visited. I don't think this is fully understood so I could be wrong in my statement - but it does seem likely that a certain amount of randomness will improve the results. Attempts to make the play-outs stronger and stronger have failed, perhaps due to the fact that this makes them less and less random, or perhaps it's due to some other phenomenon. - Don > For example a stone gets put into atari and it finds an escaping move. > With plain playouts the stone gets captured anyway roughly 50% of the > time. With heavy playouts it gives a weighting so that the chance of > escaping becomes larger? Or does it always play the escaping move and > just the rest of the playout is random? > > Mark > > ------------------------------------------------------------------------ > > _______________________________________________ > computer-go mailing list > computer-go@computer-go.org > http://www.computer-go.org/mailman/listinfo/computer-go/ _______________________________________________ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ ________________________________________________________________________ More new features than ever. Check out the new AOL Mail ! - http://webmail.aol.com
_______________________________________________ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
