I'd really like to see how this does against traditional programs. If Don sets up a 19x19 server I'll put Many Faces of Go on it.
David > -----Original Message----- > From: [EMAIL PROTECTED] > [mailto:[EMAIL PROTECTED] On Behalf Of > Chaslot G (MICC) > Sent: Tuesday, December 12, 2006 1:17 AM > To: computer-go > Subject: [computer-go] Scaling to 19x19 > > > Hi David, > > I have scaled my Monte-Carlo program (Mango) to 19x19. > My idea is to bias the UCT distribution including big > patterns. This seems to work quite well, and to lead to a > much more human-like kind of play then pure Monte Carlo. The > patterns were learnt automatically from professional games as > described in this paper: > http://www.math-info.univ-paris5.fr/~bouzy/publications/bouzy- > chaslot-ci > g05.pdf > > I am currently writing a paper describing the techniques I > used to scale on 19x19. You can see Mango's results on KGS > (it is 11k at the moment). It is a preliminary version, still > slow and with stupid bugs (no shisho management, which leads > to funny games sometimes :)). > > Cheers, > > Guillaume > > -----Original Message----- > From: [EMAIL PROTECTED] > [mailto:[EMAIL PROTECTED] On Behalf Of > David Fotland > Sent: Tuesday, December 12, 2006 6:10 AM > To: [EMAIL PROTECTED]; 'computer-go' > Subject: RE: [computer-go] 19x19 CGOS? > > I'd like to see 19x19. No one plays the game on any other > board size than 19x19, so the other sizes are not very > interesting. The current strong programs are all tuned only > for 19x19, and the patterns and strategy are quite different > at other board sizes. > > Of course you should keep the 9x9 server running as well, > since it's probably easier to tune algorithms on the smaller > board size. You probably want longer games at 19x19. I'd > like to see 30 minutes per side, since that's pretty typical > for human games. > > -David > > > -----Original Message----- > > From: [EMAIL PROTECTED] > > [mailto:[EMAIL PROTECTED] On Behalf Of Don Dailey > > Sent: Monday, December 11, 2006 8:49 PM > > To: computer-go > > Subject: Re: [computer-go] 19x19 CGOS? > > > > > > Did you have a specific size in mind? It seems like a > 19x19 server > > would be the natural thing. > > > > I could run the old server until I get the new one finished. > > > > - Don > > > > > > On Mon, 2006-12-11 at 20:29 -0800, David Doshay wrote: > > > A few months ago I suggested a number of stepwise increases > > in board > > > size to see how the algorithms scaled. It seems to me > > having just 2 > > > data points does not say enough about how the MC (or any other) > > > algorithm scales, so I wanted to be able to graph some measure of > > > strength against increasing board size. > > > > > > The responses showed some interest in the normal board > > sizes, but not > > > enough for me to think it was worth it to host the servers. > > Several of > > > the program authors indicated that they could only run one > > size at a > > > time, and I assumed that this meant that having multiple > > CGOS servers > > > would only detract from the usefulness of the 9x9, so I did > > not start > > > any other size of server. > > > > > > If things have changed, I have the resources to be a CGOS-N host. > > > > > > Cheers, > > > David > > > > > > > > > > > > On 11, Dec 2006, at 7:53 PM, David Fotland wrote: > > > > > > > > > > > Hi Don, > > > > > > > > Clearly UCT and monte carlo is very well suited to 9x9 go. It > > > > works much better than the traditional computer go > algorithms, and > > it is much, > > > > much > > > > simpler. > > > > > > > > Do you have any plans to set up a CGOS server for 19x19 > go? I'd > > > > like to see how well UCT/MC scales to 19x19 go. I > don't think it > > > > will work > > > > well at > > > > 19x19, but I'd like to see some experiments to see if I'm wrong. > > > > > > > > David > > > > > > > > > > > > _______________________________________________ > > > > 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/ > > > > _______________________________________________ > > 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/ > _______________________________________________ > 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/
