On 23, May 2009, at 4:03 AM, Gunnar Farnebäck wrote:
Joshua Shriver wrote:
> Perhaps I'm mistaken in my reading, but isn't Mogo a clusterized and
> highly tuned version of gnugo?
You are mistaken.
You may have mixed things up with SlugGo, which at least at some time
could be loosely described as a clusterized GNU Go, although I
don't believe highly tuned fits. I don't know what the current status
of SlugGo is.
/Gunnar
Gunnar is correct. SlugGo is not highly tuned, but rather has additional
heuristics for combining the multiple lines of play that are spread over
the cluster. The point of SlugGo is not to clusterize GNU Go, but to see
what can be done with multiple "brains" working on the same problem.
We just started with GNU Go because it is open source.
SlugGo moves in fits and starts ... Grad students come and go, and I
have
been dealing with heart problems lately, so progress has been slow, when
there has been progress at all.
Because my background is physics I have been bothered by MC methods.
My thesis was on MC methods to investigate phase transitions, so I am
fine with MC methods, but in physics we have theory that gives us the
correct probability distributions. I am impressed with the MCTS methods
that do so well without a prior distribution, but we spent the last 2
years
trying different methods of extracting distributions from Go games. We
had no success. We are trying to figure out an appropriate way to
publish
our negative results.
On 23, May 2009, at 8:15 AM, Ian Osgood wrote:
You are thinking of the cluster research program SlugGo. That
developer and the GNU Go team have the friendly agreement not to
both compete in the same tournament at the same time. GNU Go only
participated in the 2008 US computer Go championship when SlugGo
could not get its new cluster working in time to participate.
Yes, I ask the GNU Go folks if they will compete, and if they do, SlugGo
does not. I entered and operated GNU Go in Portland for the reason
stated by Ian.
Cheers,
David
_______________________________________________
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/
