On Jul 28, 2008, at 5:04 AM, Darren Cook <[EMAIL PROTECTED]> wrote:
Personally, I think the next big strength jump would come from
combining
localized searches/sequences with the global search's MC playouts.
Curiously, my guess is the opposite: using UCT as the node
evaluation in
a more traditional alpha-beta searcher. (It's been mentioned a few
times
here but I don't think anyone has given it a serious try yet?)
I have an alpha beta searcher that uses MC node evaluations. The last
time I played with it was before I got my core 10x faster.
I don't expect grand things from it yet. For example I don't have
CrazyStone's ELO move ratings would be great for-based move ordering.
Also, I'm unsure if MC noise would dominate the alpha-beta search.
(BTW, David, the new Many Faces combines traditional algorithms and
UCT;
how are they working together?)
I always recommend to new developers that they join forces with other
developers to reduce the total work to get a strong bot. I think the
more people we have starting from a solid bot implementation, the
faster
we'll discover the next great strength breakthrough.
There are lots of competing projects, some open source, some in
universities, some commercial. The thinking behind my question is
perhaps I can help them all by working on a really good opening
library
(or connection patterns, or optimized UCT implementation, or
whatever is
needed most).
Darren
--
Darren Cook, Software Researcher/Developer
http://dcook.org/mlsn/ (English-Japanese-German-Chinese-Arabic
open source dictionary/semantic network)
http://dcook.org/work/ (About me and my work)
http://darrendev.blogspot.com/ (blog on php, flash, i18n, linux, ...)
_______________________________________________
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/
