This is the direction in which we are moving with SlugGo. We also
expect it to be difficult to integrate different approaches, but this
has always been our research direction: when there are multiple
codes which will each give an evaluation of a situation, how does
one design an arbitrator that makes the final decision?
While SlugGo started off with multiple instantiations of GNU Go,
it has always been our intent to add engines that evaluate, select,
and rank moves in completely different ways. It just made some
sense to start getting the cluster infrastructure worked out with
only one engine.
There are many details ... each engine has a different bias in its
selection, and also a different representation for the values of
those choices, so simple algebraic combinations of the values
does not in general work, even though it did when we were using
only GNU Go engines. But one thing is reasonable obvious: if a
bunch of completely different engines pick the same move, the
confidence that it is your best choice goes up.
Cheers,
David
On 1, Feb 2008, at 8:55 AM, Don Dailey wrote:
There is much to think about with Jason's and Michaels ideas. I
favor
a more integrated approach than Michael suggests because I think it
would be very difficult to essentially have 2 different programs
playing
the same game (ever play non-consultation doubles in chess or go?
It's
fun but the level of play stinks. You take turns making a move with
your partner and no consultation is allowed.)
I also favor focusing more on the tree portion, but no doubt the
play-out portion could improve. I say this because search is more
sensitive to early mistakes and work near the root is cheap
compared to
work nearer leaf nodes. But by all means we should do research on
what it takes in the play-outs.
- Don
Jason House wrote:
I wouldn't stop there. I'd like a static analyzer to add tactical
smarts to playouts. If there's a pre-existing nakade, seki, etc, the
playouts should get it right.
On Feb 1, 2008 10:34 AM, Michael Williams
<[EMAIL PROTECTED]
<mailto:[EMAIL PROTECTED]>> wrote:
I think we would all agree that UCT+MC is quite good for strategy
but not so good with tactics. I'd like to see this hybrid
engine:
One that starts with a
traditional full-board static analysis (with local tactical
searches), looking for urgent moves. If it finds an urgent move,
it makes it. If the position is
relatively quiet, it uses UCT+MC to find a good strategic play.
_______________________________________________
computer-go mailing list
computer-go@computer-go.org <mailto: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/
