We are thinking about it, but others have code that is far ahead
of whatever our first effort is likely to produce, so I favor a
collaboration.
We are still investigating behavior that we do not expect from
SlugGo, and we are working on a number of fronts to move
from where we have been (using multiple GNU Gos) towards
our original intent of several completely different Go engines.
Adding MC-UCT makes sense, but there are only so many
hours in a day, and only a few of us working, so MC-UCT
has not yet become our highest priority.
But we have not considered moving to pure MC-UCT.
Cheers,
David
On 25, Jan 2007, at 12:30 PM, Chris Fant wrote:
Have you considered refocusing towards MC Go?
On 1/25/07, David Doshay <[EMAIL PROTECTED]> wrote:
On 25, Jan 2007, at 10:14 AM, terry mcintyre wrote:
> So what would it take to get corporate sponsorship of the sort
which
> drove the chess computing field? Where is the Go equivalent of Deep
> Thought?
The Japanese Govt and industrial sponsorship of the Fifth Generation
Project, which did have playing Go on its list, collapsed very soon
after the Japanese real estate bubble burst, over 15 years ago.
> Near as I can tell, David Doshay's Sluggo is the only large-scale
> parallel effort. Mogo uses at most 4 CPUs. What might be
accomplished
> with one of the top500.org clusters of hundreds or thousands of
CPUs?
Corporate sponsorship would be required for access to a top500
cluster, but for a smaller scale attempt I have already offered the
72 CPUs in my largest cluster for a scaling study. Nobody took
me up on the offer, so I gave the time to a physics computation.
That computation should be done in less than a month, and I
will repeat the offer now. We can also put a few of my clusters
together and scale up a bit more. However, if nobody responds
in a short while, I will again turn the cluster back to physics.
If you have a MC-UCT code that can scale up to multiple CPUs
then contact me off the list. I am open to collaboration.
Repeating my experience to date: with the algorithms in use by
SlugGo, more that 20 or so processors and deeper lookahead
than just a few moves not only does not play better, but it plays
worse. MC-UCT clearly has better scaling behavior.
Cheers,
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/
