[Computer-go] UCT parameters and application to other games

Daniel Shawul dshawul at gmail.com
Tue Apr 5 22:55:11 PDT 2011


Yes we agree. I tried to point that out by  'nps wise' (nodes per second).
I guess in Go it is playouts per second. The real scaling is of course as
you described.

On Tue, Apr 5, 2011 at 7:50 PM, Michael Williams <
michaelwilliams75 at gmail.com> wrote:

> On Tue, Apr 5, 2011 at 4:50 AM, Daniel Shawul <dshawul at gmail.com> wrote:
>
>> Oh no I actually agree with you. But I share concerns of Joona about the
>> kind of algorithm to be used.
>> He probably used root parallelization scheme like I just did. It scales
>> almost perfectly on a cluster of 16 cpus nps wise.
>>  But then I realized this might not be as good as it seems because the
>> different processors might not produce
>> significantly different trees..
>>
>
> Obviously you mean the number of simulations scales almost perfectly.  More
> important is how well the strength scales in your parallel implementation vs
> how well it scales in your sequential implementation.
>
> But, of course, that takes much longer to determine.
>
>
> _______________________________________________
> Computer-go mailing list
> Computer-go at dvandva.org
> http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://computer-go.org/pipermail/computer-go/attachments/20110406/521fe2ba/attachment.html>


More information about the Computer-go mailing list