[computer-go] New version of Crazy Stone

[Łukasz Lew]

Congratularions Rémi :)

And thank You for sharing this knowledge with us.

Łukasz

On 6/2/06, Rémi Coulom <Remi.Coulom at univ-lille3.fr> wrote:
> Hi,
>
> I have just come back from the Turin Computer Game Olympiad. I enjoyed
> this event very much. It was great to meet all the programmers. The
> version of Crazy Stone that participated in the tournament, and the
> games that it played, are now available for download:
> http://remi.coulom.free.fr/CrazyStone/
> The site of the Olympiad is there:
> http://www.cs.unimaas.nl/olympiad2006/index.htm
>
> Main changes in Crazy Stone are multiprocessor support, and expected
> territory replaced by probability of winning. Using the probability of
> winning increased the strength of Crazy Stone tremendously. By changing
> one line of code, Crazy Stone improved from scoring 35% against GNU Go
> to more than 60%. After more fine-tuning of the search parameters, it is
> now even stronger.
>
> Up to 4 processors, the speed seems to scale rather well. In case any of
> you has a machine with more than 4 CPUs, I'd be very curious to know the
>    speed you get. In order to measure it, run the ::MCSearch::Benchmark
> Analyze command (in gogui) with different values of the "Processors"
> parameter in ::MCSearch::Parameters. I found that it is possible to get
> a small speed increase on Intel processors with hyperthreading by
> setting the number of processors to 2 (I get about 30% speedup on my P4
> 3.4GHz).
>
> I will let Crazy Stone play on CGOS with the 4-CPU machine for some time.
>
> Greetings,
>
> Rémi
> _______________________________________________
> computer-go mailing list
> computer-go at computer-go.org
> http://www.computer-go.org/mailman/listinfo/computer-go/
>