[computer-go] The dominance of search (Suzie v. GnuGo)

[Rémi Coulom]

dhillismail at netscape.net wrote:
>  
> >-----Original Message-----
> >From: Remi.Coulom at univ-lille3.fr
> >To: computer-go at computer-go.org
> >Sent: Mon, 16 Apr 2007 5:26 AM
> >Subject: Re: [computer-go] The dominance of search (Suzie v. GnuGo)
>
> >dhillismail at netscape.net 
> <javascript:parent.ComposeTo("dhillismail%40netscape.net", "");> wrote: 
> >> I also find this kind of information very interesting and useful. 
> Now > I have a better feel for what kind of scaling is realistic to 
> try >for > and how to measure it. 
> >> > Putting some recent data points together, it look like giving 
> Mogo 2 > orders of magnitude more computer power would result >in low 
> dan level > 19x19 play? Not the sort of thing one can pull out of a 
> back pocket, > but tantalizing. 
> >> > I would be very interested to see equivalent scaling numbers from 
> > CrazyStone, if Remi would be so kind. 
> >> > - Dave Hillis 
> >Hi, 
> > 
> >Here is some data, each result measured over about 200 games, on a 
> single CPU (AMD Opteron 2.2 GHz): 
> >9x9, 2 minutes per game, GNU Go level 10: 87.0% 
> >13x13, 16 minutes per game, GNU Go level 10: 72.4% 
> >19x19, 32 minutes per game, GNU Go level 0: 53.6% 
> >19x19, 32 minutes per game, GNU Go level 8: 28.1% 
> >19x19, 64 minutes per game, GNU Go level 8: 35.4% 
> > 
> >(GNU Go version 3.6) 
> > 
> >Rémi 
>  
>  
> Thanks! These are interesting. I assume that the number of playouts 
> per move was variable. If it was a fixed number, or easily 
> characterized, it would be a helpful statistic too.
>  
> - Dave Hillis
Yes, it was variable. On that machine, Crazy Stone does about 15,000 
playouts/second on 9x9 with UCT. BTW, CrazyStone-Fast on CGOS is 5k 
simulations, and CrazyStone-Slower is 20k.

Rémi