Re: [computer-go] Solved boards

[Erik van der Werf <E.vanderWerf@xxxxxxxxxxxxx>]

Alejandro Lujan wrote:

> > > In particular, Im interested in knowing if GnuGo uses this information on a
> > >      
> > 5x5
> >    
> > > board (what would make it impossible to beat it playing white, right?)
> > >      
> > Not AFAIK. Although it could be done, at least for very slow games, I 
> > don't think the Gnugo developers are interested in such a thing.
> >
> > BTW, it is not necessary to solve the game to play perfect, it is only 
> > needed if you want to be completely sure. I would not be surprised to 
> > find that (maybe after some additional tuning) Gnugo's heuristic play is 
> > good enough to be unbeatable after starting with black in the center on 5x5.
> >
> > Erik
> >
> >    
> OK, you are giving me a clue here.  My interest comes from a work Im doing with
> evolving neural networks in wich I have to measure my program's level against
> gnugo.  To compare to past works I was using 5x5 boards, but found that my
> program always (99%) loses when playing white, and has a good average (75%
> after a few generations) when playing black.  I assume this has a strong
> relation to that particular board size being a solved problem.  Right now Im
> starting to test on 9x9 boards but the evolution is sloooooooooow.
>
>  
Do you randomize the opening moves, or at least use a large number of 
fixed opening positions?
If so, 99% seems strange...

Erik


_______________________________________________
computer-go mailing list
computer-go@xxxxxxxxxxxxxxxxx
http://www.computer-go.org/mailman/listinfo/computer-go/