Re: [computer-go] question about Gnugo

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

Myriam Abramson wrote:

> Is Gnugo a "worthy opponent" as white on a 5x5 for machine learning
> program? 
>  
IMO 5x5 is only interesting for a machine learning approach if you have 
a sufficiently large number of balanced opening positions. A simple way 
to do this is to create a set of fixed opening positions by randomly 
selecting the first n moves (and playing each for both sides). However, 
it might be that preventing only the first move in the center is already 
good enough...


Gunnar Farnebäck wrote:

> If I remember correctly, by the way, the most interesting way to play
> on 5x5 is to force black to start on 2-2.
>  
Yes, (2,2) is the most balanced opening move. (3,2) is also interesting. 
The (2,2) and (3,2) points are the only opening moves on the empty board 
where both sides can make a living group (under perfect play).

Erik

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