RE: [computer-go] Learning : was Chess programs versus go programs

[Arend Bayer <arend.bayer@xxxxxx>]

Hi David and Anders,

On Sun, 12 Dec 2004, Anders Kierulf wrote:

> > Mark, [...] I think you are correct that using 4 liberties for
> > tactical stability is better than using 5 liberties once you have
> > good life and death evaluation.
> > The next version of Many Faces will have this change.
> > 
> > David
> 
> I tried the same change in SmartGo, and so far it looks positive. The gain
> in global lookahead due to increased speed seems to outweigh the loss of
> accuracy in the capture information.

Do you use a depth cutoff beyond which strings with 4 liberties are
considered tactically stable? GNU Go currently uses 7 at the default level 10.

This depth cutoff is relative to the full-board position we are
analyzing, i.e. if we are at depth 4 in life-and-death reading and call
out to the tactical reading, it will only read 3 plys with 5 liberties
at the safety cutoff. I didn't get around to making the same experiment
with GNU Go but that's why I don't expect it to be a drastic speed
improvement.

Arend

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