Re: [gnugo-devel] Connection point not seen (D1320b)

[Arend Bayer]

On Thu, 5 Jun 2003, SP LEE wrote:

> > Attached is a game I just played with the latest GNUGO (from the cvs). There
> > was an obvious blunder on move 158, where GNUGO thought its group L9 is
> > dead, while this group can be simply connected with outside on O14. In GTP
> > mode, owl_attack L9 showed that O14 is an attack point. However, it didn't
> > even consider the move O14 in "owl_defend L9" and thought the group L9 is
> > dead without any defense. O14 is generated by the move generator, but almost
> > without any value.
>
> I found that GNUGO did find D1320b for O14. However this move was discarded 
> because the neighboring dragon (O15) is not strong enough (only one eye was 
> seen ;) ). As I had suspected, it seems that the pattern needs to be revised 
> instead of adding more patterns.

> I think 'strong dragon' may not have clearly two eyes, especially it's hard 
> for GNUGO to find where these two eyes are. Even if it has only one eye, and 
> the dragon under attack has one eye, it's crucial to connect two dragons 
> together. I would suggest to read the liberties as a measurement for 
> 'strong'. Maybe liberties against number of stones.
>
> Please comment.

One way to solve this is to modify owl_strong_dragon, in such a way
that P14 in this position gets considered strong. You could use the
dragon.safety or the dragon.weakness field to get a better criterion
than the 2 eyes-check.

Arend