Re: [gnugo-devel] False eye territory

[Gunnar Farneback]

Arend wrote:
> > However, this is NOT a complete solution to the problem. It does,
> > e.g., not suffice to understand that in
> >
> > |.OOOOO
> > |.O.*XO
> > |OOX.XO
> > |OOXXXO
> > |OX.XOO
> > |X.XOO.
> > |.XXO..
> > +------
> >
> > the move at * is worth two points of territory.
> 
> Couldn't this be solved by our usual paradigm "assume the other player
> moves first in all local positions"? I.e., upgrade half eyes to full
> eyes for the color that is assumed to move next?

It wouldn't help here. The major limitation of my solution is that it
only involves a static analysis of what topologically false (or half)
eyes have territorial value. The point of the move at * is that it
changes the conditions for that analysis.

If the analysis was redone for each considered move the situation
would be different. I chose not to try that approach because it would
require substantially more complex code and might potentially be too
slow.

/Gunnar