Re: [gnugo-devel] endgame tuning

[Arend Bayer]

Gunnar wrote:
> Arend wrote:
> > How do we learn that the block at F5 is trying to block influence from
> > F6? Either we add this information to the patterns in barriers.db.
> 
> My guess is that doing it through patterns is most feasible. Possibly
> special patterns for this purpose. In fact some intrusion patterns try
> to work around this problem already, e.g. Intrusion46, although rather
> inexactly.

Do you have a suggestion how such patterns would look like? I fear that
with the current framework of pattern, we would just need too many of
them to make this maintainable.

I see Nonterritory46 is an example of doing this by hand. But of course
there are many similar situations...

> > Or we check it _after_ spreading the white influence: We just check all
> > intersections with white influence, whether it has two neighbors with
> > blocks.
> 
> This sounds too general. If nothing else I fear it would require much
> more careful influence tuning to avoid accidentally getting into this
> situation.

You may be right, but I think I would still like to try it out. A bit
more careful would be to do this only for two blocks directly neighboring
an influence source. (I.e. pretty much a hardcoded pattern like

        ,!
        ?,

        a!
        ?b

        >remove_one_block(a,b)

where the "!" refers to an intrusion source added through another
pattern.

> > If yes, we remove one of them, and respread the influence from this
> > intersection.
> 
> This begs the question how we decide which of them to remove...

Yes. The best might be to just try out both of them and remove the one
where the territorial effect is smaller. This would mean running
value_territory() some hundreds of times, but I would hope this is not
a performance issue.


Arend