Re: [gnugo-devel] attack5

[Arend Bayer]

Dan wrote:

> > This probably needs a lot more analysis before putting it in CVS.  I'm
> > unsure of whether it's a good idea to put it in without defend5.  Clearly,
> > defend5 is needed before more sophisticated attacks can be tried in
> > attack5.
>
> We probably don't want to do this. The tactical reading assumes that
> a string with 5 liberties is alive, and if it has more than 5 liberties
> attacking it is an issue for the owl code.

I agree that in the current form, the patch is probably too expensive.

However, there are numerous cases where the tactical reading won't
get corrected by the owl code (e.g. reading constraints).
I've encountered a couple of times situations where tactically capturing
a 5-liberty string would have helped. Typically it was a string
extending on the 2nd line, or a string already enclosed from all sides.

One could test for this by counting 2nd order liberties. It seems a lot
easier to capture a 5-liberty string with three 2nd order liberties,
than capture a 4-liberty string with eight 2nd order liberties.

So maybe one could refine the "give up"-criteria by using 2nd order
liberties. I would be quite confident that one could both save reading
nodes _and_ get better results by tuning them.

Btw, I don't think defend5 would be necessary. Having 5 liberties _and_
being in turn to move is a lot better than having 5 liberties at the
opponent's turn. (Corollary: attack5 does nothing else but playing
liberties.)

Arend