Re: computer-go: Re: null move

[Don Dailey <drd@xxxxxxxxxxxxxxxxxxxxx>]

It's all  in the wording!  The  classic null move algorithm (in chess)
is used to estimate a  lower bound on  the value likely to be returned
by  a given search.   If a given move  survives a beta cutoff, then it
must be searched full width.  It's a test to determine  if you HAVE to
go full width.

I say  it's all in  the wording, the description  below says  the same
thing at a higher and "fuzzier" level!

It's unclear (at least to me) whether  the classic null move algorithm
would make sense in  GO.  Perhaps some  modification of it would.  But
the right way  to think of this  kind of selectivity is  to substitute
"null move heuristic" for "bounds  estimate heuristic."  Think of null
move as just a (cheap) way to estimate a lower (or upper) bound on the
score, but not the only way.  I suspect in  go you might use something
much more  evaluation based (but I'm only   guessing.)  The quality of
your selectivity would  be  very much dependant  on  the cost  and the
quality of this estimator.

- Don

> >Re: the interesting subject of the null move heuristic, the reason it is
> >so effective is, I think, if you assume an extra free move, then your
> >plan had better work pretty obviously, or it won't work at all. So
> >people often formulate a "draft" plan of action using simple heuristics
> >and assuming an extra free (null) move, and if it works, try to refine
> >it to a plan which does not involve a free move.
> >
> >That's my 2 eurocents worth!
> >
> >Julian
>