Re: [gnugo-devel] DFA should be "best" NFA

[Tanguy URVOY]

Hi,

> > Please don't rely on stones almost only being added to the board.
> > Obviously captured stones will be removed but more importantly the
> > matcher is used repeatedly in search trees
>
>
> Good point. I augment my observation that the most common changes are adding
> stones or undoing recently added stones. Captures are still rare(r).

I am not so sure that captures will be so rare.
A most reasonable approach is to compute the
invariants before reading.


for example :

If we intend to read only on a small
region of the board to solve a semeai between two groups.
we know that no new stone will be put out of
this region. We also know that the capture of one of these groups
is the end of the race.


1) we can consider any intersection out of this region as invariant
2) we can consider any worm which has at last one liberty out of this 
region as invariant.
3) we can consider the two involved groups as invariants.

We can filter the pattern database and --->

1) remove any pattern incompatible with the hypotheses 1,2,3
2) precompute any test based on 1,2,3.


Tanguy

--
---------------------------------------
Tanguy Urvoy http://www.irisa.fr/galion
---------------------------------------