Re: [gnugo-devel] Eye pattern 304

[Paul Pogonyshev]

Gunnar wrote:
> Paul wrote:
> > - find_half_and_false_eyes() bugfixed
> > - improve_lunch_attack() revised
> >
> > smaller changes include:
> > - new type of marginal vertices INESSENTIAL_MARGINAL which don't
> > (marginals that don't participate in graph matching)
> > - D219 pattern retired
>
> To help understanding the patch, could you explain the characteristics
> of an inessential margin and how they help to solve the problem in the
> original message of this thread?

sorry, i was going to comment the changes once i became sure it worked,
but apparently forgot to do that. i'll resubmit the patch with appropriate
comments in 3-4 hours (don't have cvs access right now).

i came up with inessential marginals to avoid revising pattern databases.
e.g. in the SP Lee's position we had the following:

  ...O.
  XXXOX
  X....

which is considered as

  ..!

by optics code after domain building. when i made the topological analyser
understand that the neighbor vertex is marginal too (as a false eye) i got
this eye:

  .!!

since we don't allow such eyespaces it didn't match. i decided the easiest
(and maybe the most logical) way would be to just ignore the right marginal,
matching

  .!

only. this uncovered a weakness in lunch attacking code and brought up
many fails, so i had to revise improve_lunch_attack().

more comments will be in the patch.

Paul