Re: [gnugo-devel] Inessentiality again

[Daniel Bump]

The idea of classifying some worms as inessential is
that they are immediately recognized as living inside
the opponent's eyespace.

Basically what we consider inessential has an element of
arbitrariness to it but anything that isn't inessential
has to be treated elsewhere. Unless the ambient dragon
has two eyes somewhere else this may mean that changing
the classification from inessential to something else
means the semeai code will be called. This should be
less of a problem if the semeai code is reliable.

The last example you gave shows that if the worm inside
can get one eye it should not be classified as inessential.
This is the reason for the edge criterion. If the worm 
has a lot of edge liberties:

OOOOOOOOO
O.XXXXX.O
.........
---------

then it can get an eye. Situations where a worm has
genus 0 but can get an eye are most common right along
the edge.

The aim of the criterion is to be conservative, that is,
seldom is a worm classified as inessential that really
isn't.

I think that revising the definition of inessential is
something that should be done but is not of high priority
compared with some of the other issues we are facing
right now. Our main concern at this moment is that this
definition should not prevent us from recognizing some
semeai as was the example (until I added the cutstone
criterion) for the top position in semeai6.sgf.

I think the right time to revise this definition is
after owl_analyze_semeai is reliable enough to be called
from semeai(), or earlier if we encounter more positions
where a semeai is not recognized because something is
missclassified as inessential.

Dan