Re[5]: [gnugo-devel] life & death (3.3.5)

[Paul Pogonyshev]

> I think a good way to start would be: try to systematically look for
> cases where the eye space analysis goes wrong for big eyes with
> hostile stones inside.

here is another example of gnu go fail. consider the position below.
white group at the top is critical: if black is to move, he can kill
it, if white is to move, he can defend.

       A B C D E F G H J K L M N
      ---------------------------
  13 | . . . O X . X . . X . . . | 13
  12 | . . X O . O . O O O X . . | 12
  11 | . . X . O X O O X O X . . | 11
  10 | . . . X X X X X X X X . . | 10
   9 | . . . . . . . . . . . . . |  9
   8 | . . X . . . . . . . X . . |  8
   7 | . . . . . . . X . . . . . |  7
   6 | . . X . . . . . . . X . . |  6
   5 | . . . . . X . . . . . . . |  5
   4 | . . . . . . . . . . . . . |  4
   3 | X X X X X X X X X X X X X |  3
   2 | O O O O O O O O . . . . . |  2
   1 | . . . . . . . . . . . . . |  1
      ---------------------------
       A B C D E F G H J K L M N

gnu go (3.3.5) plays as follows (first move):
  white G12 (incorrect, should be F13),
  black G12 (correct, might be F13).

i still don't know how this can be detected, but to connect stone at
F12 to the rest of the dragon, white has to play E12 and G12 sooner or
later. following the concept i proposed, we have to analyze the
following eye shape (i'm not sure, is it correct?):

       X.X.!

this matches pattern 525 from eyes.db and gives a critical point:

       x*X.!

which is exactly at F13!

since G12 is a correct attack point in this position, playing at
connection points should be considered at least when attacking (i
wrote that "on other hand, it's probably no use for black to play B5"
- this time B5 becomes E12 and G12). however, owl code seems to be
able to find such attacks.

another point where this concept might be implemented is seki reading
(which doesn't seem good yet). consider this:

       O O O O
       O . . O
       O X X O O
       O X X X O
      -----------

in this situation, groups are alive in seki (in case white groups
doesn't have eyes elsewhere).

however, if change the situation a bit:

       O O O X X
       O . a O X
       O X X O O
       O X X X O
      -----------

the white dragon becomes dead. it has to connect at point a, so its
eye shape looks like

       .
       XX
       XXX

and yields only one eye.

> You have hit an important and interesting problem here, I think.

so, the idea is probably worth implementing. the hardest thing seems
to be detecting of "necessary" connections.

Paul Pogonyshev