Re: [gnugo-devel] Problem with final_status_list

[Gunnar Farneback]

Gilles wrote:
> I had some problem with the "final_status_list dead" command in gtp mode. 
> [...]
> The stone F6 is considered as prisoner what is incorrect as you may verify 
> displaying the position.

That's to some extent a question of philosophy. There really is a
problem in the position since black has too many cutting points and a
combination attack can be carried out. The best black can do if white
attacks is to sacrifice F6, so it's not completely unreasonable to say
that it's dead.

At level 10 GNU Go does find this combination attack and how to defend
against it but at level 5 it does not. Instead it tries to play white
E3 for other reasons, and there's no corresponding pattern for black
to defend against it.

In general final_status_list can give unexpected results when used in
unstable positions, since GNU Go first finishes the game by self-play.
That said, this particular problem can be solved by disabling pattern
ED22 when scoring a game.

/Gunnar

Index: patterns/patterns.db
===================================================================
RCS file: /cvsroot/gnugo/gnugo/patterns/patterns.db,v
retrieving revision 1.124
diff -u -r1.124 patterns.db
--- patterns/patterns.db        24 Jan 2004 04:04:57 -0000      1.124
+++ patterns/patterns.db        19 Feb 2004 00:02:56 -0000
@@ -9549,6 +9549,7 @@
 Pattern ED22
 # gf Added b classification. (3.1.18)
 # gf Split pattern and revised constraint. (3.3.3)
+# gf Not when doing scoring. (3.5.5)
 
 ??*X          sente endgame (maybe rather middle game) move
 ..XO
@@ -9562,7 +9563,7 @@
 ..A?
 ----
 
-;oplay_attack(*,A) && !oplay_attack_either(*,*,b)
+;!doing_scoring && oplay_attack(*,A) && !oplay_attack_either(*,*,b)
 
 
 Pattern ED22b