Re: [gnugo-devel] corner problem

[Gunnar Farneback]

Max wrote:
> move 40 :All GNUgos play badly here .

Owl misread. It gets to this position,

   A B C D E F G H J
 9 . . . . X O O O . 9
 8 . O . . X O X X X 8
 7 . . X X . O X O O 7
 6 . O . O X X O . . 6
 5 O X X . + X O . . 5
 4 . O X X . X X O . 4
 3 O . O X . . + O . 3
 2 O O O O X X O . . 2
 1 . X X X . . . . . 1
   A B C D E F G H J

where it runs into "getting deep, looks lively".

> move 42 : gnugo 3.2 does not consider this strange move

Yes, it does consider it. In fact it's valued 6.18 points and is the
second choice.

The problem here is that the pattern DD4 comes up with a bogus move
reason. The appended patch solves the problem by adding connection
tests to the constraint. The performance effects are extremely small.
For the full regressions it adds 2 connection nodes and 6 reading
nodes (somewhere in trevor.tst) with no breakage.

> BTW , as i have already reported here , GG wins by far against Winhonte
> program on 19*19 ...

That's good to know. Thanks for doing these tests.

> But it seems to have huge problems on 9*9 board.

There's a lot to wish for when it comes to 9x9 tuning.

/Gunnar

Index: patterns/patterns2.db
===================================================================
RCS file: /cvsroot/gnugo/gnugo/patterns/patterns2.db,v
retrieving revision 1.64
diff -u -r1.64 patterns2.db
--- patterns/patterns2.db       23 Aug 2003 14:04:25 -0000      1.64
+++ patterns/patterns2.db       22 Oct 2003 21:37:05 -0000
@@ -3243,6 +3243,7 @@
 
 
 Pattern DD4
+# gf Revised constraint. (3.5.2)
 
 XOO
 O..
@@ -3256,6 +3257,7 @@
 
 ;xplay_attack(a,d) && xplay_attack_either(a,*,d,e)
 ;&& !xplay_defend_both(a,b,c,d,e) && !xplay_defend_both(a,c,b,d,e)
+;&& !xplay_connect(a,d,e) && oplay_connect(*,a,d,e)
 
 >add_defend_both_move(d,e);