Re: [gnugo-devel] A fuseki test case

[Evan Berggren Daniel]

On Sun, 24 Aug 2003, SP LEE wrote:

> The following test case is from a recent game on nngs.
>
>    A B C D E F G H J K L M N O P Q R S T
> 19 . . . . . . . . . . . . . . . . . . . 19
> 18 . . . . . . . . . . . . . . . . . . . 18
> 17 . . . O . . . . . O . . . O X . . . . 17
> 16 . . . + . . . . . + . . . O . X . . . 16
> 15 . . O . . . . . . . . . . . . . . . . 15
> 14 . . . . . . . . . . . . . . . . X . . 14
> 13 . . X . . . . . . . . . . . . . . . . 13
> 12 . . . . . . . . . . . . . . . . . . . 12
> 11 . . . . . . . . . . . . . . . . . . . 11
> 10 . . . X . . . . . + . . . . . + . . . 10
> 9 . . . . . . . . . . . . . . . . . . . 9
> 8 . . . . . . . . . . . . . . . . . . . 8
> 7 . . . . . . . . . . . . . . O . . . . 7
> 6 . . . . . . . . . . . X X X X O O . . 6
> 5 . . X . . . . . . . . . O O O X O . . 5
> 4 . . O + . O . . . + . . . . . X X . . 4
> 3 . . . O . . . X . . . . X . . X . . . 3
> 2 . . . . . . . . . . . . . . . . . . . 2
> 1 . . . . . . . . . . . . . . . . . . . 1
>    A B C D E F G H J K L M N O P Q R S T
>
> Gnugo played at D5, which let black answer at C6, not only live but also
> have chance to build territory on left side. To generalize the problem, it's
> important not to play near a light opponent tone and pincer is usually a
> good choice. In the above situation, the black stone at C5 was classified as
> alive. After adding a white stone at C7, C5 was seen dead by gnugo. Of
> course, it might live but white is in a easy position to chase it. It seems
> to me that it's hard to add a pattern to solve the problem. We might add
> another dragon status, light, to use special patterns to treat them
> separately.

While not ideal, the following patch improves things in this case.  There
are several things that go wrong with the move valuation here.  First,
F704 matches when it shouldn't.  This results in the move being considered
an invasion, which is clearly wrong, and also gives it the Ft
classification, which mandates a value of 15.07.  So the first thing to do
is fix F704.

Secondly, we need the move to be suggested by some other mechanism.  The
patch adds two new patterns, F604a and F604b, which suggest a 1- or
2-space pincer, respectively.  F604b does not match here because of the
stone at D10.

The end result of the patch is that the move at C7 is suggested with Fae
classification, which is probably very reasonable here.  Gnugo then values
C7 at 23.15, which is much more reasonable.  The remaining problem is that
D5 is still valued at 26.88, which is obviously too high.

I think the F704 portion of this patch is worth considering for cvs, but
the F604 patterns probably require further checking that they don't
produce overconcentrated moves.

I think you are correct that an algorithmic solution is needed to improve
gnugo's understanding of the situation in the general case.

Is there a good way to generate a negative strategic effect on the C5
stone, since the move at D5 forces black to defend it, and the defense is
highly effective?

Evan Daniel

Index: patterns/fuseki.db
===================================================================
RCS file: /cvsroot/gnugo/gnugo/patterns/fuseki.db,v
retrieving revision 1.32
diff -u -d -r1.32 fuseki.db
--- patterns/fuseki.db  11 Aug 2003 13:29:02 -0000      1.32
+++ patterns/fuseki.db  25 Aug 2003 04:58:08 -0000
@@ -1420,6 +1420,56 @@
 -----

 ;x_alive_somewhere(a,b,c)
+
+
+Pattern F604a
+#evand new pattern. (3.5.1)
+#see nngs4:750
+
+.....o     pincer a stone against a wall
+.....o
+.....o
+..*.XO
+.....o
+.....o
+------
+
+:8,Fae
+
+.....a
+.....b
+f....c
+e.*.XG
+.....o
+.....o
+------
+
+;o_alive_somewhere(a,b,c) && !omoyo(e) && !omoyo(f) && !weak(G)
+
+
+Pattern F604b
+#evand new pattern. (3.5.1)
+#see nngs4:750
+
+.......o           pincer a stone against a wall
+.......o
+.......o
+...*..XO
+.......o
+.......o
+--------
+
+:8,Fae
+
+.......a
+.......b
+f......c
+e..*..XG
+.......o
+.......o
+--------
+
+;o_alive_somewhere(a,b,c) && !omoyo(e) && !omoyo(f) && !weak(G)


 #####################################
@@ -1478,24 +1528,26 @@
 Pattern F704
 # gf Added constraint. (3.1.22)
 # See nngs:1030.
+# evand revised. (3.5.1)
+# See nngs4:750

-???????
-?.....?       invade 4 space extension
-.X.....
-....*.X
-.......
-.......
--------
+????????
+?.....??       invade 4 space extension
+.X......
+....*.X.
+........
+........
+--------

 :8,sFIt,shape(-1)

-???c?d?
-?.....?
-.A.....
-....*.B
-.......
-.......
--------
+???c?d??
+?.....??
+.A......
+....*.B.
+........
+........
+--------

 ;weak(A) || weak(B) || !xmoyo(c) || !xmoyo(d)