[gnugo-devel] t patterns audit

[Arend Bayer]

In the game against Go++, GNU Go fell behind early on by playing a lot
of slow moves, and catched up later on by consistently making fewer
mistakes than Go++. (In the end, Go++ won by 10 pts.)

So the game was lost in the beginning. Looking at the slow moves, they
were all generated by old t patterns, that are simply overconcentrated.
I first tried throwing out all t patterns, but that is probably not a
good idea.

So instead I audited all "t" patterns in fuseki.db and removed the "t"
classification whereever I felt that more context is needed to determine
the value (and I am confident that GNU Go will usually get this right).

The only regression fail is safety:5. H16 is now valued 26 pts instead
of 17, which seems a lot closer to the truth.
Also, all the slow moves at the bottom in the Go++ game disappear.
E.g. instead of H5, the top moves are now C12, R8, F9 -- pretty much
the moves I'd consider (although R8 should get the top place).

Arend

- remove t classification from a couple of patterns in fuseki.db

Index: patterns/fuseki.db
===================================================================
RCS file: /cvsroot/gnugo/gnugo/patterns/fuseki.db,v
retrieving revision 1.31
diff -u -p -U9 -r1.31 fuseki.db
--- patterns/fuseki.db  28 Jul 2003 14:38:47 -0000      1.31
+++ patterns/fuseki.db  9 Aug 2003 11:23:00 -0000
@@ -763,19 +763,19 @@ Pattern F207
 Pattern F208

 .O.......         3 space extension from 1 space jump
 .....*...
 .O......O
 .........
 .........
 ---------

-:8,FeEt,shape(-2)
+:8,eE,shape(-2)


 Pattern F209

 |...................|      San ren sei!
 |...................|
 |...O.....*.....O...|
 |...................|
 |...................|
@@ -821,33 +821,33 @@ Pattern F302

 .......           jump up to claim space from 4 point extension
 ?....*.
 .......
 O....O.
 .......
 .......
 -------

-:8,FEt
+:8,E


 Pattern F303

 ......      seal the largest area
 ..*...
 O....O
 ......
 O....O
 ......
 ......
 ------

-:8,FEt,shape(-2)
+:8,E,shape(-2)


 Pattern F304

 ......       light reducing move
 ......
 .....X
 .*....
 X....X
@@ -874,19 +874,19 @@ Pattern F306

 oo......       jump to expand
 oo...*..
 oo.....x
 .....O.x
 ........
 ........
 --------

-:8,Ft
+:8,E

 oo......
 oo...*..
 ab.....x
 .....O.x
 ........
 ........
 --------

@@ -940,19 +940,19 @@ x.*..O.cd

 Pattern F402

 ......x?    make 2 space extension on third row
 .O..*.x?
 ......x?
 ........
 --------

-:8,edt,jump_out_helper
+:8,ed


 Pattern F403
 # This should only be j or J if the O stones do not already
 #      have a base, t otherwise. But they usually do not
 #      have a base as this pattern most naturally comes up.

 ......o?  make 3 space extension from two stones
 ......O?
@@ -981,19 +981,19 @@ Pattern F405
 #   Probably best to remove this pattern.

 ?.....o?
 ...*..O?  make 3 space extension from two stones
 O.....O?
 ........
 ........
 --------

-:8,Fedt
+:8,e

 ?.....o?
 ...*..O?
 O..a..O?
 ........
 ........
 --------

 ;!oterri(*) && !oterri(a)
@@ -1019,32 +1019,32 @@ Pattern F407
 #     (see nngs:1150, nngs:1030, nngs:1230)

 .....???
 ......??   make 2 space extension on third row
 .O..*...
 ......??
 ......??
 --------

-:8,edt
+:8,ed


 Pattern F408
 # gf Added constraint to avoid overconcentration. See 13x13:12,
 #    13x13b:42. (3.3.17)

 .......        extend
 oO..*..
 .......
 .......
 -------

-:8,edt
+:8,ed

 .....a.
 oO..*..
 .......
 .......
 -------

 ;!omoyo(a)

@@ -1052,31 +1052,31 @@ oO..*..
 Pattern F409

 ?......??     3 space high extension
 .....*...
 .O.......
 .........
 .........
 ---------

-:8,FEt,shape(1)
+:8,Ed,shape(1)


 Pattern F410

 ???......?            extend
 ???.O....?
 ???....*.?
 ???.......
 ???.......
 ----------

-:8,FEd
+:8,Ed


 Pattern F410a

 ?xx......?            extend
 ?xx.O....?
 ?xx....*.?
 ?xx.......
 ?xx.......
@@ -1352,19 +1352,19 @@ bd.*....O

 Pattern F505

 .......        extend
 .O.....
 ....*.X
 .......
 -------

-:8,Feadt,shape(1)
+:8,ead,shape(1)


 Pattern F506

 ......     Extend along edge
 O..*.X
 o....x
 ......
 ------