[gnugo-devel] hane tuning

[Arend Bayer]

- pattern tuning

Degrading t patterns to 16 helped in a few regression tests, and this
confirms with my general impression that this was too high.

Other than that, I mainly tried to make gnugo unlearn premature unsupported
first line hanes as typically in this
+.....
+O..X.
+..OX.
+...*..
+-----
position. As I reluctantly used a replace-action for this, I tried to
relief my bad conscience by removing another replace pattern :-) (after
checking with kombilo that it seems unnecessary now).

Arend



Index: engine/shapes.c
===================================================================
RCS file: /cvsroot/gnugo/gnugo/engine/shapes.c,v
retrieving revision 1.45
diff -u -p -r1.45 shapes.c
--- engine/shapes.c     28 Apr 2003 17:45:04 -0000      1.45
+++ engine/shapes.c     20 May 2003 12:01:41 -0000
@@ -36,7 +36,7 @@
 #define U_VALUE 40.0
 #define J_VALUE 35.0
 #define j_VALUE 24.0
-#define t_VALUE 17.0
+#define t_VALUE 16.0


 /*
Index: patterns/barriers.db
===================================================================
RCS file: /cvsroot/gnugo/gnugo/patterns/barriers.db,v
retrieving revision 1.52
diff -u -p -r1.52 barriers.db
--- patterns/barriers.db        16 Apr 2003 22:03:43 -0000      1.52
+++ patterns/barriers.db        20 May 2003 12:01:42 -0000
@@ -999,14 +999,14 @@ Pattern Barrier66

 Pattern Intrusion1

-xxxx
+???x
 ...X
 !!.Q
 ----

 :8,B,value(30)

-xxxx
+???x
 ..cX
 !!ba
 ----
@@ -1249,21 +1249,22 @@ ba


 Pattern Intrusion10
+# ab Modified action constraint. (3.3.20)

-xxx?
-..XO
-!!.Q
-----
+xxx??
+..XO?
+!!.Q?
+-----

 :8,B,value(30)

-xxx?
-..CO
-!bad
-----
+xxxef
+..COg
+!badh
+-----

 ;oplay_attack_either(a,b,b,C)
->return (!xplay_attack(a,d));
+>return (!xplay_attack(a,d) && o_somewhere(g,e,f,h));


 Pattern Intrusion11
@@ -2129,6 +2130,27 @@ bXfo
 ?.de

 ;!oplay_disconnect(f,g) && !oplay_disconnect(a,b,c,d,e,c,f)
+
+
+Pattern Intrusion59
+# New pattern (3.3.20)
+# It is a little drastic, but see arend2:120 for where it is necessary.
+
+|x!xx
+|.!.x
+|.X.X
+|XQO?
+|.oo?
+
+:8,B,value(30)
+
+|x!xx
+|.!.x
+|.B.C
+|AQO?
+|.oo?
+
+;!attack(A) && oplay_disconnect(A,C) && !attack(C)


 Pattern Nonterritory1
Index: patterns/endgame.db
===================================================================
RCS file: /cvsroot/gnugo/gnugo/patterns/endgame.db,v
retrieving revision 1.49
diff -u -p -r1.49 endgame.db
--- patterns/endgame.db 21 Apr 2003 21:40:36 -0000      1.49
+++ patterns/endgame.db 20 May 2003 12:01:43 -0000
@@ -210,12 +210,19 @@ XOO.

 Pattern EE5

+???
 XO?       descend to edge
 .*?       may quite possibly be 0 points gote
 ---

 :8,OXe,terri(0.1)

+?ab
+XOc       descend to edge
+.*d       may quite possibly be 0 points gote
+---
+
+;o_somewhere(a,b,c,d)

 Pattern EE6

@@ -292,7 +299,7 @@ Pattern EE103
 ?.*.
 ----

-:8,OXe,terri(1)
+:8,OXe

 ?OX?
 ?.*a
@@ -339,7 +346,7 @@ Pattern EE104
 ..*.
 ----

-:8,OXe,terri(2)
+:8,OXe

 ?OX?
 b.*a
@@ -359,7 +366,7 @@ Pattern EE105
 ..*..
 -----

-:8,OXe,terri(2)
+:8,OXe

 ?OE..
 bc*a.
@@ -374,7 +381,7 @@ Pattern EE106
 ...*.
 -----

-:8,OXe,terri(2),reverse_followup(6)
+:8,OXe,reverse_followup(6)

 ..dX?
 .bc*a
@@ -819,7 +826,7 @@ Pattern EE504
 *.O
 ---

-:8,Oe,terri(2)
+:8,Oe

 ...
 *.O
Index: patterns/owl_attackpats.db
===================================================================
RCS file: /cvsroot/gnugo/gnugo/patterns/owl_attackpats.db,v
retrieving revision 1.83
diff -u -p -r1.83 owl_attackpats.db
--- patterns/owl_attackpats.db  12 May 2003 10:06:51 -0000      1.83
+++ patterns/owl_attackpats.db  20 May 2003 12:01:44 -0000
@@ -1953,6 +1953,7 @@ Pattern A513
 #   This may be just too generic, though often useful.
 # tm Modified (3.1.16)
 #   Added ??? rows, make sure not at upper edge.
+# ab This is indeed too generic.

 ???|
 ???|
Index: patterns/owl_vital_apats.db
===================================================================
RCS file: /cvsroot/gnugo/gnugo/patterns/owl_vital_apats.db,v
retrieving revision 1.37
diff -u -p -r1.37 owl_vital_apats.db
--- patterns/owl_vital_apats.db 12 May 2003 08:50:42 -0000      1.37
+++ patterns/owl_vital_apats.db 20 May 2003 12:01:45 -0000
@@ -831,6 +831,8 @@ Pattern VA48

 Pattern VA49
 # tm New Pattern (3.1.18) (see trevor:430)
+# FIXME: This is non-sense if the ataried stone has neighbours
+# in atari (see trevorb:360)

 X*    atari outside eyespace
 oX
Index: patterns/patterns.db
===================================================================
RCS file: /cvsroot/gnugo/gnugo/patterns/patterns.db,v
retrieving revision 1.108
diff -u -p -r1.108 patterns.db
--- patterns/patterns.db        11 May 2003 12:59:07 -0000      1.108
+++ patterns/patterns.db        20 May 2003 12:01:49 -0000
@@ -6107,24 +6107,21 @@ x.*.o
 Pattern EB1403
 # gf Revised pattern. (3.1.3)
 # gf Revised constraint. (3.3.13)
+# ab Removed followup value (handled by influence code). (3.3.20)

 ?oX??          sente hane
 ..XO?
 ?.*.o
 -----

-:8,Xe,followup_value(8)
+:8,Xe

 ?oD??
-ecDf?
-?a*bo
+b.Da?
+?.*..
 -----

-# The oplay_defend test is a workaround for oplay_defend_both not
-# correctly handling ko results.
-;alive(f) && xmoyo(e)
-;&& (!oplay_defend_both(*,a,b,?,c,a,D) || oplay_defend(*,a,b,?,c,a) != WIN)
-;&& !oplay_attack(*,a,b,c,*)
+;alive(a) && xmoyo(b)


 Pattern EB1404a
@@ -6138,12 +6135,10 @@ Pattern EB1404a
 # In practice this pattern doesn't always kick in at a high enough
 # value to guarantee that EB1403 will be followed up.
 #
-# This has the same followup value as EB1403 has. This might make
-#    EB1404b superfluous.
-#
 # gf Revised pattern. (3.1.3)
 # eb revised constraint (3.3.13)
 # see nngs3:1110
+# ab FIXME: Is this still necessary?

 ?oX??          Connect after hane.
 ..XO?
@@ -6161,11 +6156,13 @@ Pattern EB1404a


 Pattern EB1404b
-# Second followup to EB1403. This confirms that EB1403 was matched
+# Second followup to EB1403.
 # and guarantees that the move is followed up.
 # gf Revised pattern. (3.1.3)
 # eb revised constraint (3.3.13)
 # see nngs3:1110
+# ab FIMXE: Still needed? Also maybe reuse code from connect_and_cut_helper
+# here?

 ?oX??          Connect after hane.
 ..XO?
@@ -7179,17 +7176,19 @@ O.....?   extend
 :8,OE,shape(-1),jump_out_far_helper


-Pattern CB102
-
-??O??     extend
-?...?
-?...?
-?...?
-?.*.?
-?...?
-?????
-
-:|,OE,shape(-1),jump_out_far_helper
+#Pattern CB102
+# ab Removed.
+# (No need for negative shape on 3rd line; CB 104 sufficient.)
+#
+#??O??    extend
+#?...?
+#?...?
+#?...?
+#?.*.?
+#?...?
+#?????
+#
+#:|,OE,shape(-1),jump_out_far_helper


 Pattern CB103
@@ -10713,7 +10712,8 @@ oOXx??

 Pattern ED93
 # db new pattern (3.3.6)
-# see century-2002:120
+# see century2002:120
+# ab Added e classification (3.3.20)

 .....X?        don't make the wrong atari
 ..X...?
@@ -10722,7 +10722,7 @@ Pattern ED93
 .......
 -------

-:8,dO
+:8,edO

 .....X?
 ..X.a.?
@@ -10775,6 +10775,28 @@ Pattern ED95
 ;!weak(a)


+Pattern ED96
+# ab New pattern (3.3.20)
+# See arend2:120
+
+?....
+O*.X.
+..OX.
+.....
+-----
+
+:8,OXe
+
+?....
+f*.X.
+hgeXd
+..cab
+-----
+
+;!oplay_connect(a,b,c,d,?,g,?,h,e,f)
+> replace(a,*)
+
+
 #####################################################################
 #
 # Center defend/attack patterns
@@ -12279,29 +12301,30 @@ X...*....cd
 # :8,Ed,shape(5)


-Pattern EJ49
-# gf Revised. (3.1.17)
-
-.....         cut
-O*X.o
-.XO.o
-.....
-.....
------
-
-:8,sEj
-
-.....
-O*X.c
-.XObd
-..a..
-.....
------
-
-;o_alive_somewhere(c,d)
-
->replace(a,*);
->replace(b,*)
+# Pattern EJ49
+# ab Removed (3.3.20)
+# # gf Revised. (3.1.17)
+#
+# .....         cut
+# O*X.o
+# .XO.o
+# .....
+# .....
+# -----
+#
+# :8,sEj
+#
+# .....
+# O*X.c
+# .XObd
+# ..a..
+# .....
+# -----
+#
+# ;o_alive_somewhere(c,d)
+#
+# >replace(a,*);
+# >replace(b,*)


 Pattern EJ49b
Index: patterns/patterns2.db
===================================================================
RCS file: /cvsroot/gnugo/gnugo/patterns/patterns2.db,v
retrieving revision 1.59
diff -u -p -r1.59 patterns2.db
--- patterns/patterns2.db       12 May 2003 08:50:42 -0000      1.59
+++ patterns/patterns2.db       20 May 2003 12:01:51 -0000
@@ -2810,7 +2810,7 @@ oOA.
 .*OB
 ----

-;oplay_attack_either(*,A,B)
+;oplay_attack_either(*,A,B) && !attack(B) && !attack(A)


 Pattern Sente5
Index: regression/arend.tst
===================================================================
RCS file: /cvsroot/gnugo/gnugo/regression/arend.tst,v
retrieving revision 1.25
diff -u -p -r1.25 arend.tst
--- regression/arend.tst        25 Feb 2003 20:37:17 -0000      1.25
+++ regression/arend.tst        20 May 2003 12:01:51 -0000
@@ -129,9 +129,10 @@ loadsgf games/arend/gnugo-gnugo5.sgf 105
 #? [!H10]

 # Low Shimari clearly better here
+# See also arend2:130
 loadsgf games/arend/gnugo-gnugo6.sgf 20
-24 gg_genmove white 4
-#? [!P16]*
+24 restricted_genmove white P16 P17
+#? [P17]*

 #CATEGORY=ENDGAME_TUNING
 # C19 is either double sente, or worth at least 5 pts in reverse sente,
Index: regression/arend2.tst
===================================================================
RCS file: /cvsroot/gnugo/gnugo/regression/arend2.tst,v
retrieving revision 1.6
diff -u -p -r1.6 arend2.tst
--- regression/arend2.tst       16 Apr 2003 22:03:43 -0000      1.6
+++ regression/arend2.tst       20 May 2003 12:01:51 -0000
@@ -44,3 +44,16 @@ loadsgf games/paul.sgf 91
 100 restricted_genmove white O16 O15
 #? [O16]

+loadsgf games/nngs/gnugo-3.3.18-agh-200304291749.sgf 64
+110 restricted_genmove white T7 T8 R8
+#? [R8]*
+
+loadsgf games/nngs/gnugo-3.3.18-joshj-200304242104.sgf 58
+120 restricted_genmove white G19 H17 J17
+#? [H17|J17]*
+
+# See also arend:24
+loadsgf games/arend/gnugo-gnugo6.sgf 20
+130 restricted_genmove white P17 G16 F16 E15
+#? [P17]
+
Index: regression/strategy2.tst
===================================================================
RCS file: /cvsroot/gnugo/gnugo/regression/strategy2.tst,v
retrieving revision 1.52
diff -u -p -r1.52 strategy2.tst
--- regression/strategy2.tst    7 May 2003 15:04:26 -0000       1.52
+++ regression/strategy2.tst    20 May 2003 12:01:53 -0000
@@ -166,7 +166,7 @@ loadsgf games/strategy14a.sgf
 # This is essentially a semeai problem
 loadsgf games/strategy22.sgf
 80 gg_genmove white
-#? [P4|Q4|Q3]*
+#? [P4|Q4|Q3|S8]*

 # Looks to me like B12 is also effective, although much more
 # complicated. /gf