[gnugo-devel] arend_3_2.2: Tests from handtalk games.

[Arend Bayer]

On Mon, 29 Apr 2002 Daniel Bump wrote:

>
> Jessie wrote:
>
> > One dreadful bad move by Gnugo was move 24 in game 21.
>
> It gets valued 17.27 points. The high rating is a function
> of the influence code, which was changed after 3.0 to
> Arend Bayer's scheme which makes GNU Go much more aggressive
> about making and reducing moyos. It probably is really worth
> 17 points but other moves are under valued. If W wants to
> play in this area capping at K15 would be better.

And probably the influence function would favor K15 if we had a pattern
suggesting it. Nevertheless I need to do s.th. about the remaining "cosmic"
but a little aimless moves like this one or 5th line keimas.

This move is included in this set of new tests, which are from the handtalk
games 12, 13, 21, 24 (can you please add them to CVS?).

Arend

 - Tests from handtalk games added in blunder, owl, connection, handtalk.tst.


(I hope creating the new file handtalk.tst works automatically by applying
the patch. I didn't insert it in the Makefiles yet as this will clash
with arend_3_2.1.)

New file: regression/handtalk.tst
===================================================================
--- regression/handtalk.tst        Tue Apr 30 17:37:07 2002
+++ regression/handtalk.tst        Tue Apr 30 15:41:52 2002
@@ -0,0 +1,102 @@
+
+reset_owl_node_counter
+reset_reading_node_counter
+reset_trymove_counter
+
+# No the right moment to start cosmic go at G7 as in the game.
+loadsgf games/handtalk/handtalk12.sgf 17
+1 gg_genmove black
+#? [R12|R13|R14|R8|R7|R6]
+
+# Correct owl defense at P5 gets in 3.1.31 due to negative shape value
+# -20 (!) by Pattern Shape13.
+loadsgf games/handtalk/handtalk12.sgf 37
+2 gg_genmove black
+#? [P5|Q2|R2]
+
+# The sagari at F18 (very much liked by 3.1.26-32 influence code) is bad
+# endgame here (as well as in many other situations). Also, defending N7
+# seems more urgent.
+loadsgf games/handtalk/handtalk12.sgf 55
+3 gg_genmove black
+#? [!F18]
+
+# Have to try this ko!
+loadsgf games/handtalk/handtalk12.sgf 91
+4 gg_genmove black
+#? [Q2]
+
+loadsgf games/handtalk/handtalk13.sgf 79
+5 gg_genmove black
+#? [!P15]
+
+loadsgf games/handtalk/handtalk13.sgf 161
+6 gg_genmove black
+#? [!F16]
+
+loadsgf games/handtalk/handtalk21.sgf 14
+7 gg_genmove white
+#? [R4]
+
+loadsgf games/handtalk/handtalk21.sgf 24
+8 gg_genmove white
+#? [!G14]
+
+loadsgf games/handtalk/handtalk21.sgf 46
+9 gg_genmove white
+#? [J17]
+
+# Other moves possible, but GNU Go should attack the D8 stone
+loadsgf games/handtalk/handtalk23.sgf 26
+10 gg_genmove white
+#? [E9|F8|D8]
+
+# B9 is quite bad here.
+loadsgf games/handtalk/handtalk23.sgf 30
+11 gg_genmove white
+#? [C11|D10]
+
+# The balance of power between J11 and H8 is the issue at the moment.
+# Maybe other moves possible.
+loadsgf games/handtalk/handtalk23.sgf 38
+12 gg_genmove white
+#? [G9|K8|J9|K9]
+
+loadsgf games/handtalk/handtalk23.sgf 48
+13 gg_genmove white
+#? [!P14]
+
+loadsgf games/handtalk/handtalk23.sgf 80
+14 gg_genmove white
+#? [K4]
+
+loadsgf games/handtalk/handtalk23.sgf 82
+15 gg_genmove white
+#? [J3]
+
+#CATEGORY=CONNECTION
+loadsgf games/handtalk/handtalk23.sgf 96
+16 gg_genmove white
+#? [J18]
+
+loadsgf games/handtalk/handtalk23.sgf 162
+17 gg_genmove white
+#? [B4|B6]
+
+############ End of Tests #################
+
+# Report number of nodes visited by the tactical reading
+10000 get_reading_node_counter
+#? [0]&
+
+# Report number of nodes visited by the owl code
+10001 get_owl_node_counter
+#? [0]&
+
+# Report number of nodes visited by the life code
+10002 get_life_node_counter
+#? [0]&
+
+# Report number of trymoves/trykos visited by the test
+10003 get_trymove_counter
+#? [0]&
Index: regression/blunder.tst
===================================================================
RCS file: /cvsroot/gnugo/gnugo/regression/blunder.tst,v
retrieving revision 1.15
diff -u -r1.15 blunder.tst
--- regression/blunder.tst      3 Feb 2002 17:54:22 -0000       1.15
+++ regression/blunder.tst      30 Apr 2002 14:53:31 -0000
@@ -86,3 +86,13 @@
 loadsgf games/blunder14.sgf
 19 gg_genmove white
 #? [P18|P19|R18|R19]
+
+# P11 got incorrectly rejected as a blunder.
+loadsgf games/handtalk/handtalk13.sgf 123
+20 gg_genmove black
+#? [P11]
+
+# L9 got incorrectly rejected as a blunder.
+loadsgf games/handtalk/handtalk13.sgf 131
+21 gg_genmove black
+#? [L9]
Index: regression/connection.tst
===================================================================
RCS file: /cvsroot/gnugo/gnugo/regression/connection.tst,v
retrieving revision 1.24
diff -u -r1.24 connection.tst
--- regression/connection.tst   26 Apr 2002 18:23:40 -0000      1.24
+++ regression/connection.tst   30 Apr 2002 14:53:32 -0000
@@ -251,6 +251,14 @@
 80 connect R2 R7
 #? [1 S6]*

+loadsgf games/handtalk/handtalk13.sgf 106
+81 connect L5 R5
+#? [1 O5]
+
+loadsgf games/handtalk/handtalk13.sgf 116
+82 same_dragon M7 M10
+#? [0]
+
 # Report number of nodes visited by the tactical reading
 10000 get_reading_node_counter
 #? [0]&
Index: regression/owl.tst
===================================================================
RCS file: /cvsroot/gnugo/gnugo/regression/owl.tst,v
retrieving revision 1.48
diff -u -r1.48 owl.tst
--- regression/owl.tst  24 Apr 2002 11:54:36 -0000      1.48
+++ regression/owl.tst  30 Apr 2002 14:53:34 -0000
@@ -932,6 +932,20 @@
 258 owl_defend O2
 #? [0]

+loadsgf games/handtalk/handtalk12.sgf 29
+259 owl_does_attack L2 M3
+#? [0]
+
+# Far too many escape ways for F10 as it could be killed.
+loadsgf games/handtalk/handtalk12.sgf 41
+260 owl_does_attack G9 F10
+#? [0]
+
+# A good example where connection handling is important in owl.
+loadsgf games/handtalk/handtalk23.sgf 186
+261 owl_attack E5
+#? [1 (O6|Q7|T4|T5|R4)]
+
 ########### end of tests #####################

 # Report number of nodes visited by the tactical reading