[gnugo-devel] (no subject)

[Arend Bayer]

Here is my list of tests in strategy?.tst that I disagree with (I looked
through all those where GnuGo fails). (Some overlap with Trevor's
recent patch in strategy4.tst.)

Further down are my suggested patches for the .tst-file (pls tell me if
I should use other format for your patch; I used diff -Naur and all patches
are relative to 3.1.13).

-Arend


Strategy.tst:

15: I agree with the previous comments. J12 looks just about perfect to
me. Since this is such an unclear problem, I suggest to drop it.

18: Black F3 seems fine to me. Best way to protect both cuts at F3 and K3.

40: M11 (apparently a suggestion by an earlier GnuGo-Version) seems
excellent -- make territory while attacking (its definitely sente).

Strategy2.tst:

68: D13 kills, too.

73, 74: This one is very difficult. However, I am almost certain that GnuGo
is right, and that R17 kill the black dragon. Black R18 definitely makes
life. (Would be quite impressive if GnuGo really read this out
correctly! :-) )

84: GnuGo sees that G18 can be defended with ko. But I agree with GnuGo that
Q13 is more urgent. (In fact, I think both moves are valued quite
reasonably.)

Strategy4.tst

197: With S18, Gnugo gets a favorable one-step ko in the corner
(if W S19, B looses the semeai). So I think this is larger than K3.

198: Same as 197. S18 should be fine.

199: Same as 197, although here it is less clear (W might live with N5,
but I tend to agree with GnuGo here, who thinks it is still dead).

200: Same as 197. Further, I don't even think P6/P7 is necessary to kill
the white N6 dragon. I assume these 4 tests were introduced before
GnuGo was able to read out UR corner, so it might be best to drop three
of them and keep the one where it is most obvious that S18 is correct
(probably 198).


Patches:

--- regression/strategy.tst     Wed Nov  7 03:07:13 2001
+++ regression/strategy-1.14.tst        Fri Nov  9 22:05:25 2001
@@ -81,19 +81,6 @@
 14 gg_genmove black
 #? [E9]
 
-#CATEGORY=BAD_PROBLEM
-#DESCRIPTION=Problem not clear enough.
-#SEVERITY=0
-#From STRATEGY:
-#     It is unclear that the recommended E9 is the best move. 
-#     W cannot cut at E11, or even threaten to cut and keep sente. 
-#     GNU Go 2.7.205 plays at G10 which is a good move.
-# incident 123
-# Again, E9 is not necessarily the best move.
-loadsgf games/incident121.sgf 60
-15 gg_genmove black
-#? [E9]*
-
 # incident 132
 loadsgf games/incident121.sgf 116
 16 gg_genmove black
@@ -121,9 +108,10 @@
 # incident 166a
 # E5 is an overplay
 # See reading test 124
+# F3 seems fine to protect the two cuts -arend
 loadsgf games/incident165.sgf 142
 18 gg_genmove black
-#? [E4]*
+#? [E4|F3]
 
 loadsgf games/strategy4.sgf
 19 gg_genmove black
@@ -244,9 +232,10 @@
 #M11 is better than L4
 #    2.7.231: Broke again. M11 is overrated at 28 points. The
 #    large followup value is responsible.
+# I disagree. M11 makes territory while attacking. -arend
 loadsgf games/nicklas/nicklas8.sgf 98
 40 gg_genmove black
-#? [E5]*
+#? [E5|M11]*
 
 
 #CATEGORY=STRATEGY
--- regression/strategy2.tst    Fri Nov  9 22:05:39 2001
+++ regression/strategy2-1.14.tst       Fri Nov  9 22:08:00 2001
@@ -94,7 +94,7 @@
 # incident 177 (from the disastrous Indigo game)
 loadsgf games/incident169.sgf 184
 68 gg_genmove white
-#? [F14]*
+#? [D13|F14]*
 
 # incident 183 (from the disastrous Indigo game)
 loadsgf games/incident169.sgf 300
@@ -118,13 +118,15 @@
 72 gg_genmove white
 #? [A8]*
 
+# I have added R17, as it seems to kill the huge black dragon --arend
 loadsgf games/semeai2.sgf 116
 73 gg_genmove white
-#? [F7]*
+#? [F7|R17]*
 
+# I have added R17, as it seems to kill the huge black dragon --arend
 loadsgf games/semeai2.sgf 118
 74 gg_genmove white
-#? [F7]*
+#? [F7|R17]*
 
 loadsgf games/semeai2.sgf 150
 # If Q11 is not undervalued F7 is acceptable (though not correct)
@@ -172,8 +174,9 @@
 
 loadsgf games/heikki/heikki01.sgf 35
 # J18 makes ko
+# But Q13 seems larger -arend
 84 gg_genmove black
-#? [J18]*
+#? [Q13]*
 
 loadsgf games/strategy24.sgf
 85 gg_genmove white