[Top][All Lists]
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
[gnugo-devel] Regression file suggestion & patch
From: |
Trevor Morris |
Subject: |
[gnugo-devel] Regression file suggestion & patch |
Date: |
Fri, 05 Oct 2001 14:17:49 -0400 |
Here's a patch that does the following:
- Cleans up a few typos & mistakes in the current regression suite.
- Replaces several occurences of genmove_black w/ gg_genmove black.
- New GoBoard::Stone Perl Module for generate PNG files.
- Updates regress.pl to generate graphics, not text in HTML view.
- Implements the following suggestion in several places.
http://www.public32.com/games/go/trevor_1_9.4.tar.gz
[This file includes the patch file, one new SGF file, and one
new Perl module for generating stone images.]
Suggestion:
Add the following 3 comment lines to existing failure cases:
#CATEGORY={the program area (of a defined set) that needs fixing.
Current options:
("JOSEKI_DATABASE, "",
"JOSEKI_PATTERN", "",
"FUSEKI_CONCEPT", "",
"DYNAMIC_CONNECTION", "Dynamic Connection Reading",
"TACTICAL_READING",
"",
"OWL_TUNING", "",
"PATTERN_TUNING", "",
"CONNECTION_TUNING", "",
"MOVE_VALUATION", "",
"ATARI_ATARI", "",
"SEMEAI_MODULE", "",
"KO_READING", "")
#DESCRIPTION=A short one-line description of the
problem.
#SEVERITY={0-10} 0:ignore; 1:low priority; 10:Very important.
For example:
#CATEGORY=OWL_TUNING
#DESCRIPTION=W group needs help & 2 black stones
attacking.
#SEVERITY=8
#CATEGORY=PATTERN_TUNING
#DESCRIPTION=No terribly serious consequences, S11
not optimal.
#SEVERITY=3
The idea is that regress.pl will sort the failures appropriately
into areas, sorted by severity.
Here's the most interesting part of the patch text, but don't
try to apply it - use the gz file listed above, instead:
Index: NICKLAS
===================================================================
RCS file: /cvsroot/gnugo/gnugo/regression/NICKLAS,v
retrieving revision 1.3
diff -u -r1.3 NICKLAS
--- NICKLAS 2001/09/06 21:34:20 1.3
+++ NICKLAS 2001/10/05 17:51:37
@@ -1,3 +1,8 @@
+****GnuGo 3.1.8 still misses these 4 problems.
+****10/5/01 - moved comments into nicklas2.tst.
+****This file now should be considerred obsolete.
+
+
nicklas1.tst: all these tests should pass
nicklas2.tst: 9x9 games where gnugo should do better.
nicklas3.tst: 9x9 games with harder problems
@@ -8,7 +13,7 @@
NICKLAS2.TST
#102: LOW PRIORITY; DIFFICULTY HARD; DEFER
-gnugo --quiet -l niclas1.sgf -L29 -t
+gnugo --quiet -l nicklas1.sgf -L29 -t
This is not as clear cut as it may seem. 2.7.228 values the endgame
move at A7 to 6.66 points which seems about correct. The fact that
Index: dniwog.tst
===================================================================
RCS file: /cvsroot/gnugo/gnugo/regression/dniwog.tst,v
retrieving revision 1.3
diff -u -r1.3 dniwog.tst
--- dniwog.tst 2001/09/06 21:34:20 1.3
+++ dniwog.tst 2001/10/05 17:51:37
@@ -16,14 +16,24 @@
4 gg_genmove white
#? [D7]
+#CATEGORY=OWL_TUNING
+#DESCRIPTION=W group needs help & 2 black stones attacking.
+#SEVERITY=8
loadsgf games/dniwog.sgf 78
5 gg_genmove white
#? [H4|G4]*
+
+#CATEGORY=PATTERN_TUNING
+#DESCRIPTION=A more clear-cut case for this tesuji might help.
+#SEVERITY=7
loadsgf games/dniwog.sgf 88
6 gg_genmove white
#? [G1]*
+#CATEGORY=PATTERN_TUNING
+#DESCRIPTION=Safing right side group is better than terrible gote move!
+#SEVERITY=10
loadsgf games/dniwog.sgf 96
7 gg_genmove white
#? [B8]*
Index: ego.tst
===================================================================
RCS file: /cvsroot/gnugo/gnugo/regression/ego.tst,v
retrieving revision 1.7
diff -u -r1.7 ego.tst
--- ego.tst 2001/09/15 00:24:22 1.7
+++ ego.tst 2001/10/05 17:51:37
@@ -13,11 +13,18 @@
2 gg_genmove white
#? [!E5]
+
+#CATEGORY=PATTERN_TUNING
+#DESCRIPTION=K17 misses H18 threat.
+#SEVERITY=8
# K18 is unreasonable. H18 is the only move locally.
loadsgf games/ego.sgf 52
3 gg_genmove white
#? [H18]*
+#CATEGORY=PATTERN_TUNING
+#DESCRIPTION=Possible replacement pattern appropriate?
+#SEVERITY=8
# The descent at G19 is bad shape because it leaves much aji around
# E18 and gives black an endgame sente move at H19.
loadsgf games/ego.sgf 54
@@ -38,6 +45,10 @@
7 gg_genmove white
#? [!Q1]
+
+#CATEGORY=BORDER_PROTECTION
+#DESCRIPTION=Tough moyo protection.
+#SEVERITY=5
loadsgf games/ego.sgf 112
8 gg_genmove white
#? [F12]*
@@ -50,6 +61,10 @@
10 gg_genmove white
#? [S18]
+
+#CATEGORY=KO_READING
+#DESCRIPTION=Tough - Black pushthrough is troublesome.
+#SEVERITY=3
loadsgf games/ego.sgf 198
11 gg_genmove white
#? [T18]*
Index: manyfaces.tst
===================================================================
RCS file: /cvsroot/gnugo/gnugo/regression/manyfaces.tst,v
retrieving revision 1.3
diff -u -r1.3 manyfaces.tst
--- manyfaces.tst 2001/09/06 21:34:20 1.3
+++ manyfaces.tst 2001/10/05 17:51:38
@@ -5,43 +5,61 @@
reset_trymove_counter
loadsgf games/mfgg1.sgf 43
-1 genmove_black
+1 gg_genmove black
#? [S13]
loadsgf games/mfgg1.sgf 49
-2 genmove_black
+2 gg_genmove black
#? [S16]
+#CATEGORY=PATTERN_TUNING
+#DESCRIPTION=No terribly serious consequences, S11 not optimal.
+#SEVERITY=3
loadsgf games/mfgg1.sgf 65
-3 genmove_black
+3 gg_genmove black
#? [S12]*
loadsgf games/mfgg1.sgf 97
-4 genmove_black
+4 gg_genmove black
#? [!T6]
+#CATEGORY=FUSEKI_STRATEGY
+#DESCRIPTION=N6 is not terrible - right concept.
+#SEVERITY=2
loadsgf games/mfgg2.sgf 19
-5 genmove_black
+5 gg_genmove black
#? [M5|M15|R3|R17]*
+
+#CATEGORY=FUSEKI_STRATEGY
+#DESCRIPTION=E6 really misses the point.
+#SEVERITY=9
loadsgf games/mfgg2.sgf 23
-6 genmove_black
+6 gg_genmove black
#? [R3]*
+#CATEGORY=OWL_TUNING
+#DESCRIPTION=Q5 dragon is under severe pressure.
+#SEVERITY=8
loadsgf games/mfgg2.sgf 45
-7 genmove_black
+7 gg_genmove black
#? [R2|S7]*
+
+#CATEGORY=OWL_TUNING
+#DESCRIPTION=GnuGo doesn't understand this type of corner position!
+#SEVERITY=10
+#GG often messes up in this type of corner position.
loadsgf games/mfgg2.sgf 117
-8 genmove_black
+8 gg_genmove black
#? [T2]*
loadsgf games/mfgg3.sgf 42
-9 genmove_black
+9 gg_genmove black
#? [!P12]
loadsgf games/mfgg3.sgf 44
-10 genmove_black
+10 gg_genmove black
#? [N3]*
Index: nicklas1.tst
===================================================================
RCS file: /cvsroot/gnugo/gnugo/regression/nicklas1.tst,v
retrieving revision 1.9
diff -u -r1.9 nicklas1.tst
--- nicklas1.tst 2001/10/01 03:49:08 1.9
+++ nicklas1.tst 2001/10/05 17:51:41
@@ -21,6 +21,9 @@
202 gg_genmove black
#? [!H8]
+#CATEGORY=OWL_TUNING
+#DESCRIPTION=Combination threat to live / connect.
+#SEVERITY=4
loadsgf games/nicklas/nicklas2.sgf 23
203 gg_genmove black
#? [J5]*
@@ -74,10 +77,13 @@
810 gg_genmove black
#? [!C19]
-# Moved K15 to K14. /gf
+# Moved K15 to K14. /gf ; Added back K14 - both work.
+#SEVERITY=7
+#CATEGORY=CONNECTION_TUNING
+#DESCRIPTION=Is GnuGo misreading the cut?
loadsgf games/nicklas/nicklas8.sgf 240
811 gg_genmove black
-#? [K14]*
+#? [K14|K15]*
loadsgf games/nicklas/nicklas9.sgf 28
901 gg_genmove black
@@ -119,6 +125,9 @@
1106 gg_genmove black
#? [!F6]
+#CATEGORY=WASTED_MOVE
+#DESCRIPTION=GnuGo 3.1.8 gets N2 on next move; R3 benign ko threat loss.
+#SEVERITY=2
loadsgf games/nicklas/nicklas11.sgf 242
1107 gg_genmove black
#? [N2]*
@@ -151,6 +160,9 @@
1206 gg_genmove black
#? [C15]
+#CATEGORY=PATTERN_TUNING
+#DESCRIPTION=Needless 2 point loss for B.
+#SEVERITY=7
loadsgf games/nicklas/nicklas12.sgf 113
1207 gg_genmove black
#? [C17]*
@@ -268,6 +280,9 @@
1605 gg_genmove black
#? [J1]
+#CATEGORY=OWL_TUNING
+#DESCRIPTION=No reason to give W a chance for ko here.
+#SEVERITY=6
loadsgf games/nicklas/nicklas18.sgf 15
1801 gg_genmove black
#? [D2]*
Index: nicklas2.tst
===================================================================
RCS file: /cvsroot/gnugo/gnugo/regression/nicklas2.tst,v
retrieving revision 1.8
diff -u -r1.8 nicklas2.tst
--- nicklas2.tst 2001/09/24 14:05:44 1.8
+++ nicklas2.tst 2001/10/05 17:51:42
@@ -11,18 +11,50 @@
reset_reading_node_counter
reset_trymove_counter
+#CATEGORY=ATARI_ATARI
+#DESCRIPTION=GnuGo misses followup to D4 push-through.
+#SEVERITY=1
+# This is not as clear cut as it may seem. 2.7.228 values the endgame
+# move at A7 to 6.66 points which seems about correct. The fact that
+# white D4 threatens both to cut off E5-F6 and to capture D3 through
+# a chain of threats (which is what happens in the game) is not that
+# easy to understand. A more general atari_atari module could
+# perhaps see it, but we are still far from that. I think this test
+# should be in nicklas3.tst (harder problems).
loadsgf games/nicklas/nicklas1.sgf 29
102 gg_genmove black
#? [D4]*
+#CATEGORY=PATTERN_TUNING
+#DESCRIPTION=I think G9 is not too terrible here. Very difficult to
see.-trevor
+#SEVERITY=1
+# This test illustrates a general weakness of GNU Go: It's poor
+# ability to defend its borders. White can enter blacks domains or
+# capture the E8 group. In the latter case, whites first move is
+# a sacrifice and the capture happens through a chain of threats.
+#Added E6, still not thinking G9 is so bad. -trevor
loadsgf games/nicklas/nicklas6.sgf 23
601 gg_genmove black
-#? [D8|C8|D6|D4]*
+#? [D8|C8|D6|D4|E6]*
+#CATEGORY=SEMEAI_MODULE
+#DESCRIPTION=Very tough position. Even if F1 for B, W E7 is tough.
+#SEVERITY=1
+# This is a very good test case for a semeai module. Black F1 threatens
+# to live and creates all the necessary liberties to win the semai.
+#I disagree; this it tough; a more clear-cut test case would be helpful
+#to better isolate the inherent problem. - trevor
loadsgf games/nicklas/nicklas7.sgf 29
701 gg_genmove black
#? [F1]*
+
+#CATEGORY=OWL_TUNING
+#DESCRIPTION=Tough for white to see attacking B, and linking w/ attack stone.
+#SEVERITY=3
+# The only way to live is to attack a neighbour dragon, e.g. the one
+# at F8. But does the owl code really do this by default?
+#
loadsgf games/nicklas/nicklas9.sgf 28
902 dragon_status E9
#? [critical]*
@@ -45,35 +77,68 @@
1301 gg_genmove black
#? [B8|B9]
+#CATEGORY=TACTICAL_READING
+#DESCRIPTION=Fighting Ko here is not unreasonable for B.
+#SEVERITY=0
+#I would delete this regression test. -trevor
loadsgf games/nicklas/nicklas14.sgf 31
1401 gg_genmove black
#? [B3]*
+#CATEGORY=OWL_TUNING
+#DESCRIPTION=Black must protect his group. H9 is thinkable also.
+#SEVERITY=9
loadsgf games/nicklas/nicklas14.sgf 43
1402 gg_genmove black
#? [J8|J6]*
+#CATEGORY=OWL_TUNING
+#DESCRIPTION=PASS here is unthinkable, though RESIGN might be appropriate!
+#SEVERITY=5
# One eye wins against no eye
+#Note: Owl & Dragon status is both critical for the W group. Why
+# would B pass when there's a critical group on the board? Owl &
+# Dragon status for all of the B stones is Dead. Can a Critical
+# group kill another group???
loadsgf games/nicklas/nicklas14.sgf 61
1407 gg_genmove black
#? [A6]*
+
+#CATEGORY=PATTERN_TUNING
+#DESCRIPTION=Protect those borders!
+#SEVERITY=8
+#Added A8, for the faint of heart.
loadsgf games/nicklas/nicklas17.sgf 47
1701 gg_genmove black
-#? [C9]*
+#? [C9|A8]*
loadsgf games/nicklas/nicklas18.sgf 17
1802 gg_genmove black
#? [E2]
+
+#CATEGORY=WASTED_MOVE
+#DESCRIPTION=Um, you're still dead after B1, but sente!
+#SEVERITY=1
+#SEE_ALSO=trevor.tst?310
loadsgf games/nicklas/nicklas18.sgf 45
1803 gg_genmove black
#? [A8]*
+
+#CATEGORY=ENDGAME_TUNING
+#DESCRIPTION=OWL & Dragon status for W is all dead. Why does B play?
+#SEVERITY=5
loadsgf games/nicklas/nicklas21.sgf 72
2102 gg_genmove black
#? [PASS]*
+
+#CATEGORY=ENDGAME_TUNING
+#DESCRIPTION=OWL & Dragon status for W is all dead. Why does B play?
+#SEVERITY=5
+#See also nicklas2.tst?2102
loadsgf games/nicklas/nicklas21.sgf 74
2103 gg_genmove black
#? [PASS]*
@@ -90,6 +155,9 @@
2202 gg_genmove black
#? [B1]
+#CATEGORY=ATARI_ATARI
+#DESCRIPTION=Black must defend here - yikes!
+#SEVERITY=10
loadsgf games/nicklas/nicklas24.sgf 41
2401 gg_genmove black
#? [G3]*
Index: niki.tst
===================================================================
RCS file: /cvsroot/gnugo/gnugo/regression/niki.tst,v
retrieving revision 1.5
diff -u -r1.5 niki.tst
--- niki.tst 2001/10/02 20:04:18 1.5
+++ niki.tst 2001/10/05 17:51:42
@@ -9,10 +9,16 @@
1 gg_genmove white
#? [E17]
+#CATEGORY=MIDDLE_STRATEGY
+#DESCRIPTION=Tough middle-game position
+#SEVERITY=3
loadsgf games/niki.sgf 40
2 gg_genmove white
#? [E7]*
+#CATEGORY=PATTERN_TUNING
+#DESCRIPTION=Must connect 5 important stones, cutting B group.
+#SEVERITY=10
loadsgf games/niki.sgf 68
3 gg_genmove white
#? [E12]*
@@ -21,14 +27,20 @@
4 gg_genmove white
#? [Q8]
+#S10 seems OK, much better than game move of S14. Added it -trevor
+#Hm, should GG3.1.9 get credit for this? I think so! -trevor
loadsgf games/niki.sgf 94
5 gg_genmove white
-#? [S9]*
+#? [S9|S10]*
loadsgf games/niki.sgf 106
6 gg_genmove white
#? [!S18]
+
+#CATEGORY=PATTERN_TUNING
+#DESCRIPTION=T4 is terrible; other moves than J12 maybe OK.
+#SEVERITY=6
loadsgf games/niki.sgf 110
7 gg_genmove white
#? [J12]*
@@ -36,6 +48,9 @@
loadsgf games/niki.sgf 124
8 gg_genmove white
#? [G16]*
+
+#gg_genmove bug requires reloading game.
+loadsgf games/niki.sgf 124
9 gg_genmove black
#? [F15]*
@@ -43,13 +58,25 @@
10 gg_genmove white
#? [S5]*
+
+#CATEGORY=TACTICAL_READING
+#DESCRIPTION=Very tough position
+#SEVERITY=3
+#Note: G16 is still very very big. See niki.tst#8 & nikit.tst#9
# FIXME: Same move listed twice.
loadsgf games/niki.sgf 158
11 gg_genmove white
#? [M6|M6|P7]*
+
+#CATEGORY=OWL_TUNING
+#DESCRIPTION=B15 does live. Problem is w/ followup - see niki.tst#14
+#SEVERITY=0
+#
# Originally H12 was listed as correct here, but that vertex is
# occupied. F14 and G15 seem sufficient for life.
+#
+#Recommend adding B15 as acceptable to live.
loadsgf games/niki.sgf 212
12 gg_genmove white
#? [F14|G15]*
@@ -57,6 +84,14 @@
loadsgf games/niki.sgf 226
13 gg_genmove white
#? [F8]
+
+
+#CATEGORY=OWL_TUNING
+#DESCRIPTION=F16 lives because of B short of liberties
+#SEVERITY=8
+loadsgf games/niki.sgf 214
+14 gg_genmove white
+#? [F16]*
# Report number of nodes visited by the tactical reading
10000 get_reading_node_counter
Index: strategy.tst
===================================================================
RCS file: /cvsroot/gnugo/gnugo/regression/strategy.tst,v
retrieving revision 1.8
diff -u -r1.8 strategy.tst
--- strategy.tst 2001/10/01 03:49:08 1.8
+++ strategy.tst 2001/10/05 17:51:46
@@ -5,7 +5,7 @@
reset_trymove_counter
loadsgf games/strategy1.sgf
-1 genmove_white
+1 gg_genmove white
#? [Q13]
loadsgf games/strategy2.sgf 5
Index: tactics.tst
===================================================================
RCS file: /cvsroot/gnugo/gnugo/regression/tactics.tst,v
retrieving revision 1.2
diff -u -r1.2 tactics.tst
--- tactics.tst 2001/10/01 03:49:08 1.2
+++ tactics.tst 2001/10/05 17:51:46
@@ -5,7 +5,7 @@
reset_trymove_counter
loadsgf games/tactics01.sgf
-1 genmove_black
+1 gg_genmove black
#? [C15|D15|B13|B11|B14]*
############ End of Tests #################
Index: trevor.tst
===================================================================
RCS file: /cvsroot/gnugo/gnugo/regression/trevor.tst,v
retrieving revision 1.5
diff -u -r1.5 trevor.tst
--- trevor.tst 2001/10/03 20:57:57 1.5
+++ trevor.tst 2001/10/05 17:51:48
@@ -305,6 +305,9 @@
301 gg_genmove black
#? [!PASS]*
-
-
-
+#CATEGORY=OWL_TUNING
+#DESCRIPTION=Black still dead after D1!
+#SEVERITY=10
+loadsgf games/trevor/trevor_31.sgf
+310 gg_genmove black
+#? [A8]*
[Prev in Thread] |
Current Thread |
[Next in Thread] |
- [gnugo-devel] Regression file suggestion & patch,
Trevor Morris <=