Updated README: Equal sign not required with `--mode` flag.
[sgk-go] / regression / arend.tst
CommitLineData
7eeb782e
AT
1# CATEGORY=FUSEKI_STRATEGY
2# SEVERITY=5
3loadsgf games/arend/gnugo-gnugo1.sgf 25
41 reg_genmove black
5#? [C4]*
6
7# CATEGORY=KO
8# SEVERITY=3
9loadsgf games/arend/gnugo-gnugo1.sgf 112
102 reg_genmove white
11#? [H15]
12
13# CATEGORY=CONNECTION
14# SEVERITY=10
15loadsgf games/arend/gnugo-gnugo1.sgf 139
163 reg_genmove black
17#? [P6|Q5]
18
19# These kind of fights might be very difficult to do well
20# with current GNU Go methods (How should it choose between the different
21# 'correct' alternatives here? - R14 seems clearly best to me).
22# But certainly it should not tenuki.
23loadsgf games/arend/exper1.sgf 7
244 reg_genmove black
25#? [R14|S15|R17]
26
27#CATEGORY=STRATEGY
28#DESCRIPTION=GNU Go should escape with its cutting stone.
29loadsgf games/arend/exper1.sgf 25
305 reg_genmove black
31#? [D7|E7|E6]*
32
33#CATEGORY=STRATEGY
34#DESCRIPTION=In this position, GNU Go plays L10 because of an enormous assumed strategic effect on the M16 dragon, which is quite safe.
35loadsgf games/arend/gnugo-gnugo2.sgf 57
367 reg_genmove black
37#? [C3]
38
39#CATEGORY=BLUNDER
40# S 16 can be captured. S17 is better, but another big move
41# should be accepted as well
42loadsgf games/arend/gnugo-gnugo2.sgf 60
438 reg_genmove white
44#? [!S16]
45
46#CATEGORY=TACTICAL_READING
47# GNU Go rejects S17 because it is "strategically or tactically unsafe".
48# Must be a misread.
49loadsgf games/arend/gnugo-gnugo2.sgf 61
509 reg_genmove black
51#? [S17]
52
53#CATEGORY=STRATEGY
54# J7 is extremely overvalued. This is caused by pattern CB11.
55loadsgf games/arend/gnugo-gnugo2.sgf 73
5610 reg_genmove black
57#? [!J7]
58
59#CATEGORY=STRATEGY
60# K6 is overvalued by GNU Go 3.0.0.
61loadsgf games/arend/gnugo-gnugo2.sgf 91
6211 reg_genmove black
63#? [!K6]
64
65#CATEGORY=BLUNDER
66loadsgf games/arend/gnugo-gnugo2.sgf 163
6712 reg_genmove black
68#? [C16|B16]
69
70#CATEGORY=STRATEGY
71# GNU Go does not see that P17 helps his group as well, and plays O13.
72loadsgf games/arend/gnugo-gnugo4.sgf 27
7314 reg_genmove black
74#? [P17]*
75
76#CATEGORY=FUSEKI_STRATEGY
77# C6 is clearly the wrong side for approaching LL corner.
78loadsgf games/arend/gnugo-gnugo4.sgf 31
7915 gg_genmove black 1
80#? [F3]*
81
82
83# GNU Go loves this move! (H2 is the point to play here.) It is impossible
84# to teach GNU Go not to play there with minor pattern tuning. Problem is
85# huge follow_up value; there is no check at all in the move valuation
86# whether a sente move is actually worth something if opponent answers
87# (here it is aji-keshi).
88# This looses only a few points, but comes up very often.
89loadsgf games/arend/gnugo-gnugo4.sgf 45
9016 reg_genmove black
91#? [!G3]
92
93#CATEGORY=TACTICAL_READING
94# GNU Go moves at C9. I do not know what it is afraid of.
95loadsgf games/arend/gnugo-gnugo4.sgf 153
9617 reg_genmove black
97#? [C7]
98
99# Maybe this problem is suboptimal, because there is a ko involved, and
100# GNU Go gets it wrong even without the ko.
101loadsgf games/arend/gnugo-gnugo4.sgf 168
10218 reg_genmove white
103#? [P18|P19|R18|R19|S18|S17|S16|S14|S13|T14|T16]
104
105#CATEGORY=PATTERN_TUNING
106# GNU Go often plays this inefficient shape. Should be correctable with an
107# additional pattern. Core of the problem is however that GNU Go does not
108# understand the essence of the iken tobi -- it can be cut, but Black does
109# not mind that, since a cut only strengthens Black.
110loadsgf games/arend/gnugo-gnugo5.sgf 39
11120 reg_genmove black
112#? [!L15]
113
114#CATEGORY=STRATEGY
115# This inefficient move is produced by owl. But of course it should not
116# be owl in charge of strengthening this big dragon.
117loadsgf games/arend/gnugo-gnugo5.sgf 78
11821 reg_genmove white
119#? [!F9]
120
121# Immediately take the double sente point!
122loadsgf games/arend/gnugo-gnugo5.sgf 80
12322 reg_genmove white
124#? [D14]*
125
126# H10 is quite overvalued here.
127loadsgf games/arend/gnugo-gnugo5.sgf 105
12823 reg_genmove black
129#? [!H10]
130
131# Low Shimari clearly better here
132# See also arend2:130
133loadsgf games/arend/gnugo-gnugo6.sgf 20
13424 restricted_genmove white P16 P17
135#? [P17]*
136
137#CATEGORY=ENDGAME_TUNING
138# C19 is either double sente, or worth at least 5 pts in reverse sente,
139# which is 10 pts. T14 (chosen by GNU Go) is dame. F7 is nearly as large
140# as C19.
141loadsgf games/arend/gnugo-gnugo6.sgf 174
14225 reg_genmove white
143#? [C19|F7]
144
145#CATEGORY=FUSEKI_STRATEGY
146# D17 is wrong direction; in this position this is really bad.
147loadsgf games/arend/gnugo-gnugo7.sgf 17
14828 reg_genmove black
149#? [C16]
150
151#CATEGORY=FUSEKI_STRATEGY
152# Black C15 is sente here, so W C14 is mandatory.
153loadsgf games/arend/gnugo-gnugo7.sgf 24
15429 reg_genmove white
155#? [B14|C14]
156
157#CATEGORY=BLUNDER
158# GNU Go played F8.
159loadsgf games/arend/gnugo-gnugo7.sgf 79
16030 reg_genmove black
161#? [C13]*
162
163#CATEGORY=OWL_DEFEND (?)
164# B14 can (and should) be saved here
165loadsgf games/arend/gnugo-gnugo7.sgf 81
16631 reg_genmove black
167#? [B13|C13]
168
169#CATEGORY=CONNECTION
170#SEVERITY=9
171loadsgf games/arend/gnugo-gnugo7.sgf 137
17232 reg_genmove black
173#? [F9|E8|G8]
174
175loadsgf games/arend/gnugo-gnugo7.sgf 140
17633 reg_genmove white
177#? [F9]*
178
179#CATEGORY=ENDGAME_TUNING
180loadsgf games/arend/gnugo-gnugo7.sgf 173
18134 reg_genmove black
182#? [!K8]
183
184# This is a serious mistake. H17 is not huge, its HUGE.
185loadsgf games/arend/gnugo-gnugo8.sgf 32
18635 reg_genmove white
187#? [H17|J19]
188
189#CATEGORY=ENDGAME_TUNING
190# Its just one point difference, but it happens a couple of times every
191# game.
192loadsgf games/arend/constructed1.sgf
19336 reg_genmove black
194#? [C5]
195
196#CATEGORY=ENDGAME_TUNING
197loadsgf games/arend/constructed2.sgf
19837 reg_genmove black
199#? [A5]
200
201#CATEGORY=STRATEGY
202# I think this problem summarizes well two common misjudgements
203# of the influence function:
204# 1. Breaking into an opponents moyo is enormously
205# undervalued (that is why patterns like LE14--safe jump towards
206# oppoents moyo--need to have a fixed minimum value, which often leads
207# bad moves)
208# 2. O5 considerably weakens the influence of N5. Maybe this is not weighted
209# highly enough by GNU Go?
210loadsgf games/arend/constructed3.sgf
21138 reg_genmove white
212#? [O5]
213
214#CATEGORY=STRATEGY
215# This is an example for part 1 of my comment of the previous problem.
216# This slightly adapted from a game (to make the problem more specific).
217loadsgf games/arend/constructed4.sgf
21839 reg_genmove black
219#? [G10|H10|F13]
220
221#CATEGORY=ENDGAME_TUNING
222# EB714 is necessary here; without it, GNU Go played B6, only for the
223# reason of filling a liberty
224loadsgf games/arend/constructed5.sgf
22540 reg_genmove white
226#? [B7]
227
228#CATEGORY=ENDGAME_TUNING
229# EB715 necessary
230loadsgf games/arend/constructed6.sgf
23141 reg_genmove white
232#? [B6]