[Top][All Lists]

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [gnugo-devel] Scoring error

From: Gunnar Farneback
Subject: Re: [gnugo-devel] Scoring error
Date: Mon, 09 Sep 2002 19:58:27 +0200
User-agent: EMH/1.14.1 SEMI/1.14.3 (Ushinoya) FLIM/1.14.2 (Yagi-Nishiguchi) APEL/10.3 Emacs/20.7 (sparc-sun-solaris2.7) (with unibyte mode)

Evan wrote:
> I played gnugo-3.3.8-evand-200209090532.sgf online.  When done scoring,
> gnugo marks N1 as dead, therefore losing the game (it should actually win
> by a few points).
> I can't reproduce this, but the sgf file shows N1 being removed.

Try this:

gnugo --quiet -l gnugo-3.3.8-evand-200209090532.sgf --score finish

173 black (X) move Q14
174 white (O) move R17
175 black (X) move Q18
176 white (O) move P16
177 black (X) move R18
178 white (O) move R13
179 black (X) move S19
180 white (O) move T17
181 black (X) move T18
182 white (O) move S18
183 black (X) move B18
184 white (O) move S17
185 black (X) move A19
186 white (O) move A9
187 black (X) move A7
188 white (O) move B19
189 black (X) move R15
190 white (O) move S15
191 black (X) move A19
192 white (O) move B6
193 black (X) move B7
194 white (O) move B19
195 black (X) move M1
196 white (O) move B17
197 black (X) move O1
198 white (O) move L1
199 black (X) move N1
200 white (O) move P1
201 black (X) move : PASS!
202 white (O) move : PASS!
White wins by 9.5 points
Result from file: 96.5
GNU Go result and result from file are different

What we see here is that GNU Go playing for both colors finishes the
game with a complicated ko fight. When that happens things tend to go
badly wrong with the scoring.

What you see on NNGS is the output of final_status_list, which
considers N1 dead because it was captured after the sequence above.

The first problem here is that for black, GNU Go thinks the Q14 move
is useful. The second problem is that white doesn't see the importance
of answering at R13.

Mistakes like this can be added to score2.tst. Although that file
isn't run as part of the regular regressions it will be tested when
someone (usually me) tries to improve the scoring.


reply via email to

[Prev in Thread] Current Thread [Next in Thread]