gnugo-devel
[Top][All Lists]
Advanced

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

[gnugo-devel] Bug report


From: Arend Bayer
Subject: [gnugo-devel] Bug report
Date: Thu, 8 Nov 2001 14:20:58 +0100 (CET)

Hello again,

I just came across a bug that caused gnugo-3.1.13 make a serious blunder.
(It does not happen with the version 3.0.0, but that could is a side
effect. The blunder is suggested by the endgame database, which is not
invoked by 3.0.0, as it values an alternatve slightly higher than 6 pts.
Including the responsible pattern CE6 in patterns.db causes GnuGo-3.0.0
to make the same mistake.)

I have attached the sgf-file, and describe the bug below.

Arend


Move 159:
Gnugo 3.1.13 move at D15, which costs quite a lot. Here are the relevant
outputs of gnugo -L159 -l exper2.sgf -t:

(...)
>Endgame pattern matcher is looking for move reasons for black!
(...)
>pattern 'CE6'+3 matched at D15
>...expands territory
>...(min) territorial value 0.10
(...)
>pattern 'CE6'+4 matched at D15
>...expands territory
>...(min) territorial value 0.10
(...)
>
>Move at D15 expands territory
(...)
>
>  D15: -14.00 - change in territory
>  D15:   0.10 - revised to meet minimum territory value
>  D15: 7.48 - influence
>  D15: -1.03 - shape (shape values +0.00(0) -3.00(1), shape factor 0.864)
>Move generation values D15 to 6.55
(...)
>
>
>Top moves:
>1. D15 6.55
(...)
>
>Checking safety of a black move at D15
>Move generation likes D15 with value 6.55
>genmove() recommends D15 with value 6.55
>black (X) move D15

Of course, one could just change move_reasons.c and not allow the change
from such a large negative territory value to the minimum territory value.
But the high positive influence value is also nonsense; I do not know
where this comes from.  Maybe this mistake and the following are caused
by the same bug?

Move 160:
gnugo -l exper2.sgf -L160 -w

lists C12 and B13 as different dragons, and B13 as alive. (At move
158, it is still considered (owl-)dead.)
Relevant output:

(...)
>B13 : (dragon B13) black string of size 1 (1.67), genus 0: (2,2,3,3) - is a 
>potential cutting stone
>- attack threat at B12, attack threat code = 3
>- attack threat at B11, attack threat code = 3
>- attack threat at A12, attack threat code = 3
>- attack threat at C13, attack threat code = 3
>C12 : (dragon C12) black string of size 1 (3.08), genus 0: (4,3,3,3)
>- attackable at D12, attack code = 1
>- defendable at D12, defend code = 3
>- defendable at A12, defend code = 3
>- defendable at C14, defend code = 3
>- defendable at E12, defend code = 3
>- defendable at C11, defend code = 3
>- defendable at B12, defend code = 3
>- defendable at C13, defend code = 3
>- is inessential
(...)
B13 : B dragon size 1 (1.67), genus 0, half eyes 0, escape factor 0, status 
dead, matcher status alive, moyo size 0 safety weakly_alive, owl status alive
... neighbors E15 C12 C10
C12 : B dragon size 1 (3.08), genus 0, half eyes 0, escape factor 0, status 
dead, matcher status dead, moyo size 0 safety dead, owl status dead
... neighbors B13 E15 C10

Attachment: exper2.sgf
Description: Text document


reply via email to

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