Re: [gnugo-devel] scoring bug?

[Terry McIntyre]

glGo does expect users to mark the dead stones. I'm not sure that this
question is related to gnugo at all. It is possible to configure glGo 
to use gnugo ( as I have on my computer ), but the scoring system 
appears to be part of glGo, not of the go-playing engine.

I just created a file using glGo playing against Gnugo, and the first 
few lines follow:

(;GM[1]FF[4]AP[glGo:1.3.1]ST[1]

SZ[9]KM[6.5]
PW[GNU Go]PB[Black]DT[2005-11-16]
   ^^^^^^^

--- so I think Todd is using glGo to play some other engine, but the 
scoring problem is not part of the engine, but of the way glGo works - 
one must explicitly mark the dead stones. glGo shows the change, and 
counts the remaining da-me - when glGo's count of da-me agrees with 
yours, the score will be accurate.


--
Terry McIntyre

On Nov 16, 2005, at 12:37 PM, Gunnar Farnebäck wrote:

> Todd Sprang wrote:
> > if you examine this game, by the typical rules i've seen everywhere
> > else, the score would be black 23:white 21 (one black stone captured
> > during play). the 4 black stones in the white territory are considered
> > dead as they stand and are added to white's territory score of 16.
>
> Yes, black wins by 2.
>
> > instead, gnugo scores the game 23:2 since only one white territory is
> > entirely surrounded by white. it really should figure that white's
> > effectively captured the black stones on its side.
>
> GNU Go has no difficulty scoring this correctly, e.g.:
>
> Sulitelma 411% gnugo -l scoring.sgf --score finish
> 49 white (O) move PASS
> 50 black (X) move PASS
> Black wins by 2.0 points
>
> In fact I'm rather curious under what circumstances it could fail.
>
> > (;GM[1]FF[4]AP[glGo:1.3.1]ST[1]SZ[9]KM[0.0]
> > PW[Human]PB[Igo for windows]
>
> Uh, isn't "Igo for windows" the 9x9 free download version of Many
> Faces? Assuming that this is just a mistake, could it be that glGo is
> scoring the game without consulting GNU Go (and maybe is expecting the
> user to mark dead stones)?
>
> /Gunnar
>
>
> _______________________________________________
> gnugo-devel mailing list
> address@hidden
> http://lists.gnu.org/mailman/listinfo/gnugo-devel