Re: [Bug-gnubg] positive error rate bug

[Joern Thyssen]

On Thu, Sep 11, 2003 at 07:02:26PM +0200, address@hidden wrote
> On Thu, 11 Sep 2003, Joern Thyssen wrote:
> 
> > On Wed, Sep 10, 2003 at 11:10:38PM +0200, address@hidden wrote
> > > This is version 1.1357 WinXP.
> > >
> > > Import bug2.mat. Analyse 2-ply. Look at stats: OK. Save game/session
> > > as bug2.sgf. Open bug2.sgf. Look at stats: one error rate is positive.
> > There is an illegal move in the file:
> >
> >  42) 43: 6/3 6/2                 64: 25/19 20/16
> >  43) 65: 4/0 4/0                 65: 4/0 4/0
> >                                  ^^^^^^^^^^^^^^^^^
> >                                  illegal move
> >      Wins 2 points
> >
> > You'll also see a warning when you import the file.
> >
> > The problem was that gnubg added the illegal move to the move list and
> > this confused it when it loaded the SGF file.
> 
> I hope you agree it's still a bug.
> 
> What is needed  (at least if you want to  do unsupervised batch analysis
> of matches)  is a format checking  module on importing  matches.  If the
> format is incorrect (and this happens a LOT when dealing with real-world
> generated .mat files) the result should not be undefined or a crash, but
> a failure to load plus the setting  of a status flag which can be polled
> from a script to detect load status.

The best I can offer without major re-programming is that gnubg ignores
invalid moves or stops reading once an illegal move is encountered.

Jørn

pgpuPZTHVvMfc.pgp
Description: PGP signature