Re: [gnugo-devel] success!

[Evan Berggren Daniel]

On Thu, 1 Aug 2002, Gunnar Farneback wrote:

> Evan wrote:
> > OK, I've got a game with two instances of illegal state changes (ALIVE ->
> > DEAD and DEAD -> ALIVE).  sgf file follows, at moves 135 (group P11 goes
> > from alive to dead) and 140 (group R16 goes from dead to alive).
> >
> > The P11 group looks like it may be an artifact of dragons combining and
> > being cut, but the R16 one looks very real.
>
> Those are both real ones.
>
> The next step when these mistakes are found is to make regression
> tests of them. Those would best be owl tests, like
>
> loadsgf xxx.sgf 139
> 1 owl_defend R16
> #? [1 (O11|T15|T13)]
>
> Unfortunately this requires some manual work and ability to read out
> the position (but it's good exercise to find out which of the
> inconsistent owl results is wrong). An alternative would be to have
> some function to just check the consistency, similar to
>
> loadsgf xxx.sgf 139
> 1 owl_consistency R16 W O11
> #? [1]
>
> where the imagined function owl_consistency would evaluate the owl
> status of R16 before and after the white move at O11 and check for
> invalid changes. Such regression tests should in principle be possible
> to generate automatically.


OK, I'll go ahead and make a regression test from that, and also from some
games I ran last night.  I'm not very confident in my reading ability, but
I'll give it a try and send things this way in a few days.

I'll probably write the owl_consistency function next, and then make the
matcher_check.pl script auto-generate regression tests.

Thanks again

Evan Daniel