Re: [gnugo-devel] helping with joseki database regression tests

[Gunnar Farnebäck]

On 01/12/11 05:15, andrewdo wrote:
> Hi,
>
> To which joseki database do you refer in your call for help. I don't
> know much about gnugo, but would be willing to at least give this a shot.

GNU Go has a small joseki database built in, generated from the files
patterns/komoku.sgf and six other sgf files in the same directory. The
moves in the files are marked in a few different ways meaning things
like normal joseki move, urgent joseki move, followup move, and
non-joseki move. Due to the way these are handled in the move
generation it may occasionally happen that other moves than the joseki
moves are preferred.

The idea of that TODO item is to play through the variations present
in the joseki sgf files, preferrably with a few different
configurations of stones in the remaining corners, and verify that
joseki moves really are generated (at least not other local moves) and
non-joseki moves are not generated.

Presumably this can be done by hand but that would be rather painful
and not easily repeatable so a scripted solution is necessary. Ideally
it would transform a joseki sgf file into a regression test file (see
e.g. regression/seki.tst for an example) which can then be run in the
same way as all other test cases.

/Gunnar