Re: [gnugo-devel] dragon state transitions

[Arend Bayer]

On Sat, 13 Jul 2002, Evan Berggren Daniel wrote:

> OK, I'm currently planning to modify twogtp to work as follows:
>
> take an option for number of games to play.
> play the two programs against each other for that number of games.
> for each game, at every move check the state of all dragons.
> if an illegal transition occurs, output a message and save the game
> as an sgf file.
>
> Anything else I should do with it?

Maybe just one more comment. I think we are mostly interested in finding
situations where the owl goes wrong in a very obvious way. I don't
think there are very many of those, but they do exist and are what we
should fix most urgently. (Sometimes they might just be due to a missing
pattern in the owl pattern database.)

I worry that these might not be noticed in GNU Go vs GNU Go games, as then
both side might miss this obvious move. Hence replaying games that GNU
Go has played against humans (you can find a lot of them on Gunnar's web
site collecting the games on NNGS -- this is linked from the development
web page; also there are quite a few games played at computer
tournaments in various subdirectories of the regression directory) might
be more useful.

But I guess once you wrote the code to look for the dragon status
changes, it should be easy to modify it in either way.


Arend