Re: [gnugo-devel] Hoshi Josekis

[Paul Pogonyshev]

Gunnar wrote:
> Evan wrote:
> > Do we want to start make regression tests for josekis? I think
> > creating a new file joseki.tst and collecting tests about correct
> > joseki play and choices would be a good idea.
>
> It should be possible to automatically generate joseki test cases from
> the joseki database files. An interesting experiment would be to run
> this with the --nojosekidb option to see how many joseki moves the
> engine comes up with without the joseki database.

while we are on joseki question. i started running a twogtp match between
gnugo 3.3.17-pre-7 (white) and gnugo-3.2 (black) with handicap 2 and komi
5.5. so far white is well ahead, but it's only several games that have
been finished.

while looking through the games, i noticed that gnugo plays this joseki
far too often:

|............
|...X........
|.....X......
|...X.XO.....
|...XOO...O..
|............
|............
-------------

besides, on nngs i saw many times that gnugo (as black) places the last
(6-4) stone even when white doesn't drop 3-10. this is often very slow.
maybe someone who knows josekis well can do something about it? e.g. if
white tries to approach from top without having 3-10 in place, certainly
there must be a way to punish her even without 6-4.

Paul