Re: [gnugo-devel] new testcase

[Evan Berggren Daniel]

On Sat, 26 Oct 2002 address@hidden wrote:

>
> Evan wrote:
>
> > Also, GNU Go doesn't respond reasonably to a variation of the avalanche
> > joseki. I think these are the correct choices.
>
> > loadsgf games/nngs/gnugo-3.3.10-joshj-200210250406.sgf 12
> > gg_genmove white
> > #? [R3|P2]
>
> W tenuki'd after B Q6. The tenuki is playable but this is not
> really a joseki position. P2 looks dangerous, maybe playable
> by a strong player, but with B Q6 on the board I doubt it. Q2
> is playable.  W should not tenuki a second time, however. So
> this test might be OK with answer R3|Q2.

I agree.  I think this is symptomatic of a real problem, in that GNU Go
has no real concept of why josekis are good moves, and so it can't
extrapolate to near-joseki positions.  IMHO this is not fixable right now.
I think adding the positions to the database of josekis is best.

Anyway, the new test is

loadsgf games/nngs/gnugo-3.3.10-joshj-200210250406.sgf 12
gg_genmove white
#? [R3|Q2]

Thanks

Evan Daniel