Re: [gnugo-devel] some tests (new patch)

[SP LEE]

> > At the end of the message, is a sgf file from 3.3.16 lv 10 playing 
> against
> > itself at
> > lv 12 (game 35 : lv 10 won by 133.5)
> > after 187 = G12 , all gnugo versions have trouble to analyse the
> > the M16 group even when they are asked precise questions
> > eg = 1 L19(*owl_defend  M17*)
> > which is wrong !
> > The problem is fairly easy
> > i have already see Gnugo solving much more complex problems.
> > md.
> >
> >
>
> I don't think this problem is easy enough for gnugo.  This would be in the 
> category of "big eye with a cutting point on the edge". There will be many 
> variations inside the eye which gnugo doesn't consider much. I have also 
> see gnugo had similar problem in one of my game posted earlier.
>
> Of course, when I tried "owl_does_defend N19 M18", the answer was yes, too.
>
> SP Lee

Yet this problem is easier than I thought. I added two patches. To my 
suprising, there are many changes in the regression tests, 8 passes and 4 
fails.

./regress.sh . owl.tst
150 unexpected PASS!
./regress.sh . strategy.tst
36 unexpected PASS!
./regress.sh . viking.tst
2 unexpected PASS!
./regress.sh . nicklas1.tst
203 unexpected PASS!
./regress.sh . buzco.tst
5 unexpected FAIL: Correct 'C10|C11|E9|A4|F10', got 'A9'
./regress.sh . nngs.tst
220 unexpected FAIL: Correct 'L1|E7|G13|K9', got 'R6'
./regress.sh . trevorc.tst
240 unexpected FAIL: Correct 'K2', got 'K3'
./regress.sh . strategy3.tst
141 unexpected PASS!
./regress.sh . global.tst
44 unexpected PASS!
./regress.sh . nngs3.tst
230 unexpected FAIL: Correct '!L12', got 'L12'
./regress.sh . ninestones.tst
630 unexpected PASS!
./regress.sh . manyfaces1.tst
30 unexpected PASS!

After brief glance at the fails, I found that the A9 in buzco 5 seems to be 
a reasonable answer too, nngs 220 is certainly wrong, nngs3 230 needs 
further checking and trevorc 240 is something I like to discuss. I'm not 
sure what the good answer K2 means, but I thought even after K2 the black 
might live (in Ko) in the corner. Before the patch gnugo thought black was 
dead after K2. Now I added the defend pattern, gnugo thought the corner is 
alive after K2. I checked many variations. Because black has many sente 
moves and the white is not yet very thick, black should have chance to live.

SP Lee

_________________________________________________________________
The new MSN 8: advanced junk mail protection and 2 months FREE*  
http://join.msn.com/?page=features/junkmail

NewPatch
Description: Text document