Re: [gnugo-devel] Uncertain defense certainly bad

[SP LEE]

> At the end of this message is a game where an uncertain owl defense is
> certainly bad.
>
> At move 104, GNU Go 3.3.22 plays a terrible move at R11.
> Not only is this worth -1 point with gote but it actually hurts
> W's eyeshape. After this move, a full eye becomes a half eye.
>
> The move is contributed by the pattern:
>
> Pattern D123
> # tm New Pattern (3.1.17)
> # tm Modified (3.1.18)
>
> ??*??   expand eye shape, not in corner
> ?O.O?
> ?...?
> -----
>
> :|,-,value(65)
>
> One possibility here is to make this pattern more
> specific. Consider:
>
> ??*??   expand eye shape, not in corner
> XO.OX
> ?...?
> -----
>
> Assuming the X stones are not dead, the move at * does
> not contribute a half eye on the edge. The pattern
> could be split as follows:
>
> Pattern D123a
>
> ??*??   expand eye shape, not in corner
> oO.O?
> ?...?
> -----
>
> :8,-,value(65)
>
> Pattern D123b
>
> ??*??   expand eye shape, not in corner
> XO.O?
> ?...?
> -----
>
> :8,-,value(65)
>
> ??*??   expand eye shape, not in corner
> AO.O?
> ?...?
> -----
>
> ; attack(A)
>
> This helps a little in the present situation. W still makes a small
> defensive move but at least it is not a harmful one.
>
> Dan

I have removed this pattern and have run all the regression tests. The 
result:

./regress.sh . owl.tst
4 unexpected FAIL: Correct '1 C12', got '1 B14'
80 unexpected PASS!
./regress.sh . neurogo.tst
13 unexpected FAIL: Correct '!P7', got 'P7'
./regress.sh . lazarus.tst
16 unexpected PASS!
./regress.sh . arend.tst
4 unexpected FAIL: Correct 'R14|S15|R17', got 'S16'
./regress.sh . nngs3.tst
390 unexpected PASS!
1000 unexpected PASS!
./regress.sh . nngs4.tst
470 unexpected PASS!
./regress.sh . arend2.tst
50 unexpected FAIL: Correct 'S14', got 'T13'

For owl:4, B14 seems also to be a good move (I even think it's better than 
C12, which is too passive), because if B answers with C12, W can play C15 
and then B has to defend the corner and W can live easily. A possible 
sequence is W:B14, B:C12, W:C15, B:D15, W:C16.

For neurogo:13, I don't think this pattern can solve the real problem, i.e. 
don't start a completely nonsense ko.

For arend:4, S16 seems to be a reasonable move too, at least it's not 
tenuki.

For arend2:50 I have to take more time to check it.

SP Lee

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