Re: [gnugo-devel] tuning

[Gunnar Farneback]

Evan wrote:
> In strategy2:77, the stone at H17 is now considered critical, and gnugo
> plays H15 as black to kill it.  This is probably an improvement over the
> current reading of the stone as unconditionally alive.

I think the essence of that test would be captured by
restricted_genmove black N4 H3 G3 J2 H2 G2
(N4 is the game move, G3, J2, H2, and G2 are inferior connections.)

> +Pattern A1022
> +# evand New pattern. (3.3.18)
> +
> +oX.O
> +..*.
> +.X.X
> +...o

Shouldn't there be a Y in this pattern?

> +
> +:8,-,value(55)
> +
> +oE.O
> +.a*.
> +.XbF
> +c.do
> +
> +;!oplay_attack(*,*) && !oplay_disconnect(*,a,b,d,E,F)
> +;&& oplay_disconnect(*,c,b,d,E,F)

This constraint is strange, in particular the middle part. Failure to
disconnect doesn't look like something we want. Also, wouldn't we want
to test the variation where O responds at b too? Another alternative
is to use !oplay_connect(*,E,F). Is this slower or less reliable?

Another thing is that you shouldn't need the !oplay_attack(*,*) test.
This is automatically done unless the pattern has an s classification.

+Pattern A1023
+# evand New pattern. (3.3.18)
+
+?Xx?
+O*.?
+?..?
+xYxx
+
+:8,-,value(45)
+
+?Bx?
+O*a?
+?..?
+xCxx
+
+;!oplay_attack(*,*) && !oplay_connect(*,a,B,C)

This also looks strange. Sure you don't mean
oplay_disconnect(*,a,B,C)? As written X is allowed to first play at a
and then has a second consecutive move to try to connect.

/Gunnar