Re: [gnugo-devel] trevor_3_12.6 (tuning)

[Gunnar Farneback]

Trevor wrote:
> A slightly different version of this patch had the following
> results, slightly positive (9 FAIL, 15 PASS) (one FAIL was
> trevor:1100 a new problem). arend:25 looks to need some other valid
> moves. F7 is fine. I did not count owl node increases.

A few of those results are already in CVS, including owl1:268,
auto04:11, and auto_handtalk:7, which all have had their correct
results revised.

> +Pattern A516
> +# tm New Pattern (3.3.12) see owl1:286
> +
> +
> +xXY*
> +xOO.
> +----
> +
> +:8,s,value(50)
> +
> +xXY*
> +xOBa
> +----
> +
> +; lib(B)>1 || !oplay_attack(*,a,B,*)

This pattern looks slightly bogus. The liberty constraint makes little
sense since the only unknown liberty is the lower left x in the
pattern; better not to have a wildcard there. Besides, as written it
doesn't match the referenced position and the reading constraint tries
to play on top of an existing stone.

/Gunnar