Re: [gnugo-devel] Owl tuning

[Gunnar Farneback]

Nando wrote:
> In case it's considered mandatory that each of these PASSes is to be
> checked carefully, then I'll do the job, but it will take me some
> time, because I'd like to work on something else for a while.

No, we don't have any set rules that all regression deltas must be
investigated. Generally speaking it depends on the nature of the patch
what level of evaluation is expected. An obvious bugfix may go in
regardless of what it does to the regressions. A controversial or very
speculative change requires a lot of proof that it really is good. In
most cases it's a good idea to verify that at least some test case is
properly solved and that the failures are not caused by bugs or
oversights in the patch.

In this case I wouldn't say it would be well spent time to examine all
those PASSes as long as you know that some of them pass thanks to your
tuning. 

> +Pattern A1017
> +# nn New pattern (3.3.11)
> +
> +X.*o     threaten to cut and mess up opponent's shape
> +xX.X
> +
> +:8,-,value(65)
> +
> +X.*o
> +xbac
> +
> +;!oplay_attack(*,a,*) && oplay_disconnect(*,?,a,b,c)

This constraint looks strange. The second half gives O three
consecutive moves which is usually not relevant. You don't mean
!oplay_connect(*,?,a,b,c) ?

> +#   FIXME: constraint is ugly
> +
> [...]
> +
> +abcde
> +....f
> +....g
> +.X..h
> +*...i
> +
> +;(owl_escape_value(a)>0) + (owl_escape_value(b)>0)
> +;+(owl_escape_value(c)>0) + (owl_escape_value(d)>0)
> +;+(owl_escape_value(d)>0) + (owl_escape_value(e)>0)
> +;+(owl_escape_value(f)>0) + (owl_escape_value(g)>0)
> +;+(owl_escape_value(h)>0) + (owl_escape_value(i)>0) < 2

I don't think you planned to count d twice.

> +Pattern A1130
> +# nn New pattern (3.3.11)
> +
> +...o?     to be considered in crosscut cases like this
> +...Ox
> +.*.OX
> +..XXO
> +...o?
> +
> +:8,-,value(65)

Shouldn't this be constrained on a specific goal string or the O
stones being short on liberties?

/Gunnar