Re: [gnugo-devel] trevor_1_18.2

[Gunnar Farneback]

Trevor wrote:
> I'm not sure the approach I took in devaluing ATTACK & DEFEND
> moves is correct, but here it is for comment.

The relevant part of the patch looks like this:
> @@ -3141,6 +3141,12 @@
>             && !move[pos].move_safety)
>           break;
>         
> +     /* No bonus if the attack is on a very dead group. 
> +        tm added (3.1.18) see trevorb:920*/
> +     if (dragon[aa].owl_status == DEAD &&
> +         dragon[aa].matcher_status == DEAD)
> +       break;
> +
>         /* FIXME: This is totally ad hoc, just guessing the value of
>           *        potential cutting points.
>          * FIXME: When worm[aa].cutstone2 == 1 we should probably add

I don't think it's safe to remove this entirely. Unfortunately
owl_status isn't quite that reliable, especially in semeai type
positions, so this part of the valuation can save us from unnecessary
trouble. But we could certainly improve the heuristics, e.g. so that
this bonus only is awarded under circumstances related to
* whether we are ahead
* whether one or more dragons surround the stones under consideration
* the safety level of the surrounding dragon(s)

/Gunnar