RE: owl improvements (Re: [gnugo-devel] Owl tuning)

[Arend Bayer]

Nando wrote:
> Arend wrote:
> > 2b) Bettter analyis of escape value.
>
> In order for me to decide whether to stop on this work or not, I'd like to
> know what you plan to do with 2b. If my approach is clearly inferior, then
> no need for me to waste too much time with it.

Sorry, I still haven't answered this question yet. What I'd suggest to
do is to seriously refine the input from the influence module.

1. The territory valuation (of the escape influence) seems an obvious
candidate instead to use instead of the core whose_moyo/area etc.
distinction.

2. It would seem worthwile to resurrect the idea to ignore influence
from the dragon that is trying to escape. (One would need to enhance the
caching to do this, but for a prototype it is of course fine to just
disable the caching.)
Also, we need to be more careful about escaping towards other dragons
that might be dead themselves, too.

3. I'd hope that the algorithm in dragon_escape can also be refined.
a) One idea would be to misuse Gunnar's compute_connection_distances()
and collect escape influence from, say, all intersection of distance
less than 3.0. This may or may not be a good idea.
b) After spreading outwards to collect these escape values, don't just
sum them up, but instead propagate them backwards to the liberties of
the dragon. This would
- tell in which direction it might be worth escaping
- allow to ignore, say, the liberty with the highest escape value.
  (Rationale: If we have only one direction to escape, this does not
   add anything to our safety.) Here I am pretty sure that this will
   help something.

Anyway, this are all just ideas, and I don't know when I'll have time
to start trying them out. So feel free to try your own ideas or to steal
some of mine.

Even if you do completely different things this will set the bar higher
before I start, which can only be good...

Arend