Re: [gnugo-devel] Weakness calculation

[Gunnar Farneback]

Arend wrote:
> The genus is far too unreliable.

Definitely.

> Just try setting the curve to values as you suggest, and I bet you
> will break lots of regression tests. One idea to improve the genus
> accuracy would be to take into account matching vital owl attack
> patterns.

And/or improved heuristics to award more than two eyes for an
eyespace.

> Another idea would be to some very quick reading for invincibility
> (i.e. minimax reading where the goal is to get a dragon into an
> invincible shape). Both the groups you mention should be seen invincible
> within 5-10 nodes.

This sounds somewhat related to my idea of doing "local life" reading,
but should be simpler to implement. Good idea.

/Gunnar