Re: Advice on evaluation

[Martin Mueller <mueller@xxxxxxxxx>]

>
>I like Mark's idea of averaging the two values.  I think will give a more
>accurate evaluation.

I think there are two kinds of values: a qualitative group strength measure
and a territory count. For example Explorer has the following types of
group strengths:

enum TTotal
{
	dead, notAlive, noTotal, almostDead, threatened,
	neutral, nearby, potential, almostSafe, safe
};

 I think Many Faces uses a similar but more elaborate scheme. I don't see
any way of 'averaging' such qualitative measures. For counting I evaluate
dead, notAlive, almostDead as -1 per empty point, nearby as 0.2, potential
as 0.5, almostSafe and safe as +1, the rest as 0.

Territory counts are easier. From the standpoint of combinatorial games,
averaging territorial counts is the most reasonable first-order
approximation. Of course in real life things are often different. Saving a
group might give the opponent many forcing moves, so it reduces the real
value. Or it might lead to dead for another group, which would make the
move invalid. On the other hand, a very efficient capturing move might even
be sente because it threatens something else. An example is a capture which
threatens another capture.

	Martin