Re: Advice on evaluation

[Christian Bauer <Christian.Bauer@xxxxxx>]

I find this a very intersting discussion.

Sir Sedrik and I thought also about having more than one (integer or real)
number for the value of the board position or of a group.
We thought, that a whole vector of numbers,  instead of a scalar could be a good
idea and usefull for several things.

We thought especially about the following vector:

BoardValueVector ::= <BoardValue, Probability>

(and of cause, the same for groups:   GroupValueVector ::= <GroupValue,
Probability>  )

where BoardValue is the result of the evaluation function (calculated directly
or using some search) and the Probability says, how certain or uncertain the
BaordValue is calculated.

Example:
If a big group in a position is unconditionally dead or alive, the probablity
for the calculated value for this position will be high.
If the position is in a very unstable state (maybe you can't search far enough,
to get a stable position), the probability will be low.

The questions at this point are:
a) how to calculate this probablity exactly?
b) how to use it?

----------
But there is another problem for every BoardValue which is not a scalar:

If you use a vector instead of a scalar, you loose the strictly ordering
relationship in your data. That means, the values or not (or not so easy)
compareable then. The standard minimax algorithm for example, works only with
scalar BoardValues.

Until now, I found no solution for this probem.

Ciao
    CeeBee

PS: special greetings to Sir Sedrik