Hi Xiangdong,

I can think of at least two ways to interpret what you are trying to do,
here is the answer for all three interpretations:

(1) I want to generate a random JPD along with its perfect map D.
To do this, it is sufficient to construct a random dag and randomly set the
parameters so that no two columns in a given table are identical.  The dag
will be a perfect map to the JPD generated by the network.

(2) Given a JPD I want to construct its perfect map D.
As far as I know, the only way to do this is to query the JPD for
independence relations along the lines of a constraint-based learning
algorithm, for example the PC algorithm (given in the book "Causation,
Prediction and Search", Spirtes, Glymour and Scheines), or Pearl and Verma's
algorithm (http://citeseer.nj.nec.com/pearl91theory.html).

Hope this helps,
Denver.
----
Denver Dash       http://www.sis.pitt.edu/~ddash

----- Original Message -----
From: "Xiangdong An" <[EMAIL PROTECTED]>
To: <[EMAIL PROTECTED]>
Sent: Wednesday, September 26, 2001 10:02 AM
Subject: [UAI] JPD for a DAG such that DAG is p-map


> Hi everybody,
>
> I am wodering if I generate a probabilisty distribution JPD
> by generating a set of conditional probability distributions
> {P(Xi|II(Xi))} for a DAG D on a topological ordering X1,X2,...,Xn
> such that II(Xi) is the minimum set of predecessors satisfying
> P(Xi|II(Xi))=P(Xi|X1,...,Xi-1).
>
> Then the DAG D is the perfect map of the JPD? If not, there
> is any practical way to generate them?
>
> Xiangdong
>
>



Reply via email to