1. pick an ordering of the nodes (any ordering will work, but picking a 
"good" ordering will reduce the size of the resulting Bayesian network 
-- maximum cardinality search is one heuristic, there are many others; 
picking the best ordering in NP-hard.)

2. Direct all existing edges from smaller numbered nodes to larger 
numbered nodes.

3. For each node i, in reverse order from n, n-1, ... 5,4,3: Fully 
connect all parents of i (directing edges from smaller numbered nodes to 
larger). Note some parents of i might have been formed during previous 
iterations.

4. Compute the required conditional probability tables by querying the 
Markov network.

I believe you can find more info in the Neapolitan book and the Pearl book:
http://www.amazon.com/exec/obidos/tg/detail/-/0471618403/
http://www.amazon.com/exec/obidos/tg/detail/-/1558604790

Hope this helps.

Dave

Michael Shigeki Onishi wrote:

> Does anyone have resources on how to transform a Markov Network into a
> Bayesian Network?
> 
> Thanks,
> 
> Michael Shigeki Onishi
> ------------------------------------------------
> Sun Certified Java Developer
> Sun Certified Java Programmer
> 

Reply via email to