From: eq2dephased@gmail.com   
      
   Hello,   
      
   I am currently working on a project which involves applying   
   probabilistic network techniques to intrusion detection systems.   
      
   The task consists in building a network from a complete set of   
   observations (about 20 discretized variables) and then use that network   
   to see the support it gives to future observations. There will be no   
   inferences at all in this network since I only need to compute the   
   support of an observation (the states are known for all the variables).   
   Also, there is no domain knowledge on the variables used in the   
   network: no given roots, no nodes ordering, etc.   
      
      
   I am currently trying to determine which type of network would be best   
   for my problem and which technique could be applied to learn this   
   network. I hope someone can help me with answering any of the two   
   questions:   
   - Should I use a cyclic or acyclic network? In the case of  acyclic   
   network, which algorithm should I investigate to learn the structure?   
   - Should I use an oriented or un-oriented network? In the case of   
   un-oriented network, how do I define the parameters of the nodes (since   
   there is no conditional probabilities) and how could I compute the   
   support for an observation?   
      
   Thanks a lot in advance for your help!   
      
   [ comp.ai is moderated ... your article may take a while to appear. ]   
      
   --- SoupGate-Win32 v1.05   
    * Origin: you cannot sedate... all the things you hate (1:229/2)