Author Topic: N< p in EM learning  (Read 8606 times)

Offline veradj

  • Newbie
  • *
  • Posts: 1
    • View Profile
N< p in EM learning
« on: April 18, 2013, 18:05:28 »
I am working with a bayesian network with continuous variables and I am trying to estimate parameters with the EM algorithm. I have a problem with the EM wizard, and it seems due to the fact that the size of the largest clique is 11 while the dataset has only 10 cases. So my question is: does HUGIN support EM  learning in the case "N<p"?

Thank you!

Offline Anders L Madsen

  • HUGIN Expert
  • Hero Member
  • *****
  • Posts: 2288
    • View Profile
Re: N< p in EM learning
« Reply #1 on: April 27, 2013, 14:44:33 »

The EM learning algorithm in HUGIN does not consider whether or not the relationship between the number of cases and number of parameters is appropriate.

Is 11 the number of variables in the clique or the state space size of the variable(s) in the clique? and, what do the variables p and N represent in your example?