Date: Feb 22, 2008 3:01 PM
Subject: Expectation Maximization question
I have written an EM algorithm for mixtures of gaussian distributions. However, my problem is that I would now like to "weight" the EM so that my output result is scaled according to some metric that I have. By metric, I actually mean a column of numbers. Would it be right to multiply those values with the posterior probabilities calculated in the E-step of the EM? Or multiply those values with the probability density calculated in the E-step again?
Thanks in advance