mvnpdf / calculate Probability Density for a Matrix / EM algorithm

Hello MATLAB Community,

You are my last resort! I need to apply EM clustering to a couple of thousands of matrices (m rows, n columns). The corresponding k-means works well.

However, I have trouble figuring out the Gaussian probability density function for entities that are matrices.

mvnpdf(X,Mu,Sigma) is supposed to receive a matrix for X. But will it be consistent with what I need?

Thanks in advance!

Answer 1

[Sam Roberts]

In R2007b, Statistics Toolbox introduced the gmdistribution class, which represents Gaussian mixture distributions. gmm = gmdistribution.fit(X,k) uses an EM algorithm to fit a Gaussian mixture model with k components to the data in matrix X. The data can then be partitioned into clusters using idx = cluster(gmm,X).

Introduction to Gaussian mixture models:

http://www.mathworks.com/help/toolbox/stats/bq_679x-24.html

gmdistribution.fit

http://www.mathworks.com/help/toolbox/stats/gmdistribution.fit.html