Question about GMM

Hello.

I'm a student of Electronic Engineering. I have a question about GMM and gmdistribution in matlab. I'm trying to fit my data with a gaussian mixture model using the gmdistribution(mu,sigma) object. My data have 13 dimensions, but my model has only 1 component. Is this right?

I think that I needed at least one component for dimension. The covariance matrix and mean vector seems well.

Thank you.

Answer 1

[Tom Lane]

Miguel, the idea is that the distribution may be a mixture of multiple components, each of which is a multivariate normal distribution. If you have just one component, you could compute the mean and covariance of your entire dataset. That defines a single multivariate distribution. You would not need to bother with a mixture distribution.