Clustering

How many clusters should be chosen to represent the data?

The demo uses an EM based based to fit Gaussians with full covariance matrices to the data. It begins with one Gaussian, and once the maximum likelihood solution is found, one of the clusters is split into two, each placed one standard deviation along the principal axis of the distribution. The final solution (choice of number of clusters) is made when adding in a extra cluster doesn't increase the likelihood.

Compare the membership of the points when classified

	Using the EM-Gaussian fitting.
	Using standard K-means.