Density based clustering by Francesco
I uploaded a little tool for density based clustering on our group area. In the folder .../data/shared_documents/matlab/francesco Matlab Code Repository/ you will find the following m-files:
- EM_GM.m - EMclust.m
The first one fits a multivariate gaussian mixture model to an n-dimensional dataset. It gives in output the parameters of the gaussians (centers and covariance matrices), the weigths of the model (the prior probabilities of each gaussian) and other optional parameters. The learning method is based on the Expectation-maximization approach. I downloaded it from the web and I slightly optimized it. The second one assign each point of the dataset to one distribution using a simple Bayesian rule assuming the weigths of the mixture model as priors and evaluating the density functions to obtain likelihoods.
Each code plots the results and contains help lines and documentation.