Using GMM for input points to find clusters, mean, and covariance

I have this data (200 x and y values) in a file that I can read in and plot. I am following https://www.mathworks.com/help/stats/clustering-using-gaussian-mixture-models.html in order to try and determine the mean and covariance of the data. I have run this code. But, I am really lost as to how to use GMM and then determine how many clusters I may need for my data. And then after that, determine the mean and covariance of each of the clusters. It seems like it's in: https://www.mathworks.com/help/stats/clustering-using-gaussian-mixture-models.html. Can anyone offer guidance as to how I can use this code to get to my number of clusters, and the mean and covariance of the clusters?

Answer 1

[Raunak Gupta]

Hi,

The number of clusters is the only parameter you need to give to fitgmdist and mean and covariance matrix will be calculated based on the data that is being processed. The documentation you mentioned does slightly advance analysis around the covariance matrix being full or diagonalized. If you need analysis on you fitting, then I think following the documentation step-by-step will help.

Otherwise, if you only want to do the clustering then these two documentations will be quick to understand.

Cluster Gaussian Mixture Data Using Hard Clustering

Cluster Gaussian Mixture Data Using Soft Clustering

For selecting the number of clusters, you can plot the data to see how many discrete clusters visually you can depict from the data. You can change your initial estimate by +-1 to see which one gives least log-likelihood.

Hope this helps!

[Sunshine]

Thanks for your help. Just now seeing your response. So, I didn't get a chance to try it out. I ended up going with a Python implementation.