I am new to using GMMs. I was not able to find any appropriate help online. Could anyone please provide me right resource on "How to decide if using GMM fits to my problem?" or in case of classification problems "How to decide if I have to use SVM classification or GMM classification?"

  • $\begingroup$ what is ur data set and what is your exact problem? It's used when data follows ( is a mixture of) more than 1 normal distribution. See another question -stats.stackexchange.com/questions/236295/… $\endgroup$ – Arpit Sisodia Feb 5 '17 at 18:08
  • $\begingroup$ You can think of it as a form of clustering where you don't have labeled data & believe the latent groupings are perfectly multivariate normal. $\endgroup$ – gung Feb 5 '17 at 18:12
  • $\begingroup$ @arpit-sisodia, We are working on feasibility of a hardware keyboard setup which seem to have specific features and we are planning to model it using GMM. We dont know the underlying process clearly and hence we are trying to model using machine learning methods. So, we are not sure if there is actually a mixture of gaussians in the underlying process. Moreover, it is multi-dimensional and we cannot visualize it to see if it is mixture of gaussians $\endgroup$ – Vinay Feb 6 '17 at 5:20
  • $\begingroup$ @arpit-sisodia, Link you have provided suggests more of trial and error method to see if GMM fits to my data. Is there a conclusive way/Thumb rule to decide up on the models to use. Trial-and-error method of playing with more mixtures can fit my data. But is there a certain way of deciding up on? Like we need to have Linear separability of data for SVM classification $\endgroup$ – Vinay Feb 6 '17 at 5:20

In my opinion, you can perform GMM when you know that the data points are mixtures of a gaussian distribution. Basically forming clusters with different mean and standard deviation. There's a nice diagram on scikit-learn website. L

GMM classification

An approach is to find the clusters using soft clustering methods and then see if they are gaussian. If they are then you can apply a GMM model which represents the whole dataset.

  • 1
    $\begingroup$ often we don't know if data points are Mixture of Gaussians. So, this is more of play-around with Gaussian and MoG and see if it fits. But there are no directions/thumb rules to go about in using GMM right $\endgroup$ – Vinay Feb 6 '17 at 6:01
  • 2
    $\begingroup$ As per my experience you do need to find the pattern in data that its a Mixture model. A nice paper to read would be this: stat.cmu.edu/~cshalizi/uADA/12/lectures/ch20.pdf $\endgroup$ – Prerit Feb 6 '17 at 20:38

GMMs are usually a good place to start if your goal is to either (1) cluster observations, (2) specify a generative model, or (3) estimate densities. In fact, for clustering, GMMs are a superset of k-means.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.