How to use MAP estimate instead of Maximum Likelihood Estimate while modelling Gaussian Mixture Model for a data set? Please read description.

9 views (last 30 days)
I want to fit GMMs to n data sets X(1), X(2), ....X(n) one-by-one. Each X(i); i ∈ (1, n), has 100 points which are plotted to be modelled by GMMs. I know that there are only two Gaussian components in each mixture. I am using fitgmdist to fit GMM to each X(i) separately. When fitgmdist reaches convergence for X(i), it returns mixing ratios of the two Gaussian components. I want to use this mixing ratio as the initial mixing ratio for the next data i.e. X(i+1). Since this is like using prior knowledge for fitting a GMM to X(i+1), MAP should be used instead of MLE. fitgmdist uses MLE by default. Is there a way to switch to MAP in the fitgmdist algorithm? Or use previous mixing ratios obtained for X(i) as initial parameter for X(i+1) while sticking with MLE? Kindly answer in detail if possible. I am new to MATLAB.

Answers (1)

Shruti Shivaramakrishnan
Shruti Shivaramakrishnan on 1 Sep 2016
Unfortunately, MATLAB currently does not have a built-in function for the MAP estimate calculation while modelling Gaussian Mixture Model for a data set. I work for MathWorks and have forwarded this feedback to the appropriate product team.

Community Treasure Hunt

Find the treasures in MATLAB Central and discover how the community can help you!

Start Hunting!

Translated by