How to fit multiple gaussians over a 3-D histogram

I have been trying to fit multiple gaussians over a 3-D histogram (Iris data feature 3 and feature 4) using custom equation in cftool (inbuilt in matlab). But results are not coming good. The gaussians are getting clustered nearby nearby the first gaussian (near the origin). But if I try doing this for only one gaussian (data points are similar to gaussian) then I am getting good results.

I am using custom equation method and writing the equation of family of gaussians. I don't want to use fitgmdist as it gives same results always as we pass the full data but I want to pass data using histogram.

Answer 1

[Bjorn Gustavsson]

I think this should get the job done:

[nHist,xHist] = hist(Data,extra,parameters);
twoGausianFCN = @(pars,x) pars(1)*exp(-(x-pars(2)).^2/pars(3)^2) + ...
                          pars(4)*exp(-(x-pars(5)).^2/pars(6)^2);
err_fcn = @(pars,fcn,hN,hX) sum((hN - fcn(pars,hX).^2);
par0 = [pick,some,reasonable,start,guess,parameters];
par = fminsearch(@(pars) err_fcn(pars,twoGausianFCN,nHist,xHist),par0);

Obviously you can write a separate function for arbitrarily many Gaussians if you want to have something more flexible, and maybe the efficiency of anonymous functions are still worse than regular functions stored in files, but this should be enough if you need to fit a few histograms...

HTH

[JOO WON,HEE]

I didn't get this can you please elaborate. Thank You