0

I know Gaussian Mixture Models are used for distribution of points and thus the functions they represent are positive (PDFs are postivie). However I want to approximate a function using GMMs which could also be negative. Is there any way to approximate negative values using GMMs? here is an example:

a function and its approximation with GMM

as it's obvious in the plot, using a mixture of two Gaussians it's much better to cover have one positive and one negative Gaussian as opposed to what matlab fit function finds. Is there any way in MATLAB to accomplish this?

Ameer Jewdaki
  • 1,758
  • 4
  • 21
  • 36
  • @Suever I don't think it's mathematically the same ... for example in the picture attached maybe after taking the `abs` it will fit them with a single Gaussian. – Ameer Jewdaki Jun 03 '16 at 15:44
  • Have you considered applying standard fitting techniques to both the negative and positive part of the signal separately? – mikkola Jun 03 '16 at 16:28
  • @mikkola but then again that might complicate things, say if I want four Gaussian approximation how many positive and how many negative should I consider? one way is to run every combination and choose the combination that gives the best fit. But I thought there might be a more convenient strategy. – Ameer Jewdaki Jun 04 '16 at 14:25

0 Answers0