Search All of the Math Forum:
Views expressed in these public forums are not endorsed by
NCTM or The Math Forum.



Re: Fitting multiple probability distributions
Posted:
Jun 16, 1996 6:19 AM


On Tue, 11 Jun 1996, Sean Hill <shill@iphysiol.unil.ch> wrote > >Does anyone have a good algorithm for fitting gaussians to multiple partially >overlapping distributions? > >I feel like there must be some way to do this well, but I can't seem to find >any!
It depends how much information you have available, and how you want to define the "fit".
If you have N variables X_i from your overlapping distributions which you are fitting with N Gaussians Z_i, the the Gaussians are completely determined by their means and their covariance matrix E[Z_i Z_j]. One approach is therefore just to calculate the mean and covariance of your X_i and use this to define the Z_i.
Other approaches would define a metric which measures "goodness" of fit in a way relevant to your problem, and then minimizes over the parameters of the Gaussian. Depends on what you need really.
Cheeers,
John.
 John Lunt, London. john@roachend.demon.co.uk "The Universe is not only stranger than you imagine, ... it's stranger than you can imagine."



