[SciPy-user] Draw a density line on an histogram

Anne Archibald peridot.faceted@gmail....
Mon Jan 28 11:58:02 CST 2008

On 28/01/2008, Charles Vejnar <charles.vejnar@isb-sib.ch> wrote:
> Thank you.
> Ok, with gaussian_kde then linspace (it's a bit different than in R).
> Is it possible to have a different kernel than "gaussian" ?

Just to be clear: the kernel is Gaussian, but the distribution that
you are fitting need not be. Try, for example, squaring all the random
values in Robert Kern's example.

Taking a Gaussian kernel is  in some maximum-entropy sense the most
conservative possible choice, assuming as little as possible about
your distribution. For periodic distributions I have implemented a von
Mises kernel (the periodic analogue of a Gaussian, again in a
maximum-entropy sense).


More information about the SciPy-user mailing list