Hello, I have recently implemented an EM algorithm for fitting mixture models of Kent distibutions (also known as Fisher-Bingham distributions). These distributions are the spherical equivalent of nonisotropic Gaussian distributions, however, they are constrained to 3D input data -- higher-dimensional input makes moment calculating intractable. (As a sidebar, the related von Mises-Fisher distribution can handle higher-dimensional data but is restricted to *isotropic* distributions.)
I have added the code to my scikit-learn fork on github: https://github.com/shoefer/scikit-learn I have not tested the code in depth, therefore I did not send a pull request. Still, it might be useful for someone requiring this type of distribution. Let me know if you have any questions! Best, Sebastian -- http://www.sebastianhoefer.de/ ------------------------------------------------------------------------------ _______________________________________________ Scikit-learn-general mailing list Scikit-learn-general@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/scikit-learn-general
