Andy <t3k...@gmail.com> wrote:
> If you want different sized clusters, you might want to look into GMMs,
> which learn a covariance structure.
The mixmod software (mixmod.org) implements CEM with various
parameterizations of the mixture model. It can be used as a spawned
subprocess or as a C++ library (e.g. call with Cython). Unlike fitting GMMs
with EM, CEM is a direct generalization of k-means. The main advantage of
CEM over EM is a faster rate of convergence and a cheaper E-step. It is
comparable to the difference between k-means ("CEM") and fuzzy k-means
("EM"). The main disadvantage is higher susceptibility to local optima.
Sturla
------------------------------------------------------------------------------
_______________________________________________
Scikit-learn-general mailing list
Scikit-learn-general@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/scikit-learn-general
