movMF: An R Package for Fitting Mixtures ofudvon Mises-Fisher Distributions

摘要

Finite mixtures of von Mises-Fisher distributions allow to apply model-based clustering methods to data which is of standardized length, i.e., all data points lie on the unit sphere. The R package movMF contains functionality to draw samples from finite mixtures of vonudMises-Fisher distributions and to fit these models using the expectation-maximization algorithmudfor maximum likelihood estimation. Special features are the possibility to useudsparse matrix representations for the input data, different variants of the expectationmaximizationudalgorithm, different methods for determining the concentration parametersudin the M-step and to impose constraints on the concentration parameters over the components.udIn this paper we describe the main fitting function of the package and illustrate itsudapplication. In addition we compare the clustering performance of finite mixtures ofudvon Mises-Fisher distributions to spherical k-means. We also discuss the resolution ofudseveral numerical issues which occur for estimating the concentration parameters and foruddetermining the normalizing constant of the von Mises-Fisher distribution. (authors' abstract)