Section: New Software and Platforms

MixtComp package for full mixed data

Participants : Vincent Kubicki, Christophe Biernacki, Serge Iovleff.

MixtComp (Mixture Computation) is another important software fot the Modal team since it concerns model-based clustering for mixed data. Main difference with the MIXMOD/Rmixmod software is that MixtComp's architecture is able to integrate easily and quickly all new univariate models, under the conditional independence assumption, that will be sequentially available from researches of the Modal team or others. Currently, central architecture of MixtComp is built and three models (Gaussian, multinomial, Poisson) are implemented with ability to natively manage missing data (completely or by interval). MixtComp stands both as a C++ library and an R package. The code is currently developed internally, and has been field-tested through two contracted partnerships.