Section: New Results

MixtComp software for full mixed data

Participants : Christophe Biernacki, Vincent Kubicki.

MixtComp (Mixture Computation) is an integration software from the MODAL team for model-based clustering of mixed data. Its computing core is written in C++ and is accessed through an R interface. Its architecture allows to easily and quickly integrate new univariate models (under the conditional independence assumption) as they are published. The first phase of development was the implementation of three basic models (Gaussian, Multinomial, Poisson) with the native management of partially observed data (including intervals). It now implements models related to ordinal data (2015), rank data (2015) and functional data (2016), still with missing or partially missing data. The code is developed internally, and has been field-tested through several contracted partnerships (see the section about contracts). It is now referenced in the BIL database and the APP. It is available through a new web interface, called MASSICCC at https://massiccc.lille.inria.fr/#/ (see also the dedicated section). MixtComp has been presented to an invited talk in October 2016 at the Academy of Sciences in Tunisia [26].