The SYLVESTRA++ software program is dedicated to the modeling of dependencies within highly correlated (discrete) variables, in the high dimensional setting. The model inferred from the data is a forest of latent tree models (FLTM), that is a forest of tree-shaped Bayesian networks. SYLVESTRA++ relies on a clustering process to build an FLTM. Three clustering methods are currently available. Moreover, three modalities are available to drive the clustering process : (i) the clustering method and its parameters are both specified by the user, (ii) the clustering method is specified by the user, and its parameter values are automatically computed, (iii) in the third fully automatic modality, a fully automatic identification of the top best configuration parameter settings over the available clustering methods is achieved, followed by the automatic construction of a consensus clustering from samples of the top best clusterings.
Direction of the development of the versions of the SYLVESTRA++ software: Christine Sinoquet (Associate Professor), project manager of the ANR SAMOGWAS project (2013-2017), LS2N / UMR CNRS 6004 (Digital Science Institute of Nantes), University of Nantes, France.