Estimation d'un modèle àblocs latent par l'algorithme SEM

Christine Keribin (1, 2) , Gérard Govaert (3) , Gilles Celeux (1)

1 SELECT - Model selection in statistical learning
2 LM-Orsay - Laboratoire de Mathématiques d'Orsay
3 Heudiasyc - Heuristique et Diagnostic des Systèmes Complexes [Compiègne]

Mixture models can be used to deal with the simultaneous clustering problem on a set of objects and a set of variables. The latent block model defines a distribution for each combinaison of an object label and a variable label and the data are supposed to be independent, given the object labels and the variable labels. But the factorization of the joint distribution of the labels, conditionally to the observed data, is not tractable, and the E-step of the EM algorithm cannot be performed. Govaert et Nadif (2008) proposed a variational approximation, which they compared to a CEM algorithm. We present here, in the case of binary data, a SEM algorithm, using a Gibbs sampling for the estimation step, and we compare the results with the other methods.

Domaines: Statistiques [math.ST]; Théorie [stat.TH]

Christine Keribin, Gérard Govaert, Gilles Celeux. Estimation d'un modèle àblocs latent par l'algorithme SEM. 42e Journées de Statistique, SFdS, May 2010, Marseille, France. ⟨hal-00554409⟩