We consider a variable selection problem in mixture models for multilocus genotype data. Indeed it may happen that some loci are not relevant for the clustering. This leads to a model selection problem which is dealt with by penalized maximum likelihood criteria. (Toussile and Gassiat, 2009) showed experimentally the benefit of variable selection, and obtained consistency results for the model selection procedure with BIC-type criteria. In a recent work, Bontemps and Toussile obtain oracle-like inequalities, in a density estimation point of view. Slope heuristics are used in practice to calibrate the penalty. An algorithm named Mixture Model for Genotype Data (MixMoGenD) implements the procedure.
- SMILE in Paris