Page 231 - Contributed Paper Session (CPS) - Volume 4

CPS2203 Thierry D. et al.
T. Dumont and A. Karina Fermin

Figure 1: Boxplots (first plot) and empirical means (second plot) of the
2
distances h for each sample size

Figure 2: Boxplots of the ratio of variables belonging to the right
group for each sample size

4. Numerical results
In Section 3 we presented our approach in nding the optimal variable
clusters unsing a n-sample of a multivariate Bernoulli distribution. Our four
steps estimation procedure consists in
1. Build the set of partitions of interest using the thresholding method
̂
described in 3.1,
2. For each considered partition m ∈ , compute the maximum log-
̂
likelihood of the associated model ℓ and the dimension of its

parameter space,
3. Use the slope heuristic method to approach the optimal penalty
constant ,
/
̂
4. select the model ̂ minimizing among the criterion −ℓ +

In Section 4.1 we apply our procedure on simulated data. It will allow us to
appreciate the performance of the procedure by comparing our estimator with
the true model used to generate the sample. In Section 4.2 we illustrate the
performance of the procedure on the MovieLens dataset. We will provide an

220 | I S I W S C 2 0 1 9