FIGURE 4.

Illustration of the lfdr2 group set used in the miRNA expression data. (A) covariates are ordered by lfdr. Covariates are first split at $Q_c$ in two groups for the cut‐off value c such that BFDR$=0.5$. The lower lfdr group is then recursively split at the median into two new groups, as this group is expected to be of more importance. The lfdr1 group set is obtained in the same way; (B) the lfdr values corresponding to the quantiles; (C) the hierarchy of the groups, which is used in the extra level of shrinkage to find a discretization that fits the data well as described in Section 3.4.1. White groups are not selected when fit on the data