Let us turn our attention to the assignation of values to the missing attribute values. There are h classes (recall
Section 3). Thus, every training case has an associated vector with 2h components: the h first components are
probabilities (one for every class) and the h last components are risks (one for every class).