Occurrence matrix (used to compute minpts of DBSCAN)

[giusepperizzo]

Per each class area in the supervised task, compute the occurrence means per each category. This stage should be done before the frequency and density computation.

As applied to density count, we can think to apply the same independence assumption for computing pairs, triples, ... sets of 10 occurrences.

Worth thinking: E(#occurrence_c1), E(#occurrence_c2) > 1. This means that E1*E2 will return a value much greater. A great value of minpts will negatively affect the creation of clusters (basically all the points in the space will be labeled as noise).
We may switch to compute just the arithmetic mean when they are aggregated.

Anyway let's use the independence assumption and go ahead to E1*E2, ... .