ABSTRACT

The criteria and algorithms in the previous chapters depend on computational parameters that were essentially arbitrary but fixed. For any number of clusters, g, and of discards, n− r, we have obtained a finite and moderate number of locally optimal (mixture model) and steady (classification model) solutions. Although this number may still be fairly large, this means a substantial reduction from the initially astronomical number of all possible

A

B

B1

C

D

partitions to a much smaller set. If there is a true solution, we hope that it is still contained there. The final task is to further reduce this number to a handful of acceptable solutions or, in the best case, even to a single one.