Step3) How to find the best fitting mbf - see Torvik and Triantaphyllou (2002), submitted, for details
minimizing the number of errors yields a maximum likelihood mbf, assuming misclassification probability < 0.5
an error minimizing mbf can be found via a maximum flow algorithm (i.e., in O(X3) time)
an incremental maximum flow algorithm which takes O(X2) time per observation is developed in TnT (2002)