Version 1.3
Changes since previous version
A possible bug was fixed which could cause problems when
using automatic starting methods (Option 3) to fit mixtures to very large
samples. Since the likelihood can be very small legitimate solutions were not
recorded as their likelihoods were smaller than the value given. The program
used to denote a failed fit. The symptom would be EMMIX always reporting no
initial start found even when stable solutions have been located.
A segmentation fault was fixed (subroutine CAPART)
A character was removed from test data
The mixing proportions are printed even when space saving options are on.
A bug was removed that stopped the bootstrap test from working during the automatic testing (option 3).
The program was modified so that the standard errors in Option 1 read in the input file in the manner described in the readme.
An error was removed from the formulation of the
hierarchical methods; Group Average, Centroid and Ward.