This change replaces the algorithm to be closer to that described by
SPSS Statistical Algorithms. It results in better classification of
"real" data, and better seperated cluster centers.
It also provides a few subcommands which were previously not
implemented.