Hierarchical sorting quickly constructs a
tree-structured clustering, but one which is typically
nonoptimal. In particular, this control strategy suffers from
* ordering* effects: different orderings of the observations
may yield different clusterings [Fisher, Xu & Zard, 1992].
Thus, after an initial clustering phase,
a (possibly offline) process of iterative optimization seeks
to uncover better clusterings.

- Seed Selection, Reordering, and Reclustering
- Iterative Redistribution of Single Observations
- Iterative Hierarchical Redistribution
- Comparisons between Iterative Optimization Strategies
- Discussion of Iterative Optimization Methods

Douglas H. Fisher

Sat Mar 30 11:37:23 CST 1996