next up previous contents
Next: Running wrappers from the Up: Meta-Wrappers Previous: Terminators   Contents

Running Hierarchical

The Hierarchical clustering algorithm is most useful in conjunction with MCCV. One can let MCCV determine the optimal number of clusters per node (and thus the number of child nodes), while the Hierarchical wrapper drives the calculations. In the case of our test dataset, we would hope that a Hierarchical/MCCV loop would produce a tree of roughly 15 clusters at the top level with three sub-clusters each. Let's try this approach using the FullEM algorithm as our inner loop as it seemed to perform best in our previous examples.



Lucas Scharenbroich 2003-08-27