I will describe a simple algorithm for constructing
hierarchical clusterings which are approximately
optimal at every level of granularity -- that is,
for all k, the induced k-clustering is close to
optimal (under a standard clustering cost function).