I don't think so. Exactly how you handle that situation in k-means is kind of up for grabs since you can't exactly compute a centroid without dividing by zero.
I think that one common strategy is to find the data point that is furthest from its centroid and put the empty cluster there. On Wed, Mar 31, 2010 at 12:19 PM, Grant Ingersoll <[email protected]>wrote: > On Mar 31, 2010, at 1:55 PM, Ted Dunning wrote: > > > Empty clusters are not that uncommon with k-means if you specify too > large a > > value for k. > > Still, it shouldn't be an error condition, right?
