This description immediately leads to an inefficient algorithm for the generation of single-link classes.
It was demonstrated in the example above.
It simply consists of thresholding the DC at increasing levels of dissimilarity.
The binary connection matrices are then calculated at each threshold level, from which the connected components can easily be extracted.
This is the basis for many published single-link algorithms.
From the point of view of IR, where one is trying to construct asearchable tree it is too inefficient (see van Rijsbergen[48] for an appropriate implementation).
|
|
