__check_convergence passes too easily.

[JonLevy]

Per its docstring, __check_convergence, "Ensure[s] that each cluster center is within the tolerance level of the last centroid." And it, returns a boolean that is "True if the cluster centers have converged, False otherwise."

This is not true. In fact, it returns an integer that is equal to the number of clusters centers that have converged within the given tolerance. The k_pod function uses __check_convergence such that if it evaluates True (i.e., at least one cluster center has converged), then iteration ends. This prematurely terminates the algorithm.