Hierarchical clustering works under Python2.7 but fails under 3.5 #28

First, thanks for providing this library, it provided a very easy way to add a geographical clustering feature to a project I am working on (not presently in an open repository, but I am happy to send you and invite to view if it is of interest).

I recently returned to the body of code after a time and was trying to upgrade it from Python2.7 to Python3.5, and found that the HierarchicalCluster class throws an exception under 3.5 on data it processes flawlessly under 2.7.

I reduced the problem to a minimal test case, which follows at the end of this description.
I added some extra logging, and proved to my own satisfaction that the exception is thrown from logic around lines 5 to 25 of linkage.py which appear to relate to caching distance function results. Under 2.7 the a and b parameters in this passage appear to be clusters, whereas under 3.5 they appear to be the key strings.

This was tested under 1.4.1, originally with the .whl file from PyPi, but also with the source unpacked to a directory alongside my code.

Test case follows:

from cluster import HierarchicalClustering
import logging

points1D = {
'p4' : 5,
'p2' : 6,
'p7' : 10,
'p9' : 120,
'p10' : 121,
'p11' : 119,
}

logging.basicConfig(level=logging.DEBUG)

distance_func = lambda a,b : abs(points1D[a]-points1D[b])
cl = HierarchicalClustering(points1D.keys(), distance_func)
clusters = cl.getlevel(20)

print(clusters)