Division by zero with 'precomputed' matrix

[ghost]

Please see the log below:

lbl = hd.HDBSCAN(min_cluster_size=3,metric="precomputed").fit_predict(dst)
File "/usr/local/lib/python2.7/dist-packages/hdbscan/hdbscan_.py", line 549, in fit_predict
self.fit(X)
File "/usr/local/lib/python2.7/dist-packages/hdbscan/hdbscan_.py", line 531, in fit
self.min_spanning_tree) = hdbscan(X, **self.get_params())
File "/usr/local/lib/python2.7/dist-packages/hdbscan/hdbscan.py", line 379, in hdbscan
return tree_to_labels(X, single_linkage_tree, min_cluster_size) + (result_min_span_tree,)
File "/usr/local/lib/python2.7/dist-packages/hdbscan/hdbscan.py", line 53, in _tree_to_labels
labels, probabilities = get_clusters(condensed_tree, stability_dict)
File "hdbscan/_hdbscan_tree.pyx", line 466, in hdbscan._hdbscan_tree.get_clusters (hdbscan/_hdbscan_tree.c:9279)
File "hdbscan/_hdbscan_tree.pyx", line 508, in hdbscan._hdbscan_tree.get_clusters (hdbscan/_hdbscan_tree.c:9142)
File "hdbscan/_hdbscan_tree.pyx", line 418, in hdbscan._hdbscan_tree.get_probabilities (hdbscan/_hdbscan_tree.c:7703)
ZeroDivisionError: float division

'dst' is a Numpy array which encodes the distance matrix. I would appreciate if you would comment on this issue.

[lmcinnes]

Yes, this is a small bug in the soft cluster calculation that can arise is slightly pathological cases (from the point of view of the soft cluster computation -- the clustering could be fairly normal otherwise). Thanks for catching this. I'll try and get a fix written some time today and try and get corresponding releases out some time after that.

[ghost]

Thank you very much, I look forward to it.