You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
I don't it's relevant, but there also seems to be versioning incompatibilities with the hdbscan library. This is despite the fact that some of the methods in cuML's HDBSCAN require hdbscan. The incompatibility is between the combination of numba, numpy, and scikit-learn
Using metric = 'manhattan' seems to kill my kernel also
The text was updated successfully, but these errors were encountered:
Thanks for opening an issue about this @RichieHakim. You are correct to notice that the bug is in the documentation- it should not state precomputed is an option for the metric argument because that feature is not yet supported. The only distance currently supported is 'euclidean' and we do plan to support other distances in the future. Since it's a duplicate of #4460, I'm going to close this issue and add a note to #4460 to also include only the euclidean metric is supported in the documentation for now. We have a pending item in #3879 to add support more distances.
Describe the bug
Attempting to use the
'precomputed'
option for cuml.cluster.HDBSCAN results in an emptyAttributeError
.Steps/Code to reproduce bug
Error received:
Expected behavior
The function should take in a distance matrix (square shaped. I'm using the output of a KNN run)
Environment details (please complete the following information):
Additional context
hdbscan
library. This is despite the fact that some of the methods in cuML's HDBSCAN requirehdbscan
. The incompatibility is between the combination ofnumba
,numpy
, andscikit-learn
metric = 'manhattan'
seems to kill my kernel alsoThe text was updated successfully, but these errors were encountered: