For the full README of the original package, please visit hdbscan

This forked repository deals with the soft clustering method described in How Soft Clustering for HDBSCAN Works with a couple of bugs fixed and trying to address the issue #628.

Additional/Updated functions in this package

1. all_points_membership_vectors: Combined soft clustering membership probability vectors for each overdensity
2. all_points_outlier_vectors: Outlier-based membership probabilities for each overdensity
3. all_points_distance_vectors: Distance-based membership probs for each overdensity
4. all_points_prob_in_some_cluster_vectors: Factor used to convert conditional to unconditional probability for each data point

These functions can be called in a similar fashion as in the original package: hdbscan.all_points_membership_vectors(clusterer). Please make sure that prediction_data is set to True when initialising clusterer

Based on the paper:

R.J.G.B. Campello, D. Moulavi, A. Zimek and J. Sander Hierarchical Density Estimates for Data Clustering, Visualization, and Outlier Detection, ACM Trans. on Knowledge Discovery from Data, Vol 10, 1 (July 2015), 1-51.

Licensing

The hdbscan package is 3-clause BSD licensed. Enjoy.