This file documents recent notable changes to this project. The format of this file is based on Keep a Changelog, and this project adheres to Semantic Versioning.
0.11.0 - 2025-03-05
HDbscan::fit
now returns an additionalVec
that contains the score of each outlier. The score is computed from the GLOSH (Global-Local Outlier Score from Hierarchies) algorithm.HDbscan::fit
now gracefully handles empty inputs and single-point inputs, returning empty clusters for these edge cases.
- The dummy parameter HDbscan::eps has been removed as it was NOT being used in the current implementation, and the removal will NOT affect the clustering result. The current cluster selection method follows the "eom" (excess of mass) approach. This assumes an equivalent setting to cluster_selection_eps=0.0 and cluster_selection_method="eom" in Scikit-learn's HDBSCAN implementation. Removing this unused parameter helps clarify the clustering behavior and avoids confusion.
0.10.0 - 2024-11-01
HDbscan
now acceptsf32
, in addion tof64
, as the element type of the input matrix.
0.9.0 - 2024-08-09
- Updated ndarray to 0.16.0.
- Updated petal-neighbors to 0.11.0
0.8.0 - 2024-02-21
- Requires Rust 1.74 or later.
- Corrected the dual-tree traversal order in the
HDbscan
implementation whenboruvka: true
is set. Previously, this misplacement causedHDbscan
withboruvka: true
to produce a heavier Minimum Spanning Tree (MST) than whenboruvka: false
was set. (PR #67 from @azizkayumov)
0.7.0 - 2023-12-21
- Requires Rust 1.70 or later.
0.6.0 - 2023-08-07
- Switched to Rust 2021 Edition.
- Requires Rust 1.64 or later.
- Cluster that is smaller than
min_samples
forDbscan
should become outliers.
0.5.1 - 2022-05-03
HDbscan
now allows non-standard layout matrix as input.
0.5.0 - 2022-04-20
- The HDBSCAN clustering algorithm.
- Update MSRV to 1.53.0.
0.4.0 - 2021-07-07
Dbscan
andOptics
now allows caller to designate customizedMetric
to compute distance,Default
is usingEuclidean
.
0.3.0 - 2020-03-29
- Requires Rust 1.49 or later.
- Upgrade ndarray to 0.15.0
- Upgrade ndarray to 0.14.0
- Upgrade petal-neighbors to 0.5.1
0.2.2 - 2020-08-28
Dbscan
andOptics
can accept an array off32
.
- Examples added to
Dbscan
andOptics
.
0.2.1 - 2020-04-13
- The input type of
Fit
andPredict
no longer has to beSized
. With this chagne, the caller may pass a slice as an input.
0.2.0 - 2020-04-10
- Clustering algorithms takes
ArrayBase
as its input, instead ofArrayView
, to allow more types in ndarray.