hdbscan_clusterer

Simplified HDBSCAN-style clusterer. It builds the mutual-reachability graph, computes a minimum spanning tree, derives the single-linkage hierarchy, condenses the hierarchy using minimum_cluster_size, and selects clusters using eom or leaf selection. Supports continuous attributes only.

The library implements the clusterer_protocol defined in the clustering_protocols library. It provides predicates for learning a clusterer from a dataset, assigning new instances to clusters, and exporting the learned clusterer as a list of predicate clauses or to a file.

Datasets are represented as objects implementing the clustering_dataset_protocol protocol from the clustering_protocols library.

API documentation

Open the ../../apis/library_index.html#hdbscan_clusterer link in a web browser.

Loading

To load this library, load the loader.lgt file:

| ?- logtalk_load(hdbscan_clusterer(loader)).

Testing

To test this library predicates, load the tester.lgt file:

| ?- logtalk_load(hdbscan_clusterer(tester)).

To run the performance benchmark suite, load the tester_performance.lgt file:

| ?- logtalk_load(hdbscan_clusterer(tester_performance)).

Features

  • Hierarchical Density Clustering: Builds the mutual-reachability graph, computes a minimum spanning tree, derives the single-linkage hierarchy, condenses it using minimum_cluster_size, and selects clusters using eom or leaf selection.

  • Continuous Datasets: Accepts datasets containing only continuous attributes.

  • Cluster Selection Methods: Supports both eom and leaf cluster selection.

  • Distance Metrics: Supports Euclidean and Manhattan distances.

  • Optional Feature Scaling: Continuous attributes can be standardized using z-score scaling.

  • Reachability-Based Prediction: New instances are assigned to the selected cluster with the nearest training point when the distance is within the learned cluster reachability threshold; otherwise the atom noise is returned.

  • Noise Detection: Points not assigned to any extracted cluster are retained as noise.

  • Portable Export: Learned clusterers can be exported as clauses or files and reused later.

Options

The following options can be passed to the learn/3 predicate:

  • minimum_points(MinimumPoints): Minimum neighborhood size used when computing core distances and mutual reachability. Default is 2.

  • minimum_cluster_size(MinimumClusterSize): Minimum number of points required for an extracted cluster. Default is 2.

  • cluster_selection_method(Method): Cluster extraction policy. Options: eom (default) or leaf.

  • distance_metric(Metric): Distance metric to use. Options: euclidean (default) or manhattan.

  • feature_scaling(FeatureScaling): Whether to standardize continuous attributes before clustering. Options: on (default) or off.

Clusterer representation

The learned clusterer is represented as a compound term with the functor chosen by the user when exporting the clusterer and arity 4. For example:

hdbscan_clusterer(Encoders, Clusters, Noise, Options)

Where:

  • Encoders: List of continuous attribute encoders storing attribute name, mean, and scale.

    • Clusters: List of cluster(Id, Points, MaxCoreDistance, Stability) terms in cluster-id order.

  • Noise: List of encoded training points classified as noise.

  • Options: Effective training options used to learn the clusterer.