clustering_protocols

This library provides protocols used in the implementation of machine learning clustering algorithms. Datasets are represented as objects implementing the clustering_dataset_protocol protocol. Clusterers are represented as objects implementing the clusterer_protocol protocol.

This library also provides test datasets and a small smoke-test suite. The shared test suite also includes cross-library comparison tests for clusterers that implement the same protocols, allowing common datasets and validation failures to be checked in one place.

Concrete clustering algorithms are intentionally out of scope for this package. The goal is to provide a portable foundation for future libraries such as kmeans_clusterer.

API documentation

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

Loading

To load all entities in this library, load the loader.lgt file:

| ?- logtalk_load(clustering_protocols(loader)).

Testing

To run the library smoke tests and shared comparison tests, load the tester.lgt file:

| ?- logtalk_load(clustering_protocols(tester)).

Test datasets

Several sample datasets are included in the test_datasets directory:

• all_noise.lgt — A synthetic 2D continuous dataset with 4 examples and 2 continuous attributes (x, y). The examples are well separated, making the dataset useful for all-noise tests where a density-based clusterer should reject every point under a small epsilon radius.

• bridge_noise.lgt — A synthetic 2D continuous dataset with 10 examples and 2 continuous attributes (x, y). The examples form two dense blobs connected by two sparse bridge points, making the dataset suitable for testing density-based algorithms that should keep the blobs separate while treating the bridge as noise.

• dead_component_blobs.lgt — A synthetic 2D continuous dataset with 6 examples and 2 continuous attributes (x, y). The examples form two tiny ordered blobs sized so an over-specified Gaussian mixture with first_k initialization can drive one component fully dead, making the dataset useful for dead-component policy regression tests.

• duplicate_points.lgt — A synthetic 2D continuous dataset with 6 examples and 2 continuous attributes (x, y). The examples include repeated coordinates forming a dense local cluster plus one isolated outlier, making the dataset useful for duplicate-point and density-threshold tests.

• imbalanced_three_modes.lgt — A synthetic 2D continuous dataset with 9 examples and 2 continuous attributes (x, y). The examples form one dense blob plus two much sparser distant modes, making the dataset useful for imbalanced-cluster and Gaussian mixture stress tests.

• iris_unlabeled.lgt — A compact Iris-derived dataset with 9 examples and 4 continuous attributes (sepal_length, sepal_width, petal_length, petal_width). It is derived from the classic Iris dataset but intentionally omits species labels so it can be used with unsupervised algorithms.

• large_two_blobs.lgt — A synthetic 2D continuous dataset with 100 examples and 2 continuous attributes (x, y). The examples form two dense 5x10 grids that are well separated, making the dataset useful for performance benchmarks that need a larger deterministic density-based clustering workload than the small smoke-test datasets.

• mixed_profiles.lgt — A mixed-feature dataset with 6 examples, 2 continuous attributes (age, income), and 2 discrete attributes (channel, region). This dataset is intended for clustering algorithms that support both numeric and categorical features.

• scaling_bands.lgt — A synthetic 2D continuous dataset with 6 examples and 2 continuous attributes (x, y). The examples form two horizontal bands with large variation along x, making the dataset useful for tests that compare clustering behavior with feature scaling turned on versus off.

• shopping_profiles.lgt — A categorical dataset with 6 examples and 4 discrete attributes (channel, region, loyalty, device). The examples form two clear shopping-profile segments suitable for categorical clustering smoke tests.

• single_blob.lgt — A synthetic 2D continuous dataset with 6 examples and 2 continuous attributes (x, y). The examples form a single compact blob suitable for one-cluster smoke tests.

• two_blobs.lgt — A synthetic 2D continuous dataset with 8 examples and 2 continuous attributes (x, y). The examples form two compact, well-separated blobs suitable for deterministic clustering smoke tests.