Crate ferrolearn_cluster 

ferrolearn-cluster

Clustering algorithms for the ferrolearn machine learning framework.

This crate provides unsupervised clustering methods:

• KMeans — K-Means clustering with k-Means++ initialization and parallelized assignment via Rayon (REQ-6).
• DBSCAN — Density-Based Spatial Clustering of Applications with Noise (REQ-7).
• GaussianMixture — Gaussian Mixture Models via the Expectation-Maximisation algorithm, with four covariance types (full, tied, diagonal, spherical) (REQ-3).
• AgglomerativeClustering — Bottom-up hierarchical clustering with Ward, Complete, Average, and Single linkage (REQ-5).
• MeanShift — Non-parametric mode-seeking clustering (REQ-23).
• SpectralClustering — Graph Laplacian eigenmap clustering (REQ-23).
• OPTICS — Ordering Points To Identify the Clustering Structure (REQ-23).
• Hdbscan — Hierarchical DBSCAN with automatic cluster detection.
• Birch — Balanced Iterative Reducing and Clustering using Hierarchies.
• LabelPropagation — Semi-supervised label propagation through a similarity graph.
• LabelSpreading — Semi-supervised label spreading via normalized graph Laplacian.
• AffinityPropagation — Exemplar-based clustering via message passing, automatically determines number of clusters.
• BisectingKMeans — Divisive hierarchical clustering that recursively bisects the largest cluster.
• BayesianGaussianMixture — Variational Bayesian GMM with automatic component pruning via Dirichlet Process or Dirichlet Distribution priors.
• FeatureAgglomeration — Hierarchical clustering of features (columns) with pooling-based dimensionality reduction.

Design

All algorithms follow the compile-time safety pattern:

• The unfitted config struct implements Fit with Y = () (unsupervised).
• KMeans produces FittedKMeans, which implements Predict (assign to nearest centroid) and Transform (distance to each centroid).
• DBSCAN produces FittedDBSCAN, which only stores labels and core sample indices — it does not implement Predict.
• GaussianMixture produces FittedGaussianMixture, which implements Predict (hard assignment) and Transform (soft responsibilities).
• AgglomerativeClustering produces FittedAgglomerativeClustering, which stores labels and the merge tree — it does not implement Predict.
• MeanShift produces FittedMeanShift, which implements Predict (assign to nearest center).
• SpectralClustering produces FittedSpectralClustering, which stores labels — it does not implement Predict.
• OPTICS produces FittedOPTICS, which stores the reachability ordering and distances — it does not implement Predict.
• Hdbscan produces FittedHdbscan, which stores labels and probabilities — it does not implement Predict.
• Birch produces FittedBirch, which stores labels and subcluster centers — it does not implement Predict.
• LabelPropagation produces FittedLabelPropagation, which implements Predict for new data via nearest-neighbor lookup.
• LabelSpreading produces FittedLabelSpreading, which implements Predict for new data via nearest-neighbor lookup.
• AffinityPropagation produces FittedAffinityPropagation, which stores exemplar indices and labels — it does not implement Predict.
• BisectingKMeans produces FittedBisectingKMeans, which implements Predict (assign to nearest center).
• BayesianGaussianMixture produces FittedBayesianGaussianMixture, which implements Predict (hard assignment).
• FeatureAgglomeration produces FittedFeatureAgglomeration, which implements Transform (pool features).

Float Generics

All algorithms are generic over F: num_traits::Float + Send + Sync + 'static, supporting both f32 and f64.

Re-exports

pub use affinity_propagation::AffinityPropagation;

pub use affinity_propagation::FittedAffinityPropagation;

pub use agglomerative::AgglomerativeClustering;

pub use agglomerative::FittedAgglomerativeClustering;

pub use agglomerative::Linkage;

pub use bayesian_gmm::BayesianCovType;

pub use bayesian_gmm::BayesianGaussianMixture;

pub use bayesian_gmm::FittedBayesianGaussianMixture;

pub use bayesian_gmm::WeightPriorType;

pub use birch::Birch;

pub use birch::FittedBirch;

pub use bisecting_kmeans::BisectingKMeans;

pub use bisecting_kmeans::BisectingStrategy;

pub use bisecting_kmeans::FittedBisectingKMeans;

pub use dbscan::DBSCAN;

pub use dbscan::FittedDBSCAN;

pub use feature_agglomeration::AgglomerativeLinkage;

pub use feature_agglomeration::FeatureAgglomeration;

pub use feature_agglomeration::FittedFeatureAgglomeration;

pub use feature_agglomeration::PoolingFunc;

pub use gmm::CovarianceType;

pub use gmm::FittedGaussianMixture;

pub use gmm::GaussianMixture;

pub use hdbscan::FittedHdbscan;

pub use hdbscan::Hdbscan;

pub use kmeans::FittedKMeans;

pub use kmeans::KMeans;

pub use label_propagation::FittedLabelPropagation;

pub use label_propagation::LabelPropagation;

pub use label_propagation::LabelPropagationKernel;

pub use label_spreading::FittedLabelSpreading;

pub use label_spreading::LabelSpreading;

pub use label_spreading::LabelSpreadingKernel;

pub use mean_shift::FittedMeanShift;

pub use mean_shift::MeanShift;

pub use mini_batch_kmeans::FittedMiniBatchKMeans;

pub use mini_batch_kmeans::MiniBatchKMeans;

pub use mini_batch_kmeans::MiniBatchKMeansInit;

pub use optics::FittedOPTICS;

pub use optics::OPTICS;

pub use spectral::FittedSpectralClustering;

pub use spectral::SpectralClustering;

Modules

affinity_propagation

Affinity Propagation clustering.

agglomerative

Agglomerative (bottom-up) hierarchical clustering.

bayesian_gmm

Bayesian Gaussian Mixture Models via Variational Inference.

birch

BIRCH (Balanced Iterative Reducing and Clustering using Hierarchies).

bisecting_kmeans

Bisecting K-Means clustering.

dbscan

DBSCAN (Density-Based Spatial Clustering of Applications with Noise).

feature_agglomeration

Feature Agglomeration — hierarchical clustering of features.

gmm

Gaussian Mixture Models (GMM) using the Expectation-Maximisation algorithm.

hdbscan

HDBSCAN (Hierarchical Density-Based Spatial Clustering of Applications with Noise).

kmeans

K-Means clustering with k-Means++ initialization.

label_propagation

Label Propagation for semi-supervised classification.

label_spreading

Label Spreading for semi-supervised classification.

mean_shift

Mean Shift clustering.

mini_batch_kmeans

Mini-Batch K-Means clustering.

optics

OPTICS — Ordering Points To Identify the Clustering Structure.

spectral

Spectral Clustering via graph Laplacian eigenmaps.