Skip to main content

Crate solow_cluster

Crate solow_cluster 

Source
Expand description

§solow-cluster

Unsupervised clustering for the Solow statistical stack.

The crate ships three families that together cover the classical unsupervised-learning use cases:

  • Centroid-based: KMeans with the Arthur-Vassilvitskii k-means++ initialisation and Lloyd’s iterative refinement. Deterministic under a caller-supplied seed via a portable MMIX-LCG PRNG.
  • Density-based: Dbscan — the Ester-Kriegel-Sander-Xu (KDD 1996) ε, min_samples classifier. Reports per-point Core / Border / Noise roles alongside the cluster label.
  • Hierarchical: AgglomerativeClustering with Single, Complete, Average, or Ward linkage. Uses the Lance-Williams update recurrence so the whole dendrogram is built in O(n² · log n) time with O(n²) space.

Every clusterer exposes the classical fit_predict(x) -> labels call, plus a richer typed result carrying centroids / linkage matrices / core-point flags when applicable. All types derive Serialize / Deserialize under the opt-in serde feature.

§Convergence and complexity

  • KMeans: each Lloyd iteration is O(n · k · d). k-means++ pre-fit is O(n · k · d) in expectation and guarantees an O(log k)-competitive initial inertia (Arthur & Vassilvitskii 2007).
  • DBSCAN: naive neighbour lookup is O(n²); use solow-neighbors KDTree externally to obtain O(n log n) for low-dimensional data.
  • AgglomerativeClustering: O(n² · log n) time, O(n²) space via Müllner (2011) Modern hierarchical, agglomerative clustering algorithms.

§Determinism

Every stochastic component (KMeans init, MiniBatchKMeans batch draws) uses a portable 64-bit MMIX linear-congruential generator seeded by the caller. Fits are bit-identical across runs and platforms given a fixed seed and input.

Re-exports§

pub use affinity_propagation::AffinityPropagation;
pub use agglomerative::AgglomerativeClustering;
pub use agglomerative::DendrogramNode;
pub use agglomerative::Linkage;
pub use bgmm::BayesianGaussianMixture;
pub use birch::Birch;
pub use dbscan::Dbscan;
pub use dbscan::DbscanResult;
pub use dbscan::PointRole;
pub use gmm::CovType;
pub use gmm::GaussianMixture;
pub use hdbscan::Hdbscan;
pub use kmeans::KMeans;
pub use kmeans::KMeansInit;
pub use kmeans::KMeansResult;
pub use meanshift::MeanShift;
pub use minibatch::BisectingKMeans;
pub use minibatch::MiniBatchKMeans;
pub use optics::Optics;
pub use spectral::SpectralClustering;

Modules§

affinity_propagation
Affinity Propagation (Frey-Dueck 2007).
agglomerative
AgglomerativeClustering — bottom-up hierarchical clustering with Single, Complete, Average, or Ward linkage.
bgmm
BayesianGaussianMixture — variational-inference GMM with a stick- breaking Dirichlet-process prior on mixture weights (Blei-Jordan 2006).
birch
BIRCH — Balanced Iterative Reducing and Clustering using Hierarchies (Zhang-Ramakrishnan-Livny 1996).
dbscan
Dbscan — Density-Based Spatial Clustering of Applications with Noise (Ester, Kriegel, Sander & Xu 1996).
gmm
Gaussian mixture models fit by Expectation-Maximisation (Dempster-Laird-Rubin 1977; Bishop 2006, §9.2).
hdbscan
HDBSCAN* — hierarchical DBSCAN with mutual-reachability distances and cluster stability, following Campello-Moulavi-Sander (2013).
kmeans
KMeans — Lloyd’s algorithm with k-means++ initialisation.
meanshift
Mean-shift clustering (Comaniciu-Meer 2002).
minibatch
MiniBatchKMeans (Sculley 2010) and BisectingKMeans (Steinbach- Karypis-Kumar 2000) — two additional centroid-based clusterers.
optics
OPTICS (Ankerst-Breunig-Kriegel-Sander 1999) — Ordering Points To Identify the Clustering Structure.
prelude
Commonly-used imports.
spectral
Spectral clustering (Shi-Malik 2000, Ng-Jordan-Weiss 2001).