Module: aprender::cluster

Public module of the aprender-core crate.

Source

crates/aprender-core/src/cluster.rs or directory.

Example

use aprender::cluster::KMeans;
// See `cargo doc -p aprender-core --open` for full API reference.

Module summary

aprender::cluster is the unsupervised-clustering grab bag: classic K-Means plus the algorithms you reach for when K-Means breaks down — DBSCAN for density-based clustering with noise rejection, Gaussian mixture models when clusters are anisotropic, agglomerative hierarchical clustering, spectral clustering for non-convex shapes, and outlier detectors (Isolation Forest, LOF). Every type implements UnsupervisedEstimator so the fit/predict surface is uniform.

Key types

TypeDescription
KMeansLloyd's algorithm with k-means++ init and configurable max_iter / random_state.
DBSCANDensity-based clustering; clusters labelled in usize with sentinel for noise.
GaussianMixtureEM-fitted GMM with CovarianceType (Full, Diag, Spherical, Tied).
AgglomerativeClusteringBottom-up hierarchical clustering with selectable Linkage (Single, Complete, Average, Ward).
SpectralClusteringEigen-decomposition based clustering with selectable Affinity.
IsolationForest, LocalOutlierFactorUnsupervised anomaly detection.

Usage patterns

Pattern 1: K-Means with deterministic seeding

use aprender::prelude::*;

let data = Matrix::from_vec(6, 2, vec![
    1.0, 1.0, 1.1, 0.9, 0.9, 1.1,    // cluster 1
    5.0, 5.0, 5.1, 4.9, 4.9, 5.1,    // cluster 2
]).expect("valid 6x2 matrix");

let mut kmeans = KMeans::new(2)
    .with_max_iter(100)
    .with_random_state(42);
kmeans.fit(&data).expect("kmeans fit");

let labels = kmeans.predict(&data);
assert_eq!(labels[0], labels[1], "cluster coherence");
assert_ne!(labels[0], labels[3], "cluster separation");

Pattern 2: DBSCAN with noise detection

use aprender::cluster::DBSCAN;
use aprender::primitives::Matrix;
use aprender::traits::UnsupervisedEstimator;

let data = Matrix::from_vec(7, 2, vec![
    0.0, 0.0, 0.1, 0.1, 0.2, 0.0,    // dense cluster
    5.0, 5.0, 5.1, 5.1,              // second dense cluster
    100.0, 100.0,                    // noise point
]).expect("valid 7x2 matrix");

let mut dbscan = DBSCAN::new(0.5, 2);  // eps=0.5, min_samples=2
dbscan.fit(&data).expect("dbscan fit");
let labels = dbscan.predict(&data);

// Noise points get a sentinel label distinct from real clusters.
println!("labels: {:?}", labels);

See also

  • preprocessing — StandardScaler is almost always applied before clustering
  • metrics — silhouette_score, inertia for evaluating cluster quality
  • decomposition — ICA / PCA-style dimensionality reduction before clustering
  • models — when you need parametric mixture-of-experts instead of unsupervised partitioning

Full API

Run cargo doc -p aprender-core --open for the rendered rustdoc, or browse docs.rs/aprender for the published version.