Advanced clustering

Algorithm overview

Birch incrementally compresses observations into clustering-feature subclusters, AffinityPropagation discovers exemplars by message passing, and BisectingKMeans recursively splits high-inertia groups.

JavaScript implementation

These estimators run locally in browser or Node.js and follow the library's fitPredict convention. Affinity propagation stores several O(n²) matrices; as a practical browser ceiling, keep it below roughly 2,000 samples unless you have benchmarked the target device's memory budget.

Interactive advanced clustering playground

Switch among Birch, Affinity Propagation, and Bisecting K-Means, then change each estimator's most important controls. Every view is fitted live with the corresponding @kanaries/ml class, including its reported centers, exemplars, or inertia.

Quick start

import { Clusters } from '@kanaries/ml';

const X = [[-2, -2], [-1.8, -2.1], [2, 2], [2.1, 1.9]];
const labels = new Clusters.Birch({ threshold: 0.5, nClusters: 3 }).fitPredict(X);
console.log(labels);

Birch also supports partialFit; AffinityPropagation exposes exemplar indices and nIter; BisectingKMeans exposes final centers and inertia.

Detailed API reference

Birch accepts threshold, branchingFactor, and nClusters. AffinityPropagation accepts damping, preference, maxIter, convergenceIter, and randomState. BisectingKMeans accepts nClusters, bisectingStrategy, the inner K-Means iteration settings, and randomState.