NearestNeighbors Search

Algorithm overview

Nearest-neighbor search finds the closest reference samples without requiring labels. It supports similarity lookup, retrieval, deduplication, local analysis, and the search layer behind many non-parametric models.

JavaScript implementation

Neighbors.NearestNeighbors provides the familiar sklearn-style fit, kneighbors, and radiusNeighbors workflow in browser and Node.js. Choose brute force explicitly or reuse the library's KD Tree and Ball Tree search structures. Only the training matrix is serialized; indexes are rebuilt safely for each query.

Quick start

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

const search = new Neighbors.NearestNeighbors({
  nNeighbors: 2,
  algorithm: 'kdTree',
});
search.fit([[0], [2], [5]]);

console.log(search.kneighbors([[1]]));
// { distances: [[1, 1]], indices: [[0, 1]] }

Detailed API reference

Constructor options are nNeighbors, radius, algorithm (auto, brute, kdTree, or ballTree), leafSize, metric, and Minkowski p. kneighbors(X?, nNeighbors?, returnDistance?) returns sorted distances and indices; omitting X queries the fitted samples while excluding each sample itself. radiusNeighbors follows the same pattern for a distance threshold.