RadiusNeighborsClassifier

Algorithm overview

RadiusNeighborsClassifier assigns a class from all training samples within a fixed distance radius. It is useful when a fixed neighborhood size is more meaningful than a fixed number of neighbors.

JavaScript implementation

@kanaries/ml exposes Neighbors.RadiusNeighborsClassifier with uniform or distance-weighted voting in browser and Node.js applications.

Quick start example

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

const clf = new Neighbors.RadiusNeighborsClassifier({
  radius: 1.5,
  weights: 'uniform',
  outlierLabel: -1,
});

clf.fit([[0], [1], [4]], [0, 0, 1]);
const pred = clf.predict([[0.5], [10]]);

Detailed API reference

new Neighbors.RadiusNeighborsClassifier(props?: {
  radius?: number;
  weights?: 'uniform' | 'distance';
  metric?: Distance.IDistanceType;
  p?: number;
  outlierLabel?: number | null;
})

Methods:

• fit(trainX: number[][], trainY: number[]): void
• predict(testX: number[][]): number[]

If no neighbors are found, predict returns outlierLabel when it is not null; otherwise it throws.