NearestCentroid

Algorithm overview

NearestCentroid represents each class by the mean feature vector of its training samples, then predicts the class with the closest centroid. It is a fast, interpretable baseline for numeric classification.

JavaScript implementation

@kanaries/ml exposes Neighbors.NearestCentroid for browser and Node.js applications. It supports the same distance type names used by Metrics.Distance.

Quick start example

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

const clf = new Neighbors.NearestCentroid({ metric: 'euclidean' });
clf.fit([[0, 0], [0, 1], [4, 4], [5, 4]], [0, 0, 1, 1]);
const pred = clf.predict([[1, 1], [4, 5]]);

Detailed API reference

new Neighbors.NearestCentroid(props?: {
  metric?: Distance.IDistanceType;
  p?: number;
})

Methods:

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

Class labels are numeric. Ties are resolved by the smaller class label.