kmeans-plusplus

Kmeans, with a nice starter

This library is:

• Dependency free!
• Typescript!
• Isomorphic!
• Palm oil free!
• Work with multidimensional points (euclidean distance)

Install

$ npm install kmeans-plusplus

Usage

const Clusterer = require('kmeans-plusplus')
 
const data = [[1, 2], [0, 3], [10, 0], [3, 10], [2, 3], [9, 2]]
 
const clusterer = new Clusterer(data)
 
const { centroids, attributions, iterations } = clusterer.clusterize(3)

API

new Clusterer(data, [validate])

data

Type: <Array<[number, number]>

Points array

validate

Type: boolean
Default: true

Set to false if you want to skip the data validation(risky!)

clusterer.clusterize([clusters, maxIterations, distanceFn]) => { centroids, attributions, iterations }

clusters

Type: number
Default: 3

Number of clusters to create

maxIterations

Type: number
Default: 1000

Max number of iterations executed if convergence is not achieved

distanceFn

Type: function: (number[], number[]) => number
Default: squared euclidean

(p1, p2) => {
  const distances = p1.map((_, i) => (p1[i] - p2[i]) * (p1[i] - p2[i]))
  return distances / p1.length
}

Function used to measure distance between points when finding nearest cluster and seeding clusters

centroids

Type: Array<Array<number>>

Final centroids

attributions

Type: Array<number>

Array of the indices of the final cluster relative to each starting point

iterations

Type: Array<{ centroids: Array<Array<number>>, attributions: Array<number> }>

Each iteration of the algorithm, sorted

License

MIT © Yeasteregg