node-kmeans

Node.js asynchronous implementation of the clustering algorithm k-means

Installation

 $ npm install node-kmeans

Example

// Data source: LinkedInconstdata=[{'company': 'Microsoft','size': 91259,'revenue': 60420},{'company': 'IBM','size': 400000,'revenue': 98787},{'company': 'Skype','size': 700,'revenue': 716},{'company': 'SAP','size': 48000,'revenue': 11567},{'company': 'Yahoo!','size': 14000,'revenue': 6426},{'company': 'eBay','size': 15000,'revenue': 8700},];// Create the data 2D-array (vectors) describing the dataletvectors=newArray();for(leti=0;i<data.length;i++){vectors[i]=[data[i]['size'],data[i]['revenue']];}constkmeans=require('node-kmeans');kmeans.clusterize(vectors,{k: 4},(err,res)=>{if(err)console.error(err);elseconsole.log('%o',res);});

Inputs

• vectors is a nXm array (n [lines] : number of points, m [columns] : number of dimensions)
• options object:
  • k : number of clusters
  • distance (optional) : custom distance function returning the distance between two points (a,b) => number, default Euclidian Distance
  • seed (optional) : value that can be provided to get repeatable cluster generation
• callback node-style callback taking error and result argument

Outputs

An array of objects (one for each cluster) with the following properties:

• centroid : array of X elements (X = number of dimensions)
• cluster : array of X elements containing the vectors of the input data
• clusterInd : array of X integers which are the indexes of the input data

To do

• Technique to avoid local optima (mutation, ...)

Author

Philmod <philippe.modard@gmail.com>