Expmax NPM | npm.io

Expmax

Statements	Branches	Functions

Description

Expmax is an expectation maximization (EM) library. It makes use of a gaussian mixture model.

This library allows data clustering of n-dimensional datasets given the amount of clusters wanted.

Overview

Directory structure

.
├── core
│   ├── expmax_core.ts
│   └── gaussian_mixture_core.ts
├── engines
│   └── expmax.ts
├── errors.ts
├── index.ts
├── types.ts
└── utils
    └── math.ts

How to build the library to be used in production-ready projects?

Please refer to the NPM custom commands section

How to use?

First install the library

npm install expmax

Example:

import {ExpMax} from 'expmax';

// Dataset should have data points with same vector space dimension
const dataset: IDataset = {
    'points': [
        [1.1,1],
        [1,2],
        [2,2],
        [2,1],
        [15,15],
        [15,16],
    ],
    'label':'test',
};

const opts: IEmOptions = {    
    'clusterQt':2, // Quantity of clusters you want to fit
    'maxEpochs':1000, // Maximum training cycles
    'threshold': 2e-16 // Threshold (epsilon) used to define convergence
}

const model = new ExpMax(dataset, opts); // Instanciate the model with random values
const trainedModel = model.train() // Train it
console.log(trainedModel);
/*
Output:
[
  {
    mu: [ 15.55, 16 ],
    sigma: [ [Array], [Array] ],
    vectorSpaceDim: 2,
    pi: 0.3333333333333333,
    gamma: [
      1.8570742387734104e-153,
      3.512200996873401e-50,
      3.128974029587457e-48,
      7.275927146729349e-54,
      1,
      1
    ]
  },
  {
    mu: [ 1.525, 1.5 ],
    sigma: [ [Array], [Array] ],
    vectorSpaceDim: 2,
    pi: 0.6666666666666666,
    gamma: [ 1, 1, 1, 1, 5.362024468745314e-82, 1.955669841306763e-88 ]
  }
]*/