Expectation–maximization (EM) algorithm for n-dimensional vectors, implemented in javascript. This allows to fit points with a multivariate gaussian mixture model.
This can be used for statistical classification of multivariate data, anomaly detection, or predictive analytics.
You have n-dimensional datapoints that belong to k groups. With this algorithm, you can detect not only which point belongs to which group, but also what the features of each group (mean, covariance) are.
The algorithm works by repetitively estimating the probability of each point to belong to each group, and maximizing the parameters of the groups so that they best fit the data. Detailed description of the algorithm on wikipedia
var expectation_maximization = require('expectation-maximization');
// n-dimensional data points are stored as a simple array of array
var points = [
[1,2], [3,4], [5,6]
];
var n_groups = 2; // We clusterize our data into two groups
var groups = expectation_maximization(points, n_groups);
////// result:
[
{
weight: .5, // weight (a-priori probabibility) of the group
mu: [2, 3], // mean vector of the group
sigma: [[1,0],[0,1]] // covariance matrix of the group
},
... // other groups
]Auto-generated jsdoc documentation is available here
A basic demo is availaible here: https://lovasoa.github.io/expectation-maximization/dist/