FastICA

Algorithm overview

FastICA recovers statistically independent sources from observed mixtures by maximizing non-Gaussianity. Common uses include blind source separation, artifact removal, and exploratory feature extraction.

JavaScript implementation

@kanaries/ml offers parallel and deflation FastICA with deterministic initialization, whitening choices, transforms, and inverse transforms for JS-only signal and analytics workflows.

Quick start example

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

const observedSignals = [[0, 1], [1, 0], [2, 1], [1, 2], [-1, 0]];
const ica = new Decomposition.FastICA({
  nComponents: 2,
  whiten: 'unit-variance',
  randomState: 0,
});
const sources = ica.fitTransform(observedSignals);
const reconstructed = ica.inverseTransform(sources);
console.log({ sources, reconstructed });

Detailed API reference

Options include nComponents, algorithm: 'parallel' | 'deflation', whiten: 'unit-variance' | 'arbitrary-variance' | false, fun: 'logcosh' | 'exp' | 'cube', funArgs.alpha, maxIter, tol, and randomState.

Methods: fit, transform, fitTransform, inverseTransform. Learned fields: components, mixing, mean, and nIter. With whiten: false, input is not centered and nComponents is ignored, matching sklearn semantics.