Covariance estimation

Algorithm overview

Covariance matrices describe joint feature variation and power Mahalanobis distance, Gaussian scoring, and graphical models. Shrinkage improves conditioning when data is limited; graphical lasso produces a sparse precision matrix.

JavaScript implementation

Choose EmpiricalCovariance for maximum-likelihood covariance, ShrunkCovariance for a fixed shrinkage amount, LedoitWolf or OAS for data-driven shrinkage, and GraphicalLasso for L1-regularized conditional dependence.

Quick start

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

const X = [[0, 1], [1, 0], [2, 1], [3, 2]];
const model = new Covariance.OAS();
model.fit(X);
console.log(model.mahalanobis([[1, 1]]));

Detailed API reference

Every estimator exposes location, covariance, and precision. The adaptive shrinkage classes expose shrinkageValue; GraphicalLasso uses block coordinate descent, accepts alpha, maxIter, tol, enetTol, and assumeCentered, and exposes nIter.