Bayesian linear regression

Algorithm overview

Bayesian linear models place probability distributions over coefficients. They are useful when you want regularized predictions plus uncertainty, or automatic relevance determination that prunes weak features.

JavaScript implementation

BayesianRidge learns a shared coefficient precision and supports predictStd. ARDRegression learns one precision per feature and zeros coefficients whose precision crosses thresholdLambda.

Interactive Bayesian regression playground

Compare Bayesian Ridge and ARD Regression while changing the data and iteration budget. The live curve and holdout metrics come from the selected @kanaries/ml model running in your browser.

Quick start

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

const X = [[0], [1], [2], [3]];
const y = [1, 3, 5, 7];
const model = new Linear.BayesianRidge();
model.fit(X, y);
const mean = model.predict([[4]]);
const standardDeviation = model.predictStd([[4]]);
console.log({ mean, standardDeviation });

Detailed API reference

Both estimators accept maxIter, tol, Gamma-prior hyperparameters, and fitIntercept, and expose coef, intercept, alpha, lambda, sigma, and nIter.