Lasso Regression

Algorithm overview

Lasso regression uses L1 regularization, which can shrink some coefficients to zero. It is useful when you want a linear model that can perform simple feature selection while controlling overfitting.

JavaScript implementation

@kanaries/ml implements Linear.LassoRegression with coordinate-descent style optimization and a JavaScript API for browser or Node.js applications.

Interactive Lasso regression playground

Change the L1 penalty, dataset, and noise to see how regularization changes the fitted trend. The model is trained live with Linear.LassoRegression in your browser.

Quick start example

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

const X = [[0, 1], [1, 1], [2, 0], [3, 0]];
const y = [1, 2, 3, 4];

const model = new Linear.LassoRegression({ alpha: 0.1, maxIter: 1000, tol: 1e-6 });
model.fit(X, y);
const pred = model.predict([[4, 0]]);
console.log(pred);

Detailed API reference

new Linear.LassoRegression(props?: {
  alpha?: number;
  fitIntercept?: boolean;
  maxIter?: number;
  tol?: number;
})

Options:

• alpha?: number, default 1. L1 penalty strength.
• fitIntercept?: boolean, default true.
• maxIter?: number, default 1000.
• tol?: number, default 1e-6.

Methods:

• fit(X: number[][], Y: number[]): void
• predict(X: number[][]): number[]

sklearn-style alias: Lasso

Linear.Lasso extends Linear.LassoRegression with identical options and methods. Use it when you prefer the scikit-learn class name:

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

const model = new Linear.Lasso({ alpha: 0.1, fitIntercept: true });
model.fit([[0, 1], [1, 1], [2, 0]], [1, 2, 3]);
const pred = model.predict([[3, 0]]);