Skip to content
Deepbox
09intermediate

Ridge & Lasso Regression

Compare L2 (Ridge) and L1 (Lasso) regularization against plain linear regression on the diabetes dataset.

What you will learn

  • Use deepbox/datasets for loadDiabetes.
  • Use deepbox/ml for LinearRegression, Ridge, Lasso.
  • Use deepbox/metrics for r2Score, mse.
  • Use deepbox/preprocess for trainTestSplit, StandardScaler.
  • Compare L2 (Ridge) and L1 (Lasso) regularization against plain linear regression on the diabetes dataset.

Source

index.ts
/** * Example 09: Ridge & Lasso Regression * * Compare L2 (Ridge) and L1 (Lasso) regularization against plain linear * regression on the diabetes dataset. Ridge shrinks every coefficient toward * zero. Lasso can set some coefficients exactly to zero, which removes those * features from the model. */ import { loadDiabetes } from "deepbox/datasets";import { mse, r2Score } from "deepbox/metrics";import { Lasso, LinearRegression, Ridge } from "deepbox/ml";import { StandardScaler, trainTestSplit } from "deepbox/preprocess"; console.log("=== Ridge & Lasso Regression ===\n"); // Load diabetes dataset for regressionconst diabetes = loadDiabetes();console.log(`Dataset: ${diabetes.data.shape[0]} samples, ${diabetes.data.shape[1]} features\n`); // Split data into training and testing setsconst [XTrain, XTest, yTrain, yTest] = trainTestSplit(diabetes.data, diabetes.target, {  testSize: 0.2,  randomState: 42,}); // Scale featuresconst scaler = new StandardScaler();scaler.fit(XTrain);const XTrainScaled = scaler.transform(XTrain);const XTestScaled = scaler.transform(XTest); console.log("Training models...\n"); // Train different modelsconst models = [  { name: "Linear Regression", model: new LinearRegression() },  { name: "Ridge (α=0.1)", model: new Ridge({ alpha: 0.1 }) },  { name: "Ridge (α=1.0)", model: new Ridge({ alpha: 1.0 }) },  { name: "Ridge (α=10.0)", model: new Ridge({ alpha: 10.0 }) },  { name: "Lasso (α=0.1)", model: new Lasso({ alpha: 0.1 }) },  { name: "Lasso (α=1.0)", model: new Lasso({ alpha: 1.0 }) },]; // Ridge penalizes the sum of squared coefficients, Lasso the sum of their absolute values.// Fit each model and report test R², test MSE and how many coefficients are exactly zero.console.log("Comparison:");console.log("-".repeat(50)); for (const { name, model } of models) {  model.fit(XTrainScaled, yTrain);  const yPred = model.predict(XTestScaled);   const r2 = r2Score(yTest, yPred);  const mseValue = mse(yTest, yPred);   const zeros = model.coef.eq(0).sum().item();  console.log(    `${name.padEnd(20)} R²: ${r2.toFixed(4)}  MSE: ${mseValue.toFixed(2)}  zero coefficients: ${zeros}`  );} console.log("\nRidge keeps all coefficients non-zero and shrinks them smoothly.");console.log("Lasso with a larger alpha sets more coefficients to exactly zero.");

Output

Console output
Console output only: one line per model with test R², test MSE and the zero-coefficient count.