Skip to content
Deepbox
07intermediate

Linear Regression

Fit a linear regression to noisy synthetic data (y = 2x + 3) and measure it on a held-out test set.

What you will learn

  • Use deepbox/ml for LinearRegression.
  • Use deepbox/ndarray for arange, chained div, mul, add, reshape.
  • Use deepbox/random for setSeed, rand.
  • Use deepbox/metrics for r2Score, mse, mae.
  • Fit a linear regression to noisy synthetic data (y = 2x + 3) and measure it on a held-out test set.

Source

index.ts
/** * Example 07: Linear Regression * * Fit a linear regression to noisy synthetic data and measure it on a held-out * test set. The true relationship is y = 2x + 3, so the fitted coefficient * should be close to 2 and the intercept close to 3. */ import { mae, mse, r2Score } from "deepbox/metrics";import { LinearRegression } from "deepbox/ml";import { arange } from "deepbox/ndarray";import { trainTestSplit } from "deepbox/preprocess";import { rand, setSeed } from "deepbox/random"; console.log("=== Linear Regression ===\n"); // Seed the generator so every run produces the same noise.setSeed(42); // x runs from 0 to 9.9 in steps of 0.1, shaped [100, 1] as one feature column.const X = arange(0, 100).div(10).reshape([100, 1]); // y = 2x + 3 plus uniform noise in [-1, 1).const noise = rand([100]).sub(0.5).mul(2);const y = X.reshape([100]).mul(2).add(3).add(noise); console.log(`Dataset: ${X.shape[0]} samples, ${X.shape[1]} feature column\n`); // Split data: 80% training, 20% testingconst [XTrain, XTest, yTrain, yTest] = trainTestSplit(X, y, {  testSize: 0.2,  randomState: 42,}); console.log(`Training set: ${XTrain.shape[0]} samples`);console.log(`Test set: ${XTest.shape[0]} samples\n`); // Create and train the linear regression modelconst model = new LinearRegression();model.fit(XTrain, yTrain); console.log("Model trained.");console.log(`Coefficients: ${model.coef?.toString()}`);console.log(`Intercept: ${model.intercept}\n`); // Generate predictions on the test setconst yPred = model.predict(XTest); // Calculate performance metrics. The metric functions return plain numbers.const r2 = r2Score(yTest, yPred);const mseValue = mse(yTest, yPred);const maeValue = mae(yTest, yPred); console.log("Model Performance:");console.log(`R² Score: ${r2.toFixed(4)}`);console.log(`MSE: ${mseValue.toFixed(4)}`);console.log(`MAE: ${maeValue.toFixed(4)}`);

Output

Console output
Console output only: coefficient, intercept and the three test metrics.