starlight-regression

A lightweight regression module for the Starlight Machine Learning ecosystem, providing simple and transparent linear regression with evaluation metrics.

Designed to be:

• Minimal
• Educational
• Dependency-free
• Pipeline-friendly

Installation

npm install starlight-regression

Features

• Linear Regression (Normal Equation)
• Automatic bias / intercept handling
• Batch predictions
• Built-in regression metrics
• No external dependencies
• Works seamlessly with other Starlight ML packages

Quick Start

import {
  LinearRegression,
  meanSquaredError,
  rootMeanSquaredError,
  r2Score
} from "starlight-regression";

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

const y = [2, 4, 6, 8];

const model = new LinearRegression();
model.fit(X, y);

const predictions = model.predictBatch(X);

console.log("Predictions:", predictions);
console.log("MSE:", meanSquaredError(y, predictions));
console.log("RMSE:", rootMeanSquaredError(y, predictions));
console.log("R²:", r2Score(y, predictions));

LinearRegression API

new LinearRegression(options?)

new LinearRegression({
  fitIntercept: true // default
});

| Option | Description | | -------------- | ------------------------------ | | fitIntercept | Automatically adds a bias term |

fit(X, y)

Train the regression model using the normal equation.

model.fit(X, y);

• X: 2D array of features
• y: 1D array of target values

predict(x)

Predict a single value.

model.predict([5]);

predictBatch(X)

Predict multiple samples at once.

model.predictBatch(X);

Regression Metrics

Mean Squared Error

meanSquaredError(yTrue, yPred);

Root Mean Squared Error

rootMeanSquaredError(yTrue, yPred);

R² Score

r2Score(yTrue, yPred);

Measures how well the model explains the variance in the data.

Ecosystem Compatibility

starlight-regression integrates naturally with:

• starlight-ml – tokenization & utilities
• starlight-vec – feature vectorization
• starlight-pipeline – ML workflows
• starlight-eval – evaluation tools

Design Philosophy

• No hidden magic
• Pure JavaScript math
• Easy to read, modify, and learn from
• Ideal for education and lightweight ML tasks

Roadmap

• Polynomial regression
• Ridge / Lasso regression
• Gradient descent solver
• Pipeline integration helpers

License

MIT License