Regression-js: Least Squares Data Fitting

Common errors

• TypeError: regression.linear is not a function

  cause Attempting to call a method on `regression` when it was imported incorrectly, or the `regression` object itself is undefined due to a CJS/ESM mixup.
  fix Ensure you are using `import regression from 'regression';` for ESM contexts, or `const regression = require('regression');` for CommonJS. If using `regression('linear', data)`, update to `regression.linear(data)` as the old API was removed in v2.0.0.

• TypeError: data is not an array

  cause The `data` argument passed to any regression method is not in the expected `[[x1, y1], [x2, y2], ...]` format.
  fix Verify that your input data is an array of arrays, where each inner array contains exactly two numbers representing an [x, y] coordinate. E.g., `[[1, 2], [3, 4], [5, 6]]`.

• TypeError: Cannot read properties of undefined (reading 'equation')

  cause This usually occurs when trying to access properties like `equation`, `string`, or `r2` from the result of a regression call that failed or returned an unexpected value (e.g., if `data` was malformed and the function returned `undefined`).
  fix Before accessing properties, check if the `result` object is valid and contains the expected properties. Ensure the input `data` format is correct and the regression function executed successfully.

Warnings

• breaking Version 2.0.0 introduced significant API changes, moving from a single `regression(type, data, options)` function to direct method calls like `regression.linear(data, options)`. Old call patterns will no longer work.
  Fix: Update usage to call specific regression methods directly on the imported `regression` object, e.g., `regression.linear(data)` instead of `regression('linear', data)`.
• gotcha Input data must be an array of `[x, y]` pairs (e.g., `[[0, 1], [32, 67]]`). Providing data in other formats (e.g., objects like `{ x: 0, y: 1 }` or separate X and Y arrays) will lead to errors.
  Fix: Ensure all input data is formatted as `Array<[number, number]>`.
• gotcha The `precision` option (defaulting to 2 significant figures) rounds the output equation coefficients. If high precision is required for intermediate calculations, be aware of this rounding or set `precision` to a higher value or `null`.
  Fix: Set the `precision` option in the configuration object to a higher number or `null` for full precision, e.g., `{ precision: 10 }` or `{ precision: null }`.
• deprecated The `lastvalue` and `linearThroughOrigin` methods were removed in version 2.0.0. Linear regression through the origin is now handled by the standard `linear` model.
  Fix: For linear regression through the origin, use `regression.linear()` and interpret the `yIntercept` as zero if that is your model assumption. The `lastvalue` functionality is no longer available directly.
• gotcha The package's last publish was over 8 years ago. While stable, this indicates that the library is in maintenance mode and unlikely to receive new features, active bug fixes, or updates for modern JavaScript language features or security concerns.
  Fix: Consider its long-term viability for new projects. For active development, assess if an actively maintained alternative might be more suitable, or be prepared to fork and maintain if specific updates are needed.

Install

• npm install regression npm
• yarn add regression yarn
• pnpm add regression pnpm

Imports

• regression
  wrong:

  const regression = require('regression');

  correct:

  import regression from 'regression';

  The library primarily uses a default export pattern. The 'regression' package provides a single object containing all regression methods. While CommonJS `require` works in Node.js environments, ESM `import` is the recommended standard for modern JavaScript.
• regression.linear
  wrong:

  import { linear } from 'regression'; // Incorrect - linear is a method of the default export
  const result = linear(data);

  correct:

  import regression from 'regression'; const result = regression.linear(data);

  Individual regression methods like `linear` are properties of the default `regression` object, not named exports themselves. They must be accessed via the imported default object.
• regression.polynomial
  wrong:

  const result = regression('polynomial', data, 3); // Old API from earlier versions

  correct:

  import regression from 'regression'; const result = regression.polynomial(data, { order: 3 });

  Since version 2.0.0, the API for calling regression methods changed from a single `regression(type, data, options)` function to directly calling `regression.type(data, options)`. Also note the `order` option for polynomial regression.

Quickstart

Demonstrates how to import the `regression` library, perform linear and polynomial regression on sample data, and predict values using the generated models.

import regression from 'regression';

// Sample data: [[x1, y1], [x2, y2], ...]
const data = [
  [0, 1],
  [32, 67],
  [12, 79],
  [5, 10],
  [15, 30],
  [25, 50],
  [40, 85],
  [50, 100]
];

// Perform linear regression
const linearResult = regression.linear(data);
console.log('Linear Regression Result:');
console.log('  Equation:', linearResult.equation); // [m, c]
console.log('  String:', linearResult.string);
console.log('  R-squared (R²):', linearResult.r2);
console.log('  Prediction for x=45:', linearResult.predict(45)[1]);

// Perform polynomial regression with order 2
const polynomialResult = regression.polynomial(data, { order: 2 });
console.log('\nPolynomial Regression (Order 2) Result:');
console.log('  Equation:', polynomialResult.equation); // [a_n, ..., a_1, a_0]
console.log('  String:', polynomialResult.string);
console.log('  R-squared (R²):', polynomialResult.r2);
console.log('  Prediction for x=45:', polynomialResult.predict(45)[1]);