README

Fit nonlinear parameterized functions to paired numeric data with a fluent, dependency-light PHP API. The package supports bounds, weights, forward or central finite differences, damping controls, error and iteration reporting, and predictions from the fitted curve.

Requirements

PHP 8.1 or later
markbaker/matrix 2.1 or later

Installation

composer require luminsports/levenberg-marquardt

Example

use LuminSports\LevenbergMarquardt\LevenbergMarquardt;

$xValues = [0, 1, 2, 3, 4];
$yValues = [1, 3, 5, 7, 9];

$model = (new LevenbergMarquardt)
    ->setParameterizedFunction(
        fn (float $slope, float $intercept) =>
            fn (float $x): float => $slope * $x + $intercept,
    )
    ->setInitialValues([1, 0])
    ->setXValues($xValues)
    ->setYValues($yValues);

$curve = $model->getCurve();

$curve->getParameters(); // approximately ['slope' => 2.0, 'intercept' => 1.0]
$curve->getError();
$curve->getIterations();
$model->predict([5, 6]); // array of Point objects

The parameterized function receives the values to fit and must return a function of the independent variable. Initial, minimum, and maximum values follow the same parameter order. predict() returns Point objects; call toArray() when an ['x' => ..., 'y' => ...] representation is needed.

Development

composer install
composer check-style
composer phpstan
composer test

Generate a text coverage report with:

composer test -- --coverage-text

Background and Credit

The implementation is based on the derivative-free nonlinear least-squares methods described by Brown and Dennis and is a PHP port of mljs/levenberg-marquardt. The upstream MIT license and attribution are retained in LICENSE.md.

License

Released under the MIT License. See LICENSE.md and COPYRIGHT.

luminsports / levenberg-marquardt

Maintainers

Package info

Statistics

Security