Irurueta Numerical

This documentation was generated with the assistance of AI. Please report any inaccuracies.

irurueta-numerical is a Java library of numerical utilities: root finding, single/multi-variable optimization, quadrature integration, interpolation, polynomial fitting and estimation, robust (RANSAC-family) estimation, curve fitting, and Kalman-filter based signal processing. It is part of the com.irurueta family of libraries and builds on top of irurueta-algebra, irurueta-statistics, and irurueta-sorting.

Why this library exists

Numerical algorithms such as root finders, optimizers, and curve fitters are easy to get subtly wrong: convergence criteria, bracketing, and failure modes vary a lot between methods. This library packages well-known, tested implementations of these algorithms behind a small set of consistent conventions, so callers can swap one algorithm for another (e.g. a BrentSingleRootEstimator for a NewtonRaphsonSingleRootEstimator) without relearning a new API shape each time.

Core concepts

Function evaluation: Algorithms don’t take lambdas; they take a listener instance (e.g. SingleDimensionFunctionEvaluatorListener, MultiDimensionFunctionEvaluatorListener) whose evaluate(…) method may throw a checked EvaluationException. Implement one of these interfaces to plug your own function into any root finder, optimizer, integrator, or fitter.
Estimator/fitter lifecycle: Most estimators and fitters share a NotReadyException / LockedException protocol: they refuse to run until required inputs are set (isReady()), and refuse to mutate their configuration while an estimation is in progress (isLocked()).
Robust estimation: The robust package provides a reusable RANSAC/LMedS/MSAC/PROSAC/PROMedS framework, used both directly and specialized by other packages (e.g. polynomials.estimators) to make an estimator resilient to outliers in the input samples.

flowchart TD A["Start: candidate samples"] --> B["Randomly select a minimal subset"] B --> C["Fit a candidate model to the subset"] C --> D["Evaluate all samples against the candidate model"] D --> E{"Better consensus than best so far?"} E -->|Yes| F["Keep as best model and inliers"] E -->|No| G["Discard candidate"] F --> H{"Stopping criterion met? (iterations or confidence)"} G --> H H -->|No| B H -->|Yes| I["Refine best model using its inliers"] I --> J["Return final model"]

Robust algorithms

The robust package provides a reusable, outlier-resilient estimation framework, built around repeatedly fitting a candidate model to a small random sample and keeping whichever candidate best explains the whole dataset. Five variants are available, trading off speed, accuracy, and how much needs to be known in advance (e.g. an inlier threshold, or a per-sample quality score):

Robust Estimation Framework — the shared hypothesize-and-verify loop, listener hierarchy, and subset selection every algorithm below builds on.
RANSAC (Random Sample Consensus) — the original algorithm: uniform random sampling, scored by inlier count against a fixed threshold.
LMedS (Least Median of Squares) — no threshold required; scores candidates by the median of their residuals instead.
MSAC (M-estimator Sample Consensus) — a bounded-loss, median-based hybrid of RANSAC and LMedS.
PROSAC (Progressive Sample Consensus) — samples progressively from the highest-quality data first, often one to two orders of magnitude faster than RANSAC.
PROMedS (Progressive Least Median of Squares) — combines PROSAC’s progressive sampling with LMedS’s threshold-free criterion.

Signal processing algorithms

The signal.processing package provides the building blocks for filtering noisy, real-time measurements: a 1D convolution utility, a discrete Kalman filter, and a running estimator for a sensor’s measurement noise — the three combine naturally, since noise measured by the third feeds directly into the second.

1D Discrete Convolution — discrete convolution of a signal with a kernel, with configurable edge-handling for positions where the kernel extends past the signal’s boundaries.
Kalman Filter — recursive predict/correct state estimator for discrete-time linear systems observed through noisy measurements.
Measurement Noise Covariance Estimation — running estimate of a sensor’s measurement noise covariance, built up one sample at a time while the sensor’s true state is held constant.

Where to go next

Installation — add the library as a dependency.
Algorithms — catalog of every algorithm the library implements, grouped by problem domain.
Reference — generated reports (Javadoc, coverage, static analysis, quality dashboards) produced by the build.