RandLib

Fit probability distributions and handle heavy tails in modern C++. RandLib is a dependency-free, header-only C++20 library for sampling, probability functions, moments, parameter estimation, model selection, goodness-of-fit testing, extreme-value analysis, and stable laws.

RandLib 2 is the current C++20 release line. The source-compatible C++17 maintenance release remains available on 1.x; see the v1 to v2 migration guide.

Quick start

#include <randlib/randlib.hpp>

#include <iostream>
#include <random>
#include <vector>

int main()
{
    std::mt19937_64 engine(42);
    const randlib::GammaRand<double> model(2.5, 0.5);
    std::vector<double> sample(1'000);
    model.Sample(engine, sample);

    const auto fitted = randlib::fit<randlib::GammaRand<double>>(sample);
    std::cout << fitted.distribution.Name() << " AIC=" << fitted.aic << '\n';
}

Every sampler accepts any standard uniform random bit generator. Fitting returns a new distribution; it never mutates the model used to create the data.

When to use RandLib

Library	Sampling	PDF/CDF/quantiles	Fitting and diagnostics	Heavy-tail focus	Prefer it when…
C++ <random>	Yes	No	No	No	You only need the standard set of portable samplers.
Boost.Math	Separate Boost.Random library	Excellent, including multiprecision and configurable policies	No general fitting API	Some heavy-tailed laws	Accuracy, multiprecision, or mature special functions matter most.
StatsLib	Yes	Yes, with optional matrix APIs	No general fitting API	Limited	You prefer an R-like functional API or Eigen/Armadillo/Blaze integration.
RandLib	Yes	Yes	MLE, moments, conjugate Bayes, model selection, goodness-of-fit, censoring and truncation	Stable, Pareto, GPD/GEV and tail estimators	One object-oriented library should cover simulation through fitted heavy-tail models.

These projects are complementary. See the standard <random> inventory, Boost.Math distribution reference, and StatsLib feature list for their current scope.

Use it

Add include/ to your compiler's include path, use the generated single header, or install with CMake 3.21 or newer:

cmake -S . -B build -DCMAKE_BUILD_TYPE=Release
cmake --build build
cmake --install build --prefix /your/prefix

find_package(RandLib 2 REQUIRED)
target_link_libraries(your_target PRIVATE RandLib::RandLib)

The project also supplies a vcpkg overlay port and a Conan 2 recipe. Build controls include:

• RANDLIB_BUILD_TESTS, RANDLIB_BUILD_EXAMPLES, RANDLIB_BUILD_BENCHMARKS, and RANDLIB_BUILD_DOCS;
• RANDLIB_WARNINGS_AS_ERRORS for project-owned targets;
• RANDLIB_PRECOMPILED to compile common explicit instantiations once; and
• RANDLIB_PORTABLE_SPECIAL_FUNCTIONS to force RandLib's Bessel, beta, and zeta implementations.

Learn more

• API reference
• Distribution and numerical-method reference
• Fitting distributions
• Parallel sampling and streams
• Stable distributions
• Accuracy and reference-data policy
• v1 to v2 migration
• Packaging and the single header
• Compiled examples
• Legacy mathematical notes

The changelog records behavior changes and known limitations. Numerical claims are backed by committed high-precision reference data; documented exceptions are tracked in tests/data/reference_tolerances.csv.

Development

cmake --preset dev
cmake --build --preset dev --parallel
ctest --preset dev-unit
ctest --preset dev-statistical

Supported CI compilers include GCC 11–14, Clang 15–18, AppleClang, and MSVC 2022. The library is MIT licensed.