math/test/test_gamma_dist.cpp
Matt Borland d01893d215
Add GPU markers to fisher f dist
Add SYCL testing of fisher f dist

Add CUDA fisher f dist testing

Add NVRTC fisher f dist testing

Add GPU support to gamma dist

Add SYCL testing of gamma dist

Add CUDA gamma dist testing

Add NVRTC gamma dist testing

Reduce number of threads per block since it can crash CI

Add GPU support to the geometric dist

Add SYCL testing of geometric dist

Add cuda::std::tie

Add GPU support to inv_discrete_quantile

Add CUDA testing of geometric dist

Add NVRTC testing of geometric dist

Add SYCL testing of inverse_chi_squared dist

Adjust tol

Add NVRTC inverse chi squared dist testing

Add CUDA inverse chi squared dist testing

Add GPU support to inverse gamma dist

Add SYCL testing to inverse gamma dist

Add NVRTC testing of inverse gamma dist

Add CUDA testing of inverse gamma dist
2024-09-04 11:07:17 -04:00

290 lines
10 KiB
C++

// Copyright John Maddock 2006.
// Copyright Paul A. Bristow 2007, 2010.
// Use, modification and distribution are subject to the
// Boost Software License, Version 1.0.
// (See accompanying file LICENSE_1_0.txt
// or copy at http://www.boost.org/LICENSE_1_0.txt)
// test_gamma_dist.cpp
// http://en.wikipedia.org/wiki/Gamma_distribution
// http://www.itl.nist.gov/div898/handbook/eda/section3/eda366b.htm
// Also:
// Weisstein, Eric W. "Gamma Distribution."
// From MathWorld--A Wolfram Web Resource.
// http://mathworld.wolfram.com/GammaDistribution.html
#ifndef SYCL_LANGUAGE_VERSION
#include <pch.hpp> // include directory libs/math/src/tr1/ is needed.
#endif
#include <boost/math/tools/config.hpp>
#ifndef BOOST_MATH_NO_REAL_CONCEPT_TESTS
#include <boost/math/concepts/real_concept.hpp> // for real_concept
#endif
#define BOOST_TEST_MAIN
#include <boost/test/unit_test.hpp> // Boost.Test
#include <boost/test/tools/floating_point_comparison.hpp>
#include <boost/math/distributions/gamma.hpp>
using boost::math::gamma_distribution;
#include "../include_private/boost/math/tools/test.hpp"
#include "test_out_of_range.hpp"
#include <iostream>
#include <iomanip>
using std::cout;
using std::endl;
using std::setprecision;
#include <limits>
using std::numeric_limits;
template <class RealType>
RealType NaivePDF(RealType shape, RealType scale, RealType x)
{
// Deliberately naive PDF calculator again which
// we'll compare our pdf function. However some
// published values to compare against would be better....
using namespace std;
RealType result = log(x) * (shape - 1) - x / scale - boost::math::lgamma(shape) - log(scale) * shape;
return exp(result);
}
template <class RealType>
void check_gamma(RealType shape, RealType scale, RealType x, RealType p, RealType q, RealType tol)
{
BOOST_CHECK_CLOSE(
::boost::math::cdf(
gamma_distribution<RealType>(shape, scale), // distribution.
x), // random variable.
p, // probability.
tol); // %tolerance.
BOOST_CHECK_CLOSE(
::boost::math::cdf(
complement(
gamma_distribution<RealType>(shape, scale), // distribution.
x)), // random variable.
q, // probability complement.
tol); // %tolerance.
if(p < 0.999)
{
BOOST_CHECK_CLOSE(
::boost::math::quantile(
gamma_distribution<RealType>(shape, scale), // distribution.
p), // probability.
x, // random variable.
tol); // %tolerance.
}
if(q < 0.999)
{
BOOST_CHECK_CLOSE(
::boost::math::quantile(
complement(
gamma_distribution<RealType>(shape, scale), // distribution.
q)), // probability complement.
x, // random variable.
tol); // %tolerance.
}
// PDF:
BOOST_CHECK_CLOSE(
boost::math::pdf(
gamma_distribution<RealType>(shape, scale), // distribution.
x), // random variable.
NaivePDF(shape, scale, x), // PDF
tol); // %tolerance.
// LOGPDF:
BOOST_CHECK_CLOSE(
boost::math::logpdf(
gamma_distribution<RealType>(shape, scale), // distribution.
x), // random variable.
log(boost::math::pdf(gamma_distribution<RealType>(shape, scale), x)), // PDF
tol); // %tolerance.
}
template <class RealType>
void test_spots(RealType)
{
// Basic sanity checks
//
// 15 decimal places expressed as a percentage.
// The first tests use values generated by MathCAD,
// and should be accurate to around double precision.
//
RealType tolerance = (std::max)(RealType(5e-14f), std::numeric_limits<RealType>::epsilon() * 20) * 100;
cout << "Tolerance for type " << typeid(RealType).name() << " is " << tolerance << " %" << endl;
check_gamma(
static_cast<RealType>(0.5),
static_cast<RealType>(1),
static_cast<RealType>(0.5),
static_cast<RealType>(0.682689492137085),
static_cast<RealType>(1-0.682689492137085),
tolerance);
check_gamma(
static_cast<RealType>(2),
static_cast<RealType>(1),
static_cast<RealType>(0.5),
static_cast<RealType>(0.090204010431050),
static_cast<RealType>(1-0.090204010431050),
tolerance);
check_gamma(
static_cast<RealType>(40),
static_cast<RealType>(1),
static_cast<RealType>(10),
static_cast<RealType>(7.34163631456064E-13),
static_cast<RealType>(1-7.34163631456064E-13),
tolerance);
//
// Some more test data generated by the online
// calculator at http://espse.ed.psu.edu/edpsych/faculty/rhale/hale/507Mat/statlets/free/pdist.htm
// This has the advantage of supporting the scale parameter as well
// as shape, but has only a few digits accuracy, and produces
// some deeply suspect values if the shape parameter is < 1
// (it doesn't agree with MathCAD or this implementation).
// To be fair the incomplete gamma is tricky to get right in this area...
//
tolerance = 1e-5f * 100; // 5 decimal places as a percentage
cout << "Tolerance for type " << typeid(RealType).name() << " is " << tolerance << " %" << endl;
check_gamma(
static_cast<RealType>(2),
static_cast<RealType>(1)/5,
static_cast<RealType>(0.1),
static_cast<RealType>(0.090204),
static_cast<RealType>(1-0.090204),
tolerance);
check_gamma(
static_cast<RealType>(2),
static_cast<RealType>(1)/5,
static_cast<RealType>(0.5),
static_cast<RealType>(1-0.287298),
static_cast<RealType>(0.287298),
tolerance);
check_gamma(
static_cast<RealType>(3),
static_cast<RealType>(2),
static_cast<RealType>(1),
static_cast<RealType>(0.014388),
static_cast<RealType>(1-0.014388),
tolerance * 10); // one less decimal place in the test value
check_gamma(
static_cast<RealType>(3),
static_cast<RealType>(2),
static_cast<RealType>(5),
static_cast<RealType>(0.456187),
static_cast<RealType>(1-0.456187),
tolerance);
RealType tol2 = boost::math::tools::epsilon<RealType>() * 5 * 100; // 5 eps as a percentage
gamma_distribution<RealType> dist(8, 3);
RealType x = static_cast<RealType>(0.125);
using namespace std; // ADL of std names.
// mean:
BOOST_CHECK_CLOSE(
mean(dist)
, static_cast<RealType>(8*3), tol2);
// variance:
BOOST_CHECK_CLOSE(
variance(dist)
, static_cast<RealType>(8*3*3), tol2);
// std deviation:
BOOST_CHECK_CLOSE(
standard_deviation(dist)
, sqrt(static_cast<RealType>(8*3*3)), tol2);
// hazard:
BOOST_CHECK_CLOSE(
hazard(dist, x)
, pdf(dist, x) / cdf(complement(dist, x)), tol2);
// cumulative hazard:
BOOST_CHECK_CLOSE(
chf(dist, x)
, -log(cdf(complement(dist, x))), tol2);
// coefficient_of_variation:
BOOST_CHECK_CLOSE(
coefficient_of_variation(dist)
, standard_deviation(dist) / mean(dist), tol2);
// mode:
BOOST_CHECK_CLOSE(
mode(dist)
, static_cast<RealType>(7 * 3), tol2);
// skewness:
BOOST_CHECK_CLOSE(
skewness(dist)
, 2 / sqrt(static_cast<RealType>(8)), tol2);
// kurtosis:
BOOST_CHECK_CLOSE(
kurtosis(dist)
, 3 + 6 / static_cast<RealType>(8), tol2);
// kurtosis excess:
BOOST_CHECK_CLOSE(
kurtosis_excess(dist)
, 6 / static_cast<RealType>(8), tol2);
BOOST_CHECK_CLOSE(
median(dist), static_cast<RealType>(23.007748327502412), // double precision test value
(std::max)(tol2, static_cast<RealType>(std::numeric_limits<double>::epsilon() * 2 * 100))); // 2 eps as percent
using std::log;
RealType expected_entropy = RealType(8) + log(RealType(3)) + boost::math::lgamma(RealType(8)) - 7*boost::math::digamma(RealType(8));
BOOST_CHECK_CLOSE(
entropy(dist), expected_entropy, tol2);
// Rely on default definition in derived accessors.
// error tests
check_out_of_range<boost::math::gamma_distribution<RealType> >(1, 1);
BOOST_MATH_CHECK_THROW(boost::math::gamma_distribution<RealType>(0, 1), std::domain_error);
BOOST_MATH_CHECK_THROW(boost::math::gamma_distribution<RealType>(-1, 1), std::domain_error);
BOOST_MATH_CHECK_THROW(boost::math::gamma_distribution<RealType>(1, 0), std::domain_error);
BOOST_MATH_CHECK_THROW(boost::math::gamma_distribution<RealType>(1, -1), std::domain_error);
} // template <class RealType>void test_spots(RealType)
BOOST_AUTO_TEST_CASE( test_main )
{
// Basic sanity-check spot values.
// (Parameter value, arbitrarily zero, only communicates the floating point type).
test_spots(0.0F); // Test float. OK at decdigits = 0 tolerance = 0.0001 %
test_spots(0.0); // Test double. OK at decdigits 7, tolerance = 1e07 %
#ifndef BOOST_MATH_NO_LONG_DOUBLE_MATH_FUNCTIONS
test_spots(0.0L); // Test long double.
#ifndef BOOST_MATH_NO_REAL_CONCEPT_TESTS
test_spots(boost::math::concepts::real_concept(0.)); // Test real concept.
#endif
#else
std::cout << "<note>The long double tests have been disabled on this platform "
"either because the long double overloads of the usual math functions are "
"not available at all, or because they are too inaccurate for these tests "
"to pass.</note>" << std::endl;
#endif
} // BOOST_AUTO_TEST_CASE( test_main )
/*
Output:
Autorun "i:\boost-06-05-03-1300\libs\math\test\Math_test\debug\test_gamma_dist.exe"
Running 1 test case...
Tolerance for type float is 0.000238419 %
Tolerance for type float is 0.001 %
Tolerance for type double is 5e-012 %
Tolerance for type double is 0.001 %
Tolerance for type long double is 5e-012 %
Tolerance for type long double is 0.001 %
Tolerance for type class boost::math::concepts::real_concept is 5e-012 %
Tolerance for type class boost::math::concepts::real_concept is 0.001 %
*** No errors detected
*/