2
0
mirror of https://github.com/boostorg/math.git synced 2026-01-27 07:02:08 +00:00
Files
math/test/test_ibeta_inv_ab.cpp
John Maddock 99929b1023 Fixed binomial quantile, added complements, various subtle changes to corner cases, tidied up and removed dead code.
Fixed error messages in beta and gamma inverses.
Updated binomial test program, refacted cdf test cases to test cdf, quantile and complements in one hit.
Added accuracy tests for ibeta_inva and ibeta_invb.
Added more docs.


[SVN r3159]
2006-08-20 09:28:27 +00:00

275 lines
10 KiB
C++

// (C) Copyright John Maddock 2006.
// 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)
#include <boost/math/concepts/real_concept.hpp>
#include <boost/test/included/test_exec_monitor.hpp>
#include <boost/test/floating_point_comparison.hpp>
#include <boost/math/special_functions/beta.hpp>
#include <boost/math/tools/stats.hpp>
#include <boost/math/tools/test.hpp>
#include <boost/math/constants/constants.hpp>
#include <boost/type_traits/is_floating_point.hpp>
#include <boost/array.hpp>
#include <boost/lambda/lambda.hpp>
#include <boost/lambda/bind.hpp>
#ifdef TEST_GSL
#include <gsl/gsl_errno.h>
#include <gsl/gsl_message.h>
#endif
#include "handle_test_result.hpp"
//
// DESCRIPTION:
// ~~~~~~~~~~~~
//
// This file tests the incomplete beta function inverses
// ibeta_inva and ibetac_inva. There are three sets of tests:
// 1) TODO!!!! Accuracy tests use values generated with NTL::RR at
// 1000-bit precision and our generic versions of these functions.
// 2) Round trip sanity checks, use the test data for the forward
// functions, and verify that we can get (approximately) back
// where we started.
//
// Note that when this file is first run on a new platform many of
// these tests will fail: the default accuracy is 1 epsilon which
// is too tight for most platforms. In this situation you will
// need to cast a human eye over the error rates reported and make
// a judgement as to whether they are acceptable. Either way please
// report the results to the Boost mailing list. Acceptable rates of
// error are marked up below as a series of regular expressions that
// identify the compiler/stdlib/platform/data-type/test-data/test-function
// along with the maximum expected peek and RMS mean errors for that
// test.
//
void expected_results()
{
//
// Define the max and mean errors expected for
// various compilers and platforms.
//
const char* largest_type;
#ifndef BOOST_MATH_NO_LONG_DOUBLE_MATH_FUNCTIONS
if(boost::math::tools::digits<double>() == boost::math::tools::digits<long double>())
{
largest_type = "(long\\s+)?double";
}
else
{
largest_type = "long double";
}
#else
largest_type = "(long\\s+)?double";
#endif
//
// Catch all cases come last:
//
// TODO!!!!
add_expected_result(
".*", // compiler
".*", // stdlib
".*", // platform
largest_type, // test type(s)
".*", // test data group
".*", 500, 500); // test function
add_expected_result(
".*", // compiler
".*", // stdlib
".*", // platform
"float|double", // test type(s)
".*", // test data group
".*", 5, 3); // test function
add_expected_result(
".*", // compiler
".*", // stdlib
".*", // platform
"real_concept", // test type(s)
".*", // test data group
".*", 1000000, 500000); // test function
//
// Finish off by printing out the compiler/stdlib/platform names,
// we do this to make it easier to mark up expected error rates.
//
std::cout << "Tests run with " << BOOST_COMPILER << ", "
<< BOOST_STDLIB << ", " << BOOST_PLATFORM << std::endl;
}
template <class T>
void test_inverses(const T& data)
{
using namespace std;
typedef typename T::value_type row_type;
typedef typename row_type::value_type value_type;
value_type precision = static_cast<value_type>(ldexp(1.0, 1-boost::math::tools::digits<value_type>()/2)) * 100;
if(boost::math::tools::digits<value_type>() < 50)
precision = 1; // 1% or two decimal digits, all we can hope for when the input is truncated
for(unsigned i = 0; i < data.size(); ++i)
{
//
// These inverse tests are thrown off if the output of the
// incomplete beta is too close to 1: basically there is insuffient
// information left in the value we're using as input to the inverse
// to be able to get back to the original value.
//
if(data[i][5] == 0)
{
BOOST_CHECK_EQUAL(boost::math::ibeta_inva(data[i][1], data[i][2], data[i][5]), boost::math::tools::max_value<value_type>());
BOOST_CHECK_EQUAL(boost::math::ibeta_invb(data[i][0], data[i][2], data[i][5]), boost::math::tools::min_value<value_type>());
}
else if((1 - data[i][5] > 0.001) && (fabs(data[i][5]) >= boost::math::tools::min_value<value_type>()))
{
value_type inv = boost::math::ibeta_inva(data[i][1], data[i][2], data[i][5]);
BOOST_CHECK_CLOSE(data[i][0], inv, precision);
inv = boost::math::ibeta_invb(data[i][0], data[i][2], data[i][5]);
BOOST_CHECK_CLOSE(data[i][1], inv, precision);
}
else if(1 == data[i][5])
{
BOOST_CHECK_EQUAL(boost::math::ibeta_inva(data[i][1], data[i][2], data[i][5]), boost::math::tools::min_value<value_type>());
BOOST_CHECK_EQUAL(boost::math::ibeta_invb(data[i][0], data[i][2], data[i][5]), boost::math::tools::max_value<value_type>());
}
if(data[i][6] == 0)
{
BOOST_CHECK_EQUAL(boost::math::ibetac_inva(data[i][1], data[i][2], data[i][6]), boost::math::tools::min_value<value_type>());
BOOST_CHECK_EQUAL(boost::math::ibetac_invb(data[i][0], data[i][2], data[i][6]), boost::math::tools::max_value<value_type>());
}
else if((1 - data[i][6] > 0.001) && (fabs(data[i][6]) >= boost::math::tools::min_value<value_type>()))
{
value_type inv = boost::math::ibetac_inva(data[i][1], data[i][2], data[i][6]);
BOOST_CHECK_CLOSE(data[i][0], inv, precision);
inv = boost::math::ibetac_invb(data[i][0], data[i][2], data[i][6]);
BOOST_CHECK_CLOSE(data[i][1], inv, precision);
}
else if(data[i][6] == 1)
{
BOOST_CHECK_EQUAL(boost::math::ibetac_inva(data[i][1], data[i][2], data[i][6]), boost::math::tools::max_value<value_type>());
BOOST_CHECK_EQUAL(boost::math::ibetac_invb(data[i][0], data[i][2], data[i][6]), boost::math::tools::min_value<value_type>());
}
}
}
template <class T>
void test_inverses2(const T& data, const char* type_name, const char* test_name)
{
typedef typename T::value_type row_type;
typedef typename row_type::value_type value_type;
typedef value_type (*pg)(value_type, value_type, value_type);
pg funcp = boost::math::ibeta_inva;
using namespace boost::lambda;
boost::math::tools::test_result<value_type> result;
std::cout << "Testing " << test_name << " with type " << type_name
<< "\n~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~\n";
//
// test ibeta_inva(T, T, T) against data:
//
result = boost::math::tools::test(
data,
bind(funcp, ret<value_type>(_1[0]), ret<value_type>(_1[1]), ret<value_type>(_1[2])),
ret<value_type>(_1[3]));
handle_test_result(result, data[result.worst()], result.worst(), type_name, "boost::math::ibeta_inva", test_name);
//
// test ibetac_inva(T, T, T) against data:
//
funcp = boost::math::ibetac_inva;
result = boost::math::tools::test(
data,
bind(funcp, ret<value_type>(_1[0]), ret<value_type>(_1[1]), ret<value_type>(_1[2])),
ret<value_type>(_1[4]));
handle_test_result(result, data[result.worst()], result.worst(), type_name, "boost::math::ibetac_inva", test_name);
//
// test ibeta_invb(T, T, T) against data:
//
funcp = boost::math::ibeta_invb;
result = boost::math::tools::test(
data,
bind(funcp, ret<value_type>(_1[0]), ret<value_type>(_1[1]), ret<value_type>(_1[2])),
ret<value_type>(_1[5]));
handle_test_result(result, data[result.worst()], result.worst(), type_name, "boost::math::ibeta_invb", test_name);
//
// test ibetac_invb(T, T, T) against data:
//
funcp = boost::math::ibetac_invb;
result = boost::math::tools::test(
data,
bind(funcp, ret<value_type>(_1[0]), ret<value_type>(_1[1]), ret<value_type>(_1[2])),
ret<value_type>(_1[6]));
handle_test_result(result, data[result.worst()], result.worst(), type_name, "boost::math::ibetac_invb", test_name);
}
template <class T>
void test_beta(T, const char* name)
{
//
// The actual test data is rather verbose, so it's in a separate file
//
// The contents are as follows, each row of data contains
// five items, input value a, input value b, integration limits x, beta(a, b, x) and ibeta(a, b, x):
//
std::cout << "Running sanity checks for type " << name << std::endl;
# include "ibeta_small_data.ipp"
test_inverses(ibeta_small_data);
# include "ibeta_data.ipp"
test_inverses(ibeta_data);
# include "ibeta_large_data.ipp"
test_inverses(ibeta_large_data);
#ifndef FULL_TEST
if(boost::is_floating_point<T>::value){
#endif
//
// This accuracy test is normally only enabled for "real"
// floating point types and not for class real_concept.
// The reason is that these tests are exceptionally slow
// to complete when T doesn't have Lanczos support defined for it.
//
# include "ibeta_inva_data.ipp"
test_inverses2(ibeta_inva_data, name, "Inverse incomplete beta");
#ifndef FULL_TEST
}
#endif
}
int test_main(int, char* [])
{
expected_results();
#ifdef TEST_GSL
gsl_set_error_handler_off();
#endif
test_beta(0.1F, "float");
test_beta(0.1, "double");
#ifndef BOOST_MATH_NO_LONG_DOUBLE_MATH_FUNCTIONS
test_beta(0.1L, "long double");
test_beta(boost::math::concepts::real_concept(0.1), "real_concept");
#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::cout;
#endif
return 0;
}