mirror of
https://github.com/boostorg/math.git
synced 2026-07-22 13:34:12 +00:00
Matt Borland
120 lines
5.1 KiB
C++
120 lines
5.1 KiB
C++
/*
|
|
* (C) Copyright Nick Thompson 2018.
|
|
* (C) Copyright Matt Borland 2021.
|
|
* 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)
|
|
*
|
|
* Module build check for boost::math::statistics::means_and_covariance,
|
|
* covariance and correlation_coefficient. It exercises the exported entry
|
|
* points when the test is built against the boost.math module. The double and
|
|
* float expected values mirror the assertions in bivariate_statistics_test.cpp.
|
|
*/
|
|
|
|
#ifndef BOOST_MATH_BUILD_MODULE
|
|
#include <boost/math/statistics/bivariate_statistics.hpp>
|
|
#else
|
|
import boost.math;
|
|
#endif
|
|
|
|
#ifndef BOOST_MATH_BUILD_MODULE
|
|
#include <vector>
|
|
#include <tuple>
|
|
#include <limits>
|
|
#include <cmath>
|
|
#endif
|
|
#include "math_unit_test.hpp"
|
|
|
|
// means_and_covariance returns (mean_u, mean_v, covariance); covariance() pulls
|
|
// out the third element. All expected values below are exact for these inputs.
|
|
template <typename Real>
|
|
void test_covariance()
|
|
{
|
|
using boost::math::statistics::means_and_covariance;
|
|
using boost::math::statistics::covariance;
|
|
|
|
const Real tol = std::numeric_limits<Real>::epsilon();
|
|
|
|
// Covariance of a single element is zero; the means are the elements themselves.
|
|
std::vector<Real> u1{static_cast<Real>(8)};
|
|
std::vector<Real> v1{static_cast<Real>(17)};
|
|
auto temp = means_and_covariance(u1, v1);
|
|
CHECK_ABSOLUTE_ERROR(static_cast<Real>(8), std::get<0>(temp), tol);
|
|
CHECK_ABSOLUTE_ERROR(static_cast<Real>(17), std::get<1>(temp), tol);
|
|
CHECK_ABSOLUTE_ERROR(static_cast<Real>(0), std::get<2>(temp), tol);
|
|
|
|
// u = {8,4}, v = {3,7}: mean_u = 6, mean_v = 5, covariance = -4.
|
|
std::vector<Real> u2{static_cast<Real>(8), static_cast<Real>(4)};
|
|
std::vector<Real> v2{static_cast<Real>(3), static_cast<Real>(7)};
|
|
temp = means_and_covariance(u2, v2);
|
|
CHECK_ABSOLUTE_ERROR(static_cast<Real>(6), std::get<0>(temp), tol);
|
|
CHECK_ABSOLUTE_ERROR(static_cast<Real>(5), std::get<1>(temp), tol);
|
|
CHECK_ABSOLUTE_ERROR(static_cast<Real>(-4), std::get<2>(temp), tol);
|
|
|
|
// v is constant, so cov(u,v) = 0 for any u; mean_u = 2, mean_v = 1.
|
|
std::vector<Real> u3{static_cast<Real>(1), static_cast<Real>(2), static_cast<Real>(3)};
|
|
std::vector<Real> v3{static_cast<Real>(1), static_cast<Real>(1), static_cast<Real>(1)};
|
|
temp = means_and_covariance(u3, v3);
|
|
CHECK_ABSOLUTE_ERROR(static_cast<Real>(2), std::get<0>(temp), tol);
|
|
CHECK_ABSOLUTE_ERROR(static_cast<Real>(1), std::get<1>(temp), tol);
|
|
CHECK_ABSOLUTE_ERROR(static_cast<Real>(0), std::get<2>(temp), tol);
|
|
|
|
// covariance() must agree with the third tuple element.
|
|
CHECK_ABSOLUTE_ERROR(static_cast<Real>(0), covariance(u3, v3), tol);
|
|
// Covariance is symmetric: cov(v,u) = cov(u,v).
|
|
CHECK_ABSOLUTE_ERROR(static_cast<Real>(0), covariance(v3, u3), tol);
|
|
// cov(u,u) = variance(u) = 2/3 for {1,2,3}.
|
|
CHECK_ABSOLUTE_ERROR(static_cast<Real>(2) / static_cast<Real>(3), covariance(u3, u3), tol);
|
|
}
|
|
|
|
// correlation_coefficient returns rho in [-1, 1], or NaN when a sample is constant.
|
|
template <typename Real>
|
|
void test_correlation_coefficient()
|
|
{
|
|
using boost::math::statistics::correlation_coefficient;
|
|
|
|
const Real tol = std::numeric_limits<Real>::epsilon();
|
|
|
|
// A single sample has an undefined correlation coefficient.
|
|
std::vector<Real> a{static_cast<Real>(1)};
|
|
std::vector<Real> b{static_cast<Real>(1)};
|
|
CHECK_NAN(correlation_coefficient(a, b));
|
|
|
|
// Two constant samples are likewise undefined.
|
|
std::vector<Real> u{static_cast<Real>(1), static_cast<Real>(1)};
|
|
std::vector<Real> v{static_cast<Real>(1), static_cast<Real>(1)};
|
|
CHECK_NAN(correlation_coefficient(u, v));
|
|
|
|
// Perfectly correlated data: rho = 1.
|
|
u = {static_cast<Real>(1), static_cast<Real>(2), static_cast<Real>(3)};
|
|
v = {static_cast<Real>(1), static_cast<Real>(2), static_cast<Real>(3)};
|
|
CHECK_ABSOLUTE_ERROR(static_cast<Real>(1), correlation_coefficient(u, v), tol);
|
|
|
|
// Perfectly anti-correlated data: rho = -1, and rho is symmetric in its arguments.
|
|
u = {static_cast<Real>(1), static_cast<Real>(2), static_cast<Real>(3)};
|
|
v = {static_cast<Real>(-1), static_cast<Real>(-2), static_cast<Real>(-3)};
|
|
CHECK_ABSOLUTE_ERROR(static_cast<Real>(-1), correlation_coefficient(u, v), tol);
|
|
CHECK_ABSOLUTE_ERROR(static_cast<Real>(-1), correlation_coefficient(v, u), tol);
|
|
|
|
// One constant dataset => correlation is undefined.
|
|
u = {static_cast<Real>(1), static_cast<Real>(2), static_cast<Real>(3)};
|
|
v = {static_cast<Real>(0), static_cast<Real>(0), static_cast<Real>(0)};
|
|
CHECK_NAN(correlation_coefficient(v, u));
|
|
|
|
// v = {0,0,3} against u = {1,2,3}: rho = sqrt(3)/2.
|
|
u = {static_cast<Real>(1), static_cast<Real>(2), static_cast<Real>(3)};
|
|
v = {static_cast<Real>(0), static_cast<Real>(0), static_cast<Real>(3)};
|
|
CHECK_ABSOLUTE_ERROR(std::sqrt(static_cast<Real>(3)) / static_cast<Real>(2), correlation_coefficient(v, u), tol);
|
|
}
|
|
|
|
int main()
|
|
{
|
|
test_covariance<double>();
|
|
test_correlation_coefficient<double>();
|
|
|
|
test_covariance<float>();
|
|
test_correlation_coefficient<float>();
|
|
|
|
return boost::math::test::report_errors();
|
|
}
|