math/test/naive_monte_carlo_test.cpp
2019-08-10 08:50:12 -04:00

545 lines
18 KiB
C++

/*
* Copyright Nick Thompson, 2017
* 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)
*/
#define BOOST_TEST_MODULE naive_monte_carlo_test
#define BOOST_NAIVE_MONTE_CARLO_DEBUG_FAILURES
#include <cmath>
#include <ostream>
#include <boost/lexical_cast.hpp>
#include <boost/type_index.hpp>
#include <boost/test/included/unit_test.hpp>
#include <boost/test/tools/floating_point_comparison.hpp>
#include <boost/math/constants/constants.hpp>
#include <boost/math/quadrature/naive_monte_carlo.hpp>
using std::abs;
using std::vector;
using std::pair;
using boost::math::constants::pi;
using boost::math::quadrature::naive_monte_carlo;
template<class Real>
void test_pi_multithreaded()
{
std::cout << "Testing pi is calculated correctly (multithreaded) using Monte-Carlo on type " << boost::typeindex::type_id<Real>().pretty_name() << "\n";
auto g = [](std::vector<Real> const & x)->Real {
Real r = x[0]*x[0]+x[1]*x[1];
if (r <= 1) {
return 4;
}
return 0;
};
std::vector<std::pair<Real, Real>> bounds{{Real(0), Real(1)}, {Real(0), Real(1)}};
Real error_goal = 0.0002;
naive_monte_carlo<Real, decltype(g)> mc(g, bounds, error_goal,
/*singular =*/ false,/* threads = */ 2, /* seed = */ 18012);
auto task = mc.integrate();
Real pi_estimated = task.get();
if (abs(pi_estimated - pi<Real>())/pi<Real>() > 0.005) {
std::cout << "Error in estimation of pi too high, function calls: " << mc.calls() << "\n";
std::cout << "Final error estimate : " << mc.current_error_estimate() << "\n";
std::cout << "Error goal : " << error_goal << "\n";
BOOST_CHECK_CLOSE_FRACTION(pi_estimated, pi<Real>(), 0.005);
}
}
template<class Real>
void test_pi()
{
std::cout << "Testing pi is calculated correctly using Monte-Carlo on type " << boost::typeindex::type_id<Real>().pretty_name() << "\n";
auto g = [](std::vector<Real> const & x)->Real
{
Real r = x[0]*x[0]+x[1]*x[1];
if (r <= 1)
{
return 4;
}
return 0;
};
std::vector<std::pair<Real, Real>> bounds{{Real(0), Real(1)}, {Real(0), Real(1)}};
Real error_goal = (Real) 0.0002;
naive_monte_carlo<Real, decltype(g)> mc(g, bounds, error_goal,
/*singular =*/ false,/* threads = */ 1, /* seed = */ 128402);
auto task = mc.integrate();
Real pi_estimated = task.get();
if (abs(pi_estimated - pi<Real>())/pi<Real>() > 0.005)
{
std::cout << "Error in estimation of pi too high, function calls: " << mc.calls() << "\n";
std::cout << "Final error estimate : " << mc.current_error_estimate() << "\n";
std::cout << "Error goal : " << error_goal << "\n";
BOOST_CHECK_CLOSE_FRACTION(pi_estimated, pi<Real>(), 0.005);
}
}
template<class Real>
void test_constant()
{
std::cout << "Testing constants are integrated correctly using Monte-Carlo on type " << boost::typeindex::type_id<Real>().pretty_name() << "\n";
auto g = [](std::vector<Real> const &)->Real
{
return 1;
};
std::vector<std::pair<Real, Real>> bounds{{Real(0), Real(1)}, { Real(0), Real(1)}};
naive_monte_carlo<Real, decltype(g)> mc(g, bounds, (Real) 0.0001,
/* singular = */ false, /* threads = */ 1, /* seed = */ 87);
auto task = mc.integrate();
Real one = task.get();
BOOST_CHECK_CLOSE_FRACTION(one, 1, 0.001);
BOOST_CHECK_SMALL(mc.current_error_estimate(), std::numeric_limits<Real>::epsilon());
BOOST_CHECK(mc.calls() > 1000);
}
template<class Real>
void test_exception_from_integrand()
{
std::cout << "Testing that a reasonable action is performed by the Monte-Carlo integrator when the integrand throws an exception on type " << boost::typeindex::type_id<Real>().pretty_name() << "\n";
auto g = [](std::vector<Real> const & x)->Real
{
if (x[0] > 0.5 && x[0] < 0.5001)
{
throw std::domain_error("You have done something wrong.\n");
}
return (Real) 1;
};
std::vector<std::pair<Real, Real>> bounds{{ Real(0), Real(1)}, { Real(0), Real(1)}};
naive_monte_carlo<Real, decltype(g)> mc(g, bounds, (Real) 0.0001);
auto task = mc.integrate();
bool caught_exception = false;
try
{
Real result = task.get();
// Get rid of unused variable warning:
std::ostream cnull(0);
cnull << result;
}
catch(std::exception const &)
{
caught_exception = true;
}
BOOST_CHECK(caught_exception);
}
template<class Real>
void test_cancel_and_restart()
{
std::cout << "Testing that cancellation and restarting works on naive Monte-Carlo integration on type " << boost::typeindex::type_id<Real>().pretty_name() << "\n";
Real exact = boost::lexical_cast<Real>("1.3932039296856768591842462603255");
BOOST_CONSTEXPR const Real A = 1.0 / (pi<Real>() * pi<Real>() * pi<Real>());
auto g = [&](std::vector<Real> const & x)->Real
{
return A / (1.0 - cos(x[0])*cos(x[1])*cos(x[2]));
};
vector<pair<Real, Real>> bounds{{ Real(0), pi<Real>()}, { Real(0), pi<Real>()}, { Real(0), pi<Real>()}};
naive_monte_carlo<Real, decltype(g)> mc(g, bounds, (Real) 0.05, true, 1, 888889);
auto task = mc.integrate();
mc.cancel();
double y = task.get();
// Super low tolerance; because it got canceled so fast:
BOOST_CHECK_CLOSE_FRACTION(y, exact, 1.0);
mc.update_target_error((Real) 0.01);
task = mc.integrate();
y = task.get();
BOOST_CHECK_CLOSE_FRACTION(y, exact, 0.1);
}
template<class Real>
void test_finite_singular_boundary()
{
std::cout << "Testing that finite singular boundaries work on naive Monte-Carlo integration on type " << boost::typeindex::type_id<Real>().pretty_name() << "\n";
using std::pow;
using std::log;
auto g = [](std::vector<Real> const & x)->Real
{
// The first term is singular at x = 0.
// The second at x = 1:
return pow(log(1.0/x[0]), 2) + log1p(-x[0]);
};
vector<pair<Real, Real>> bounds{{Real(0), Real(1)}};
naive_monte_carlo<Real, decltype(g)> mc(g, bounds, (Real) 0.01, true, 1, 1922);
auto task = mc.integrate();
double y = task.get();
BOOST_CHECK_CLOSE_FRACTION(y, 1.0, 0.1);
}
template<class Real>
void test_multithreaded_variance()
{
std::cout << "Testing that variance computed by naive Monte-Carlo integration converges to integral formula on type " << boost::typeindex::type_id<Real>().pretty_name() << "\n";
Real exact_variance = (Real) 1/(Real) 12;
auto g = [&](std::vector<Real> const & x)->Real
{
return x[0];
};
vector<pair<Real, Real>> bounds{{ Real(0), Real(1)}};
naive_monte_carlo<Real, decltype(g)> mc(g, bounds, (Real) 0.001, false, 2, 12341);
auto task = mc.integrate();
Real y = task.get();
BOOST_CHECK_CLOSE_FRACTION(y, 0.5, 0.01);
BOOST_CHECK_CLOSE_FRACTION(mc.variance(), exact_variance, 0.05);
}
template<class Real>
void test_variance()
{
std::cout << "Testing that variance computed by naive Monte-Carlo integration converges to integral formula on type " << boost::typeindex::type_id<Real>().pretty_name() << "\n";
Real exact_variance = (Real) 1/(Real) 12;
auto g = [&](std::vector<Real> const & x)->Real
{
return x[0];
};
vector<pair<Real, Real>> bounds{{ Real(0), Real(1)}};
naive_monte_carlo<Real, decltype(g)> mc(g, bounds, (Real) 0.001, false, 1, 12341);
auto task = mc.integrate();
Real y = task.get();
BOOST_CHECK_CLOSE_FRACTION(y, 0.5, 0.01);
BOOST_CHECK_CLOSE_FRACTION(mc.variance(), exact_variance, 0.05);
}
template<class Real, uint64_t dimension>
void test_product()
{
std::cout << "Testing that product functions are integrated correctly by naive Monte-Carlo on type " << boost::typeindex::type_id<Real>().pretty_name() << "\n";
auto g = [&](std::vector<Real> const & x)->Real
{
double y = 1;
for (uint64_t i = 0; i < x.size(); ++i)
{
y *= 2*x[i];
}
return y;
};
vector<pair<Real, Real>> bounds(dimension);
for (uint64_t i = 0; i < dimension; ++i)
{
bounds[i] = std::make_pair<Real, Real>(0, 1);
}
naive_monte_carlo<Real, decltype(g)> mc(g, bounds, (Real) 0.001, false, 1, 13999);
auto task = mc.integrate();
Real y = task.get();
BOOST_CHECK_CLOSE_FRACTION(y, 1, 0.01);
using std::pow;
Real exact_variance = pow(4.0/3.0, dimension) - 1;
BOOST_CHECK_CLOSE_FRACTION(mc.variance(), exact_variance, 0.1);
}
template<class Real, uint64_t dimension>
void test_alternative_rng_1()
{
std::cout << "Testing that alternative RNGs work correctly using naive Monte-Carlo on type " << boost::typeindex::type_id<Real>().pretty_name() << "\n";
auto g = [&](std::vector<Real> const & x)->Real
{
double y = 1;
for (uint64_t i = 0; i < x.size(); ++i)
{
y *= 2*x[i];
}
return y;
};
vector<pair<Real, Real>> bounds(dimension);
for (uint64_t i = 0; i < dimension; ++i)
{
bounds[i] = std::make_pair<Real, Real>(0, 1);
}
std::cout << "Testing std::mt19937" << std::endl;
naive_monte_carlo<Real, decltype(g), std::mt19937> mc1(g, bounds, (Real) 0.001, false, 1, 1882);
auto task = mc1.integrate();
Real y = task.get();
BOOST_CHECK_CLOSE_FRACTION(y, 1, 0.01);
using std::pow;
Real exact_variance = pow(4.0/3.0, dimension) - 1;
BOOST_CHECK_CLOSE_FRACTION(mc1.variance(), exact_variance, 0.05);
std::cout << "Testing std::knuth_b" << std::endl;
naive_monte_carlo<Real, decltype(g), std::knuth_b> mc2(g, bounds, (Real) 0.001, false, 1, 1883);
task = mc2.integrate();
y = task.get();
BOOST_CHECK_CLOSE_FRACTION(y, 1, 0.01);
std::cout << "Testing std::ranlux48" << std::endl;
naive_monte_carlo<Real, decltype(g), std::ranlux48> mc3(g, bounds, (Real) 0.001, false, 1, 1884);
task = mc3.integrate();
y = task.get();
BOOST_CHECK_CLOSE_FRACTION(y, 1, 0.01);
}
template<class Real, uint64_t dimension>
void test_alternative_rng_2()
{
std::cout << "Testing that alternative RNGs work correctly using naive Monte-Carlo on type " << boost::typeindex::type_id<Real>().pretty_name() << "\n";
auto g = [&](std::vector<Real> const & x)->Real
{
double y = 1;
for (uint64_t i = 0; i < x.size(); ++i)
{
y *= 2*x[i];
}
return y;
};
vector<pair<Real, Real>> bounds(dimension);
for (uint64_t i = 0; i < dimension; ++i)
{
bounds[i] = std::make_pair<Real, Real>(0, 1);
}
std::cout << "Testing std::default_random_engine" << std::endl;
naive_monte_carlo<Real, decltype(g), std::default_random_engine> mc4(g, bounds, (Real) 0.001, false, 1, 1884);
auto task = mc4.integrate();
Real y = task.get();
BOOST_CHECK_CLOSE_FRACTION(y, 1, 0.01);
std::cout << "Testing std::minstd_rand" << std::endl;
naive_monte_carlo<Real, decltype(g), std::minstd_rand> mc5(g, bounds, (Real) 0.001, false, 1, 1887);
task = mc5.integrate();
y = task.get();
BOOST_CHECK_CLOSE_FRACTION(y, 1, 0.01);
std::cout << "Testing std::minstd_rand0" << std::endl;
naive_monte_carlo<Real, decltype(g), std::minstd_rand0> mc6(g, bounds, (Real) 0.001, false, 1, 1889);
task = mc6.integrate();
y = task.get();
BOOST_CHECK_CLOSE_FRACTION(y, 1, 0.01);
}
template<class Real>
void test_upper_bound_infinite()
{
std::cout << "Testing that infinite upper bounds are integrated correctly by naive Monte-Carlo on type " << boost::typeindex::type_id<Real>().pretty_name() << "\n";
auto g = [](std::vector<Real> const & x)->Real
{
return 1.0/(x[0]*x[0] + 1.0);
};
vector<pair<Real, Real>> bounds(1);
for (uint64_t i = 0; i < bounds.size(); ++i)
{
bounds[i] = std::make_pair<Real, Real>(0, std::numeric_limits<Real>::infinity());
}
naive_monte_carlo<Real, decltype(g)> mc(g, bounds, (Real) 0.001, true, 1, 8765);
auto task = mc.integrate();
Real y = task.get();
BOOST_CHECK_CLOSE_FRACTION(y, boost::math::constants::half_pi<Real>(), 0.01);
}
template<class Real>
void test_lower_bound_infinite()
{
std::cout << "Testing that infinite lower bounds are integrated correctly by naive Monte-Carlo on type " << boost::typeindex::type_id<Real>().pretty_name() << "\n";
auto g = [](std::vector<Real> const & x)->Real
{
return 1.0/(x[0]*x[0] + 1.0);
};
vector<pair<Real, Real>> bounds(1);
for (uint64_t i = 0; i < bounds.size(); ++i)
{
bounds[i] = std::make_pair<Real, Real>(-std::numeric_limits<Real>::infinity(), 0);
}
naive_monte_carlo<Real, decltype(g)> mc(g, bounds, (Real) 0.001, true, 1, 1208);
auto task = mc.integrate();
Real y = task.get();
BOOST_CHECK_CLOSE_FRACTION(y, boost::math::constants::half_pi<Real>(), 0.01);
}
template<class Real>
void test_lower_bound_infinite2()
{
std::cout << "Testing that infinite lower bounds (2) are integrated correctly by naive Monte-Carlo on type " << boost::typeindex::type_id<Real>().pretty_name() << "\n";
auto g = [](std::vector<Real> const & x)->Real
{
// If x[0] = inf, this should blow up:
return (x[0]*x[0])/(x[0]*x[0]*x[0]*x[0] + 1.0);
};
vector<pair<Real, Real>> bounds(1);
for (uint64_t i = 0; i < bounds.size(); ++i)
{
bounds[i] = std::make_pair<Real, Real>(-std::numeric_limits<Real>::infinity(), 0);
}
naive_monte_carlo<Real, decltype(g)> mc(g, bounds, (Real) 0.001, true, 1, 1208);
auto task = mc.integrate();
Real y = task.get();
BOOST_CHECK_CLOSE_FRACTION(y, boost::math::constants::half_pi<Real>()/boost::math::constants::root_two<Real>(), 0.01);
}
template<class Real>
void test_double_infinite()
{
std::cout << "Testing that double infinite bounds are integrated correctly by naive Monte-Carlo on type " << boost::typeindex::type_id<Real>().pretty_name() << "\n";
auto g = [](std::vector<Real> const & x)->Real
{
return 1.0/(x[0]*x[0] + 1.0);
};
vector<pair<Real, Real>> bounds(1);
for (uint64_t i = 0; i < bounds.size(); ++i)
{
bounds[i] = std::make_pair<Real, Real>(-std::numeric_limits<Real>::infinity(), std::numeric_limits<Real>::infinity());
}
naive_monte_carlo<Real, decltype(g)> mc(g, bounds, (Real) 0.001, true, 1, 1776);
auto task = mc.integrate();
Real y = task.get();
BOOST_CHECK_CLOSE_FRACTION(y, boost::math::constants::pi<Real>(), 0.01);
}
template<class Real, uint64_t dimension>
void test_radovic()
{
// See: Generalized Halton Sequences in 2008: A Comparative Study, function g1:
std::cout << "Testing that the Radovic function is integrated correctly by naive Monte-Carlo on type " << boost::typeindex::type_id<Real>().pretty_name() << "\n";
auto g = [](std::vector<Real> const & x)->Real
{
using std::abs;
Real alpha = (Real)0.01;
Real z = 1;
for (uint64_t i = 0; i < dimension; ++i)
{
z *= (abs(4*x[i]-2) + alpha)/(1+alpha);
}
return z;
};
vector<pair<Real, Real>> bounds(dimension);
for (uint64_t i = 0; i < bounds.size(); ++i)
{
bounds[i] = std::make_pair<Real, Real>(0, 1);
}
Real error_goal = (Real) 0.001;
naive_monte_carlo<Real, decltype(g)> mc(g, bounds, error_goal, false, 1, 1982);
auto task = mc.integrate();
Real y = task.get();
if (abs(y - 1) > 0.01)
{
std::cout << "Error in estimation of Radovic integral too high, function calls: " << mc.calls() << "\n";
std::cout << "Final error estimate: " << mc.current_error_estimate() << std::endl;
std::cout << "Error goal : " << error_goal << std::endl;
std::cout << "Variance estimate : " << mc.variance() << std::endl;
BOOST_CHECK_CLOSE_FRACTION(y, 1, 0.01);
}
}
BOOST_AUTO_TEST_CASE(naive_monte_carlo_test)
{
std::cout << "Default hardware concurrency = " << std::thread::hardware_concurrency() << std::endl;
#if !defined(TEST) || TEST == 1
test_finite_singular_boundary<double>();
test_finite_singular_boundary<float>();
#endif
#if !defined(TEST) || TEST == 2
test_pi<float>();
test_pi<double>();
#endif
#if !defined(TEST) || TEST == 3
test_pi_multithreaded<float>();
test_constant<float>();
#endif
//test_pi<long double>();
#if !defined(TEST) || TEST == 4
test_constant<double>();
//test_constant<long double>();
test_cancel_and_restart<float>();
#endif
#if !defined(TEST) || TEST == 5
test_exception_from_integrand<float>();
test_variance<float>();
#endif
#if !defined(TEST) || TEST == 6
test_variance<double>();
test_multithreaded_variance<double>();
#endif
#if !defined(TEST) || TEST == 7
test_product<float, 1>();
test_product<float, 2>();
#endif
#if !defined(TEST) || TEST == 8
test_product<float, 3>();
test_product<float, 4>();
test_product<float, 5>();
#endif
#if !defined(TEST) || TEST == 9
test_product<float, 6>();
test_product<double, 1>();
#endif
#if !defined(TEST) || TEST == 10
test_product<double, 2>();
#endif
#if !defined(TEST) || TEST == 11
test_product<double, 3>();
test_product<double, 4>();
#endif
#if !defined(TEST) || TEST == 12
test_upper_bound_infinite<float>();
test_upper_bound_infinite<double>();
#endif
#if !defined(TEST) || TEST == 13
test_lower_bound_infinite<float>();
test_lower_bound_infinite<double>();
#endif
#if !defined(TEST) || TEST == 14
test_lower_bound_infinite2<float>();
#endif
#if !defined(TEST) || TEST == 15
test_double_infinite<float>();
test_double_infinite<double>();
#endif
#if !defined(TEST) || TEST == 16
test_radovic<float, 1>();
test_radovic<float, 2>();
#endif
#if !defined(TEST) || TEST == 17
test_radovic<float, 3>();
test_radovic<double, 1>();
#endif
#if !defined(TEST) || TEST == 18
test_radovic<double, 2>();
test_radovic<double, 3>();
#endif
#if !defined(TEST) || TEST == 19
test_radovic<double, 4>();
test_radovic<double, 5>();
#endif
#if !defined(TEST) || TEST == 20
test_alternative_rng_1<float, 3>();
#endif
#if !defined(TEST) || TEST == 21
test_alternative_rng_1<double, 3>();
#endif
#if !defined(TEST) || TEST == 22
test_alternative_rng_2<float, 3>();
#endif
#if !defined(TEST) || TEST == 23
test_alternative_rng_2<double, 3>();
#endif
}