1//===----------------------------------------------------------------------===// 2// 3// The LLVM Compiler Infrastructure 4// 5// This file is dual licensed under the MIT and the University of Illinois Open 6// Source Licenses. See LICENSE.TXT for details. 7// 8//===----------------------------------------------------------------------===// 9// 10// REQUIRES: long_tests 11 12// <random> 13 14// template<class RealType = double> 15// class piecewise_constant_distribution 16 17// template<class _URNG> result_type operator()(_URNG& g, const param_type& parm); 18 19#include <random> 20#include <algorithm> 21#include <vector> 22#include <iterator> 23#include <numeric> 24#include <cassert> 25#include <cstddef> 26 27template <class T> 28inline 29T 30sqr(T x) 31{ 32 return x*x; 33} 34 35int main() 36{ 37 { 38 typedef std::piecewise_constant_distribution<> D; 39 typedef D::param_type P; 40 typedef std::mt19937_64 G; 41 G g; 42 double b[] = {10, 14, 16, 17}; 43 double p[] = {25, 62.5, 12.5}; 44 const size_t Np = sizeof(p) / sizeof(p[0]); 45 D d; 46 P pa(b, b+Np+1, p); 47 const int N = 1000000; 48 std::vector<D::result_type> u; 49 for (int i = 0; i < N; ++i) 50 { 51 D::result_type v = d(g, pa); 52 assert(10 <= v && v < 17); 53 u.push_back(v); 54 } 55 std::vector<double> prob(std::begin(p), std::end(p)); 56 double s = std::accumulate(prob.begin(), prob.end(), 0.0); 57 for (std::size_t i = 0; i < prob.size(); ++i) 58 prob[i] /= s; 59 std::sort(u.begin(), u.end()); 60 for (std::size_t i = 0; i < Np; ++i) 61 { 62 typedef std::vector<D::result_type>::iterator I; 63 I lb = std::lower_bound(u.begin(), u.end(), b[i]); 64 I ub = std::lower_bound(u.begin(), u.end(), b[i+1]); 65 const size_t Ni = ub - lb; 66 if (prob[i] == 0) 67 assert(Ni == 0); 68 else 69 { 70 assert(std::abs((double)Ni/N - prob[i]) / prob[i] < .01); 71 double mean = std::accumulate(lb, ub, 0.0) / Ni; 72 double var = 0; 73 double skew = 0; 74 double kurtosis = 0; 75 for (I j = lb; j != ub; ++j) 76 { 77 double dbl = (*j - mean); 78 double d2 = sqr(dbl); 79 var += d2; 80 skew += dbl * d2; 81 kurtosis += d2 * d2; 82 } 83 var /= Ni; 84 double dev = std::sqrt(var); 85 skew /= Ni * dev * var; 86 kurtosis /= Ni * var * var; 87 kurtosis -= 3; 88 double x_mean = (b[i+1] + b[i]) / 2; 89 double x_var = sqr(b[i+1] - b[i]) / 12; 90 double x_skew = 0; 91 double x_kurtosis = -6./5; 92 assert(std::abs((mean - x_mean) / x_mean) < 0.01); 93 assert(std::abs((var - x_var) / x_var) < 0.01); 94 assert(std::abs(skew - x_skew) < 0.01); 95 assert(std::abs((kurtosis - x_kurtosis) / x_kurtosis) < 0.01); 96 } 97 } 98 } 99} 100