1 | //===----------------------------------------------------------------------===// |
2 | // |
3 | // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. |
4 | // See https://llvm.org/LICENSE.txt for license information. |
5 | // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception |
6 | // |
7 | //===----------------------------------------------------------------------===// |
8 | // |
9 | // REQUIRES: long_tests |
10 | |
11 | // <random> |
12 | |
13 | // template<class RealType = double> |
14 | // class weibull_distribution |
15 | |
16 | // template<class _URNG> result_type operator()(_URNG& g, const param_type& parm); |
17 | |
18 | #include <random> |
19 | #include <cassert> |
20 | #include <vector> |
21 | #include <numeric> |
22 | #include <cstddef> |
23 | |
24 | #include "test_macros.h" |
25 | |
26 | template <class T> |
27 | inline |
28 | T |
29 | sqr(T x) |
30 | { |
31 | return x * x; |
32 | } |
33 | |
34 | int main(int, char**) |
35 | { |
36 | { |
37 | typedef std::weibull_distribution<> D; |
38 | typedef D::param_type P; |
39 | typedef std::mt19937 G; |
40 | G g; |
41 | D d(0.5, 2); |
42 | P p(1, .5); |
43 | const int N = 1000000; |
44 | std::vector<D::result_type> u; |
45 | for (int i = 0; i < N; ++i) |
46 | { |
47 | D::result_type v = d(g, p); |
48 | assert(d.min() <= v); |
49 | u.push_back(x: v); |
50 | } |
51 | double mean = std::accumulate(u.begin(), u.end(), 0.0) / u.size(); |
52 | double var = 0; |
53 | double skew = 0; |
54 | double kurtosis = 0; |
55 | for (std::size_t i = 0; i < u.size(); ++i) |
56 | { |
57 | double dbl = (u[i] - mean); |
58 | double d2 = sqr(dbl); |
59 | var += d2; |
60 | skew += dbl * d2; |
61 | kurtosis += d2 * d2; |
62 | } |
63 | var /= u.size(); |
64 | double dev = std::sqrt(x: var); |
65 | skew /= u.size() * dev * var; |
66 | kurtosis /= u.size() * var * var; |
67 | kurtosis -= 3; |
68 | double x_mean = p.b() * std::tgamma(1 + 1/p.a()); |
69 | double x_var = sqr(p.b()) * std::tgamma(1 + 2/p.a()) - sqr(x_mean); |
70 | double x_skew = (sqr(p.b())*p.b() * std::tgamma(1 + 3/p.a()) - |
71 | 3*x_mean*x_var - sqr(x_mean)*x_mean) / |
72 | (std::sqrt(x: x_var)*x_var); |
73 | double x_kurtosis = (sqr(sqr(p.b())) * std::tgamma(1 + 4/p.a()) - |
74 | 4*x_skew*x_var*sqrt(x: x_var)*x_mean - |
75 | 6*sqr(x_mean)*x_var - sqr(sqr(x_mean))) / sqr(x_var) - 3; |
76 | assert(std::abs((mean - x_mean) / x_mean) < 0.01); |
77 | assert(std::abs((var - x_var) / x_var) < 0.01); |
78 | assert(std::abs((skew - x_skew) / x_skew) < 0.01); |
79 | assert(std::abs((kurtosis - x_kurtosis) / x_kurtosis) < 0.01); |
80 | } |
81 | { |
82 | typedef std::weibull_distribution<> D; |
83 | typedef D::param_type P; |
84 | typedef std::mt19937 G; |
85 | G g; |
86 | D d(1, .5); |
87 | P p(2, 3); |
88 | const int N = 1000000; |
89 | std::vector<D::result_type> u; |
90 | for (int i = 0; i < N; ++i) |
91 | { |
92 | D::result_type v = d(g, p); |
93 | assert(d.min() <= v); |
94 | u.push_back(v); |
95 | } |
96 | double mean = std::accumulate(u.begin(), u.end(), 0.0) / u.size(); |
97 | double var = 0; |
98 | double skew = 0; |
99 | double kurtosis = 0; |
100 | for (std::size_t i = 0; i < u.size(); ++i) |
101 | { |
102 | double dbl = (u[i] - mean); |
103 | double d2 = sqr(dbl); |
104 | var += d2; |
105 | skew += dbl * d2; |
106 | kurtosis += d2 * d2; |
107 | } |
108 | var /= u.size(); |
109 | double dev = std::sqrt(x: var); |
110 | skew /= u.size() * dev * var; |
111 | kurtosis /= u.size() * var * var; |
112 | kurtosis -= 3; |
113 | double x_mean = p.b() * std::tgamma(1 + 1/p.a()); |
114 | double x_var = sqr(p.b()) * std::tgamma(1 + 2/p.a()) - sqr(x_mean); |
115 | double x_skew = (sqr(p.b())*p.b() * std::tgamma(1 + 3/p.a()) - |
116 | 3*x_mean*x_var - sqr(x_mean)*x_mean) / |
117 | (std::sqrt(x: x_var)*x_var); |
118 | double x_kurtosis = (sqr(sqr(p.b())) * std::tgamma(1 + 4/p.a()) - |
119 | 4*x_skew*x_var*sqrt(x: x_var)*x_mean - |
120 | 6*sqr(x_mean)*x_var - sqr(sqr(x_mean))) / sqr(x_var) - 3; |
121 | assert(std::abs((mean - x_mean) / x_mean) < 0.01); |
122 | assert(std::abs((var - x_var) / x_var) < 0.01); |
123 | assert(std::abs((skew - x_skew) / x_skew) < 0.01); |
124 | assert(std::abs((kurtosis - x_kurtosis) / x_kurtosis) < 0.03); |
125 | } |
126 | { |
127 | typedef std::weibull_distribution<> D; |
128 | typedef D::param_type P; |
129 | typedef std::mt19937 G; |
130 | G g; |
131 | D d(2, 3); |
132 | P p(.5, 2); |
133 | const int N = 1000000; |
134 | std::vector<D::result_type> u; |
135 | for (int i = 0; i < N; ++i) |
136 | { |
137 | D::result_type v = d(g, p); |
138 | assert(d.min() <= v); |
139 | u.push_back(v); |
140 | } |
141 | double mean = std::accumulate(u.begin(), u.end(), 0.0) / u.size(); |
142 | double var = 0; |
143 | double skew = 0; |
144 | double kurtosis = 0; |
145 | for (std::size_t i = 0; i < u.size(); ++i) |
146 | { |
147 | double dbl = (u[i] - mean); |
148 | double d2 = sqr(dbl); |
149 | var += d2; |
150 | skew += dbl * d2; |
151 | kurtosis += d2 * d2; |
152 | } |
153 | var /= u.size(); |
154 | double dev = std::sqrt(x: var); |
155 | skew /= u.size() * dev * var; |
156 | kurtosis /= u.size() * var * var; |
157 | kurtosis -= 3; |
158 | double x_mean = p.b() * std::tgamma(1 + 1/p.a()); |
159 | double x_var = sqr(p.b()) * std::tgamma(1 + 2/p.a()) - sqr(x_mean); |
160 | double x_skew = (sqr(p.b())*p.b() * std::tgamma(1 + 3/p.a()) - |
161 | 3*x_mean*x_var - sqr(x_mean)*x_mean) / |
162 | (std::sqrt(x: x_var)*x_var); |
163 | double x_kurtosis = (sqr(sqr(p.b())) * std::tgamma(1 + 4/p.a()) - |
164 | 4*x_skew*x_var*sqrt(x: x_var)*x_mean - |
165 | 6*sqr(x_mean)*x_var - sqr(sqr(x_mean))) / sqr(x_var) - 3; |
166 | assert(std::abs((mean - x_mean) / x_mean) < 0.01); |
167 | assert(std::abs((var - x_var) / x_var) < 0.01); |
168 | assert(std::abs((skew - x_skew) / x_skew) < 0.01); |
169 | assert(std::abs((kurtosis - x_kurtosis) / x_kurtosis) < 0.03); |
170 | } |
171 | |
172 | return 0; |
173 | } |
174 | |