1 //  Copyright 2014 Marco Guazzone (marco.guazzone@gmail.com)
2 //
3 //  Use, modification and distribution are subject to the
4 //  Boost Software License, Version 1.0. (See accompanying file
5 //  LICENSE_1_0.txt or copy at http://www.boost.org/LICENSE_1_0.txt)
6 //
7 // This module implements the Hyper-Exponential distribution.
8 //
9 // References:
10 // - "Queueing Theory in Manufacturing Systems Analysis and Design" by H.T. Papadopolous, C. Heavey and J. Browne (Chapman & Hall/CRC, 1993)
11 // - http://reference.wolfram.com/language/ref/HyperexponentialDistribution.html
12 // - http://en.wikipedia.org/wiki/Hyperexponential_distribution
13 //
14 
15 #ifndef BOOST_MATH_DISTRIBUTIONS_HYPEREXPONENTIAL_HPP
16 #define BOOST_MATH_DISTRIBUTIONS_HYPEREXPONENTIAL_HPP
17 
18 
19 #include <boost/config.hpp>
20 #include <boost/math/distributions/complement.hpp>
21 #include <boost/math/distributions/detail/common_error_handling.hpp>
22 #include <boost/math/distributions/exponential.hpp>
23 #include <boost/math/policies/policy.hpp>
24 #include <boost/math/special_functions/fpclassify.hpp>
25 #include <boost/math/tools/precision.hpp>
26 #include <boost/math/tools/roots.hpp>
27 #include <boost/range/begin.hpp>
28 #include <boost/range/end.hpp>
29 #include <boost/range/size.hpp>
30 #include <boost/type_traits/has_pre_increment.hpp>
31 #include <cstddef>
32 #include <iterator>
33 #include <limits>
34 #include <numeric>
35 #include <utility>
36 #include <vector>
37 
38 #if !defined(BOOST_NO_CXX11_HDR_INITIALIZER_LIST)
39 # include <initializer_list>
40 #endif
41 
42 #ifdef _MSC_VER
43 # pragma warning (push)
44 # pragma warning(disable:4127) // conditional expression is constant
45 # pragma warning(disable:4389) // '==' : signed/unsigned mismatch in test_tools
46 #endif // _MSC_VER
47 
48 namespace boost { namespace math {
49 
50 namespace detail {
51 
52 template <typename Dist>
53 typename Dist::value_type generic_quantile(const Dist& dist, const typename Dist::value_type& p, const typename Dist::value_type& guess, bool comp, const char* function);
54 
55 } // Namespace detail
56 
57 
58 template <typename RealT, typename PolicyT>
59 class hyperexponential_distribution;
60 
61 
62 namespace /*<unnamed>*/ { namespace hyperexp_detail {
63 
64 template <typename T>
normalize(std::vector<T> & v)65 void normalize(std::vector<T>& v)
66 {
67    if(!v.size())
68       return;  // Our error handlers will get this later
69     const T sum = std::accumulate(v.begin(), v.end(), static_cast<T>(0));
70     T final_sum = 0;
71     const typename std::vector<T>::iterator end = --v.end();
72     for (typename std::vector<T>::iterator it = v.begin();
73          it != end;
74          ++it)
75     {
76         *it /= sum;
77         final_sum += *it;
78     }
79     *end = 1 - final_sum;  // avoids round off errors, ensures the probs really do sum to 1.
80 }
81 
82 template <typename RealT, typename PolicyT>
check_probabilities(char const * function,std::vector<RealT> const & probabilities,RealT * presult,PolicyT const & pol)83 bool check_probabilities(char const* function, std::vector<RealT> const& probabilities, RealT* presult, PolicyT const& pol)
84 {
85     BOOST_MATH_STD_USING
86     const std::size_t n = probabilities.size();
87     RealT sum = 0;
88     for (std::size_t i = 0; i < n; ++i)
89     {
90         if (probabilities[i] < 0
91             || probabilities[i] > 1
92             || !(boost::math::isfinite)(probabilities[i]))
93         {
94             *presult = policies::raise_domain_error<RealT>(function,
95                                                            "The elements of parameter \"probabilities\" must be >= 0 and <= 1, but at least one of them was: %1%.",
96                                                            probabilities[i],
97                                                            pol);
98             return false;
99         }
100         sum += probabilities[i];
101     }
102 
103     //
104     // We try to keep phase probabilities correctly normalized in the distribution constructors,
105     // however in practice we have to allow for a very slight divergence from a sum of exactly 1:
106     //
107     if (fabs(sum - 1) > tools::epsilon<RealT>() * 2)
108     {
109         *presult = policies::raise_domain_error<RealT>(function,
110                                                        "The elements of parameter \"probabilities\" must sum to 1, but their sum is: %1%.",
111                                                        sum,
112                                                        pol);
113         return false;
114     }
115 
116     return true;
117 }
118 
119 template <typename RealT, typename PolicyT>
check_rates(char const * function,std::vector<RealT> const & rates,RealT * presult,PolicyT const & pol)120 bool check_rates(char const* function, std::vector<RealT> const& rates, RealT* presult, PolicyT const& pol)
121 {
122     const std::size_t n = rates.size();
123     for (std::size_t i = 0; i < n; ++i)
124     {
125         if (rates[i] <= 0
126             || !(boost::math::isfinite)(rates[i]))
127         {
128             *presult = policies::raise_domain_error<RealT>(function,
129                                                            "The elements of parameter \"rates\" must be > 0, but at least one of them is: %1%.",
130                                                            rates[i],
131                                                            pol);
132             return false;
133         }
134     }
135     return true;
136 }
137 
138 template <typename RealT, typename PolicyT>
check_dist(char const * function,std::vector<RealT> const & probabilities,std::vector<RealT> const & rates,RealT * presult,PolicyT const & pol)139 bool check_dist(char const* function, std::vector<RealT> const& probabilities, std::vector<RealT> const& rates, RealT* presult, PolicyT const& pol)
140 {
141     BOOST_MATH_STD_USING
142     if (probabilities.size() != rates.size())
143     {
144         *presult = policies::raise_domain_error<RealT>(function,
145                                                        "The parameters \"probabilities\" and \"rates\" must have the same length, but their size differ by: %1%.",
146                                                        fabs(static_cast<RealT>(probabilities.size())-static_cast<RealT>(rates.size())),
147                                                        pol);
148         return false;
149     }
150 
151     return check_probabilities(function, probabilities, presult, pol)
152            && check_rates(function, rates, presult, pol);
153 }
154 
155 template <typename RealT, typename PolicyT>
check_x(char const * function,RealT x,RealT * presult,PolicyT const & pol)156 bool check_x(char const* function, RealT x, RealT* presult, PolicyT const& pol)
157 {
158     if (x < 0 || (boost::math::isnan)(x))
159     {
160         *presult = policies::raise_domain_error<RealT>(function, "The random variable must be >= 0, but is: %1%.", x, pol);
161         return false;
162     }
163     return true;
164 }
165 
166 template <typename RealT, typename PolicyT>
check_probability(char const * function,RealT p,RealT * presult,PolicyT const & pol)167 bool check_probability(char const* function, RealT p, RealT* presult, PolicyT const& pol)
168 {
169     if (p < 0 || p > 1 || (boost::math::isnan)(p))
170     {
171         *presult = policies::raise_domain_error<RealT>(function, "The probability be >= 0 and <= 1, but is: %1%.", p, pol);
172         return false;
173     }
174     return true;
175 }
176 
177 template <typename RealT, typename PolicyT>
178 RealT quantile_impl(hyperexponential_distribution<RealT, PolicyT> const& dist, RealT const& p, bool comp)
179 {
180     // Don't have a closed form so try to numerically solve the inverse CDF...
181 
182     typedef typename policies::evaluation<RealT, PolicyT>::type value_type;
183     typedef typename policies::normalise<PolicyT,
184                                          policies::promote_float<false>,
185                                          policies::promote_double<false>,
186                                          policies::discrete_quantile<>,
187                                          policies::assert_undefined<> >::type forwarding_policy;
188 
189     static const char* function = comp ? "boost::math::quantile(const boost::math::complemented2_type<boost::math::hyperexponential_distribution<%1%>, %1%>&)"
190                                        : "boost::math::quantile(const boost::math::hyperexponential_distribution<%1%>&, %1%)";
191 
192     RealT result = 0;
193 
194     if (!check_probability(function, p, &result, PolicyT()))
195     {
196         return result;
197     }
198 
199     const std::size_t n = dist.num_phases();
200     const std::vector<RealT> probs = dist.probabilities();
201     const std::vector<RealT> rates = dist.rates();
202 
203     // A possible (but inaccurate) approximation is given below, where the
204     // quantile is given by the weighted sum of exponential quantiles:
205     RealT guess = 0;
206     if (comp)
207     {
208         for (std::size_t i = 0; i < n; ++i)
209         {
210             const exponential_distribution<RealT,PolicyT> exp(rates[i]);
211 
212             guess += probs[i]*quantile(complement(exp, p));
213         }
214     }
215     else
216     {
217         for (std::size_t i = 0; i < n; ++i)
218         {
219             const exponential_distribution<RealT,PolicyT> exp(rates[i]);
220 
221             guess += probs[i]*quantile(exp, p);
222         }
223     }
224 
225     // Fast return in case the Hyper-Exponential is essentially an Exponential
226     if (n == 1)
227     {
228         return guess;
229     }
230 
231     value_type q;
232     q = detail::generic_quantile(hyperexponential_distribution<RealT,forwarding_policy>(probs, rates),
233                                  p,
234                                  guess,
235                                  comp,
236                                  function);
237 
238     result = policies::checked_narrowing_cast<RealT,forwarding_policy>(q, function);
239 
240     return result;
241 }
242 
243 }} // Namespace <unnamed>::hyperexp_detail
244 
245 
246 template <typename RealT = double, typename PolicyT = policies::policy<> >
247 class hyperexponential_distribution
248 {
249     public: typedef RealT value_type;
250     public: typedef PolicyT policy_type;
251 
252 
hyperexponential_distribution()253     public: hyperexponential_distribution()
254     : probs_(1, 1),
255       rates_(1, 1)
256     {
257         RealT err;
258         hyperexp_detail::check_dist("boost::math::hyperexponential_distribution<%1%>::hyperexponential_distribution",
259                                     probs_,
260                                     rates_,
261                                     &err,
262                                     PolicyT());
263     }
264 
265     // Four arg constructor: no ambiguity here, the arguments must be two pairs of iterators:
266     public: template <typename ProbIterT, typename RateIterT>
hyperexponential_distribution(ProbIterT prob_first,ProbIterT prob_last,RateIterT rate_first,RateIterT rate_last)267             hyperexponential_distribution(ProbIterT prob_first, ProbIterT prob_last,
268                                           RateIterT rate_first, RateIterT rate_last)
269     : probs_(prob_first, prob_last),
270       rates_(rate_first, rate_last)
271     {
272         hyperexp_detail::normalize(probs_);
273         RealT err;
274         hyperexp_detail::check_dist("boost::math::hyperexponential_distribution<%1%>::hyperexponential_distribution",
275                                     probs_,
276                                     rates_,
277                                     &err,
278                                     PolicyT());
279     }
280 
281     // Two arg constructor from 2 ranges, we SFINAE this out of existance if
282     // either argument type is incrementable as in that case the type is
283     // probably an iterator:
284     public: template <typename ProbRangeT, typename RateRangeT>
hyperexponential_distribution(ProbRangeT const & prob_range,RateRangeT const & rate_range,typename boost::disable_if_c<boost::has_pre_increment<ProbRangeT>::value||boost::has_pre_increment<RateRangeT>::value>::type * =0)285             hyperexponential_distribution(ProbRangeT const& prob_range,
286                                           RateRangeT const& rate_range,
287                                           typename boost::disable_if_c<boost::has_pre_increment<ProbRangeT>::value || boost::has_pre_increment<RateRangeT>::value>::type* = 0)
288     : probs_(boost::begin(prob_range), boost::end(prob_range)),
289       rates_(boost::begin(rate_range), boost::end(rate_range))
290     {
291         hyperexp_detail::normalize(probs_);
292 
293         RealT err;
294         hyperexp_detail::check_dist("boost::math::hyperexponential_distribution<%1%>::hyperexponential_distribution",
295                                     probs_,
296                                     rates_,
297                                     &err,
298                                     PolicyT());
299     }
300 
301     // Two arg constructor for a pair of iterators: we SFINAE this out of
302     // existance if neither argument types are incrementable.
303     // Note that we allow different argument types here to allow for
304     // construction from an array plus a pointer into that array.
305     public: template <typename RateIterT, typename RateIterT2>
hyperexponential_distribution(RateIterT const & rate_first,RateIterT2 const & rate_last,typename boost::enable_if_c<boost::has_pre_increment<RateIterT>::value||boost::has_pre_increment<RateIterT2>::value>::type * =0)306             hyperexponential_distribution(RateIterT const& rate_first,
307                                           RateIterT2 const& rate_last,
308                                           typename boost::enable_if_c<boost::has_pre_increment<RateIterT>::value || boost::has_pre_increment<RateIterT2>::value>::type* = 0)
309     : probs_(std::distance(rate_first, rate_last), 1), // will be normalized below
310       rates_(rate_first, rate_last)
311     {
312         hyperexp_detail::normalize(probs_);
313 
314         RealT err;
315         hyperexp_detail::check_dist("boost::math::hyperexponential_distribution<%1%>::hyperexponential_distribution",
316                                     probs_,
317                                     rates_,
318                                     &err,
319                                     PolicyT());
320     }
321 
322 #if !defined(BOOST_NO_CXX11_HDR_INITIALIZER_LIST)
323       // Initializer list constructor: allows for construction from array literals:
hyperexponential_distribution(std::initializer_list<RealT> l1,std::initializer_list<RealT> l2)324 public: hyperexponential_distribution(std::initializer_list<RealT> l1, std::initializer_list<RealT> l2)
325       : probs_(l1.begin(), l1.end()),
326         rates_(l2.begin(), l2.end())
327       {
328          hyperexp_detail::normalize(probs_);
329 
330          RealT err;
331          hyperexp_detail::check_dist("boost::math::hyperexponential_distribution<%1%>::hyperexponential_distribution",
332             probs_,
333             rates_,
334             &err,
335             PolicyT());
336       }
337 
hyperexponential_distribution(std::initializer_list<RealT> l1)338 public: hyperexponential_distribution(std::initializer_list<RealT> l1)
339       : probs_(l1.size(), 1),
340         rates_(l1.begin(), l1.end())
341       {
342          hyperexp_detail::normalize(probs_);
343 
344          RealT err;
345          hyperexp_detail::check_dist("boost::math::hyperexponential_distribution<%1%>::hyperexponential_distribution",
346             probs_,
347             rates_,
348             &err,
349             PolicyT());
350       }
351 #endif // !defined(BOOST_NO_CXX11_HDR_INITIALIZER_LIST)
352 
353     // Single argument constructor: argument must be a range.
354     public: template <typename RateRangeT>
hyperexponential_distribution(RateRangeT const & rate_range)355     hyperexponential_distribution(RateRangeT const& rate_range)
356     : probs_(boost::size(rate_range), 1), // will be normalized below
357       rates_(boost::begin(rate_range), boost::end(rate_range))
358     {
359         hyperexp_detail::normalize(probs_);
360 
361         RealT err;
362         hyperexp_detail::check_dist("boost::math::hyperexponential_distribution<%1%>::hyperexponential_distribution",
363                                     probs_,
364                                     rates_,
365                                     &err,
366                                     PolicyT());
367     }
368 
probabilities() const369     public: std::vector<RealT> probabilities() const
370     {
371         return probs_;
372     }
373 
rates() const374     public: std::vector<RealT> rates() const
375     {
376         return rates_;
377     }
378 
num_phases() const379     public: std::size_t num_phases() const
380     {
381         return rates_.size();
382     }
383 
384 
385     private: std::vector<RealT> probs_;
386     private: std::vector<RealT> rates_;
387 }; // class hyperexponential_distribution
388 
389 
390 // Convenient type synonym for double.
391 typedef hyperexponential_distribution<double> hyperexponential;
392 
393 
394 // Range of permissible values for random variable x
395 template <typename RealT, typename PolicyT>
range(hyperexponential_distribution<RealT,PolicyT> const &)396 std::pair<RealT,RealT> range(hyperexponential_distribution<RealT,PolicyT> const&)
397 {
398     if (std::numeric_limits<RealT>::has_infinity)
399     {
400         return std::make_pair(static_cast<RealT>(0), std::numeric_limits<RealT>::infinity()); // 0 to +inf.
401     }
402 
403     return std::make_pair(static_cast<RealT>(0), tools::max_value<RealT>()); // 0 to +<max value>
404 }
405 
406 // Range of supported values for random variable x.
407 // This is range where cdf rises from 0 to 1, and outside it, the pdf is zero.
408 template <typename RealT, typename PolicyT>
support(hyperexponential_distribution<RealT,PolicyT> const &)409 std::pair<RealT,RealT> support(hyperexponential_distribution<RealT,PolicyT> const&)
410 {
411     return std::make_pair(tools::min_value<RealT>(), tools::max_value<RealT>()); // <min value> to +<max value>.
412 }
413 
414 template <typename RealT, typename PolicyT>
415 RealT pdf(hyperexponential_distribution<RealT, PolicyT> const& dist, RealT const& x)
416 {
417     BOOST_MATH_STD_USING
418     RealT result = 0;
419 
420     if (!hyperexp_detail::check_x("boost::math::pdf(const boost::math::hyperexponential_distribution<%1%>&, %1%)", x, &result, PolicyT()))
421     {
422         return result;
423     }
424 
425     const std::size_t n = dist.num_phases();
426     const std::vector<RealT> probs = dist.probabilities();
427     const std::vector<RealT> rates = dist.rates();
428 
429     for (std::size_t i = 0; i < n; ++i)
430     {
431         const exponential_distribution<RealT,PolicyT> exp(rates[i]);
432 
433         result += probs[i]*pdf(exp, x);
434         //result += probs[i]*rates[i]*exp(-rates[i]*x);
435     }
436 
437     return result;
438 }
439 
440 template <typename RealT, typename PolicyT>
441 RealT cdf(hyperexponential_distribution<RealT, PolicyT> const& dist, RealT const& x)
442 {
443     RealT result = 0;
444 
445     if (!hyperexp_detail::check_x("boost::math::cdf(const boost::math::hyperexponential_distribution<%1%>&, %1%)", x, &result, PolicyT()))
446     {
447         return result;
448     }
449 
450     const std::size_t n = dist.num_phases();
451     const std::vector<RealT> probs = dist.probabilities();
452     const std::vector<RealT> rates = dist.rates();
453 
454     for (std::size_t i = 0; i < n; ++i)
455     {
456         const exponential_distribution<RealT,PolicyT> exp(rates[i]);
457 
458         result += probs[i]*cdf(exp, x);
459     }
460 
461     return result;
462 }
463 
464 template <typename RealT, typename PolicyT>
465 RealT quantile(hyperexponential_distribution<RealT, PolicyT> const& dist, RealT const& p)
466 {
467     return hyperexp_detail::quantile_impl(dist, p , false);
468 }
469 
470 template <typename RealT, typename PolicyT>
471 RealT cdf(complemented2_type<hyperexponential_distribution<RealT,PolicyT>, RealT> const& c)
472 {
473     RealT const& x = c.param;
474     hyperexponential_distribution<RealT,PolicyT> const& dist = c.dist;
475 
476     RealT result = 0;
477 
478     if (!hyperexp_detail::check_x("boost::math::cdf(boost::math::complemented2_type<const boost::math::hyperexponential_distribution<%1%>&, %1%>)", x, &result, PolicyT()))
479     {
480         return result;
481     }
482 
483     const std::size_t n = dist.num_phases();
484     const std::vector<RealT> probs = dist.probabilities();
485     const std::vector<RealT> rates = dist.rates();
486 
487     for (std::size_t i = 0; i < n; ++i)
488     {
489         const exponential_distribution<RealT,PolicyT> exp(rates[i]);
490 
491         result += probs[i]*cdf(complement(exp, x));
492     }
493 
494     return result;
495 }
496 
497 
498 template <typename RealT, typename PolicyT>
499 RealT quantile(complemented2_type<hyperexponential_distribution<RealT, PolicyT>, RealT> const& c)
500 {
501     RealT const& p = c.param;
502     hyperexponential_distribution<RealT,PolicyT> const& dist = c.dist;
503 
504     return hyperexp_detail::quantile_impl(dist, p , true);
505 }
506 
507 template <typename RealT, typename PolicyT>
508 RealT mean(hyperexponential_distribution<RealT, PolicyT> const& dist)
509 {
510     RealT result = 0;
511 
512     const std::size_t n = dist.num_phases();
513     const std::vector<RealT> probs = dist.probabilities();
514     const std::vector<RealT> rates = dist.rates();
515 
516     for (std::size_t i = 0; i < n; ++i)
517     {
518         const exponential_distribution<RealT,PolicyT> exp(rates[i]);
519 
520         result += probs[i]*mean(exp);
521     }
522 
523     return result;
524 }
525 
526 template <typename RealT, typename PolicyT>
527 RealT variance(hyperexponential_distribution<RealT, PolicyT> const& dist)
528 {
529     RealT result = 0;
530 
531     const std::size_t n = dist.num_phases();
532     const std::vector<RealT> probs = dist.probabilities();
533     const std::vector<RealT> rates = dist.rates();
534 
535     for (std::size_t i = 0; i < n; ++i)
536     {
537         result += probs[i]/(rates[i]*rates[i]);
538     }
539 
540     const RealT mean = boost::math::mean(dist);
541 
542     result = 2*result-mean*mean;
543 
544     return result;
545 }
546 
547 template <typename RealT, typename PolicyT>
548 RealT skewness(hyperexponential_distribution<RealT,PolicyT> const& dist)
549 {
550     BOOST_MATH_STD_USING
551     const std::size_t n = dist.num_phases();
552     const std::vector<RealT> probs = dist.probabilities();
553     const std::vector<RealT> rates = dist.rates();
554 
555     RealT s1 = 0; // \sum_{i=1}^n \frac{p_i}{\lambda_i}
556     RealT s2 = 0; // \sum_{i=1}^n \frac{p_i}{\lambda_i^2}
557     RealT s3 = 0; // \sum_{i=1}^n \frac{p_i}{\lambda_i^3}
558     for (std::size_t i = 0; i < n; ++i)
559     {
560         const RealT p = probs[i];
561         const RealT r = rates[i];
562         const RealT r2 = r*r;
563         const RealT r3 = r2*r;
564 
565         s1 += p/r;
566         s2 += p/r2;
567         s3 += p/r3;
568     }
569 
570     const RealT s1s1 = s1*s1;
571 
572     const RealT num = (6*s3 - (3*(2*s2 - s1s1) + s1s1)*s1);
573     const RealT den = (2*s2 - s1s1);
574 
575     return num / pow(den, static_cast<RealT>(1.5));
576 }
577 
578 template <typename RealT, typename PolicyT>
579 RealT kurtosis(hyperexponential_distribution<RealT,PolicyT> const& dist)
580 {
581     const std::size_t n = dist.num_phases();
582     const std::vector<RealT> probs = dist.probabilities();
583     const std::vector<RealT> rates = dist.rates();
584 
585     RealT s1 = 0; // \sum_{i=1}^n \frac{p_i}{\lambda_i}
586     RealT s2 = 0; // \sum_{i=1}^n \frac{p_i}{\lambda_i^2}
587     RealT s3 = 0; // \sum_{i=1}^n \frac{p_i}{\lambda_i^3}
588     RealT s4 = 0; // \sum_{i=1}^n \frac{p_i}{\lambda_i^4}
589     for (std::size_t i = 0; i < n; ++i)
590     {
591         const RealT p = probs[i];
592         const RealT r = rates[i];
593         const RealT r2 = r*r;
594         const RealT r3 = r2*r;
595         const RealT r4 = r3*r;
596 
597         s1 += p/r;
598         s2 += p/r2;
599         s3 += p/r3;
600         s4 += p/r4;
601     }
602 
603     const RealT s1s1 = s1*s1;
604 
605     const RealT num = (24*s4 - 24*s3*s1 + 3*(2*(2*s2 - s1s1) + s1s1)*s1s1);
606     const RealT den = (2*s2 - s1s1);
607 
608     return num/(den*den);
609 }
610 
611 template <typename RealT, typename PolicyT>
612 RealT kurtosis_excess(hyperexponential_distribution<RealT,PolicyT> const& dist)
613 {
614     return kurtosis(dist) - 3;
615 }
616 
617 template <typename RealT, typename PolicyT>
618 RealT mode(hyperexponential_distribution<RealT,PolicyT> const& /*dist*/)
619 {
620     return 0;
621 }
622 
623 }} // namespace boost::math
624 
625 #ifdef BOOST_MSVC
626 #pragma warning (pop)
627 #endif
628 // This include must be at the end, *after* the accessors
629 // for this distribution have been defined, in order to
630 // keep compilers that support two-phase lookup happy.
631 #include <boost/math/distributions/detail/derived_accessors.hpp>
632 #include <boost/math/distributions/detail/generic_quantile.hpp>
633 
634 #endif // BOOST_MATH_DISTRIBUTIONS_HYPEREXPONENTIAL
635