| 1 |
/* |
| 2 |
* Copyright Nick Thompson, 2019 |
| 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 |
|
| 8 |
#ifndef BOOST_MATH_STATISTICS_RUNS_TEST_HPP |
| 9 |
#define BOOST_MATH_STATISTICS_RUNS_TEST_HPP |
| 10 |
|
| 11 |
#include <cmath> |
| 12 |
#include <algorithm> |
| 13 |
#include <utility> |
| 14 |
#include <boost/math/statistics/univariate_statistics.hpp> |
| 15 |
#include <boost/math/distributions/normal.hpp> |
| 16 |
|
| 17 |
namespace boost::math::statistics { |
| 18 |
|
| 19 |
template<class RandomAccessContainer> |
| 20 |
auto runs_above_and_below_threshold(RandomAccessContainer const & v, |
| 21 |
typename RandomAccessContainer::value_type threshold) |
| 22 |
{ |
| 23 |
using Real = typename RandomAccessContainer::value_type; |
| 24 |
using std::sqrt; |
| 25 |
using std::abs; |
| 26 |
if (v.size() <= 1) |
| 27 |
{ |
| 28 |
throw std::domain_error("At least 2 samples are required to get number of runs."); |
| 29 |
} |
| 30 |
typedef boost::math::policies::policy< |
| 31 |
boost::math::policies::promote_float<false>, |
| 32 |
boost::math::policies::promote_double<false> > |
| 33 |
no_promote_policy; |
| 34 |
|
| 35 |
decltype(v.size()) nabove = 0; |
| 36 |
decltype(v.size()) nbelow = 0; |
| 37 |
|
| 38 |
decltype(v.size()) imin = 0; |
| 39 |
|
| 40 |
// Take care of the case that v[0] == threshold: |
| 41 |
while (imin < v.size() && v[imin] == threshold) { |
| 42 |
++imin; |
| 43 |
} |
| 44 |
|
| 45 |
// Take care of the constant vector case: |
| 46 |
if (imin == v.size()) { |
| 47 |
return std::make_pair(std::numeric_limits<Real>::quiet_NaN(), Real(0)); |
| 48 |
} |
| 49 |
|
| 50 |
bool run_up = (v[imin] > threshold); |
| 51 |
if (run_up) { |
| 52 |
++nabove; |
| 53 |
} else { |
| 54 |
++nbelow; |
| 55 |
} |
| 56 |
decltype(v.size()) runs = 1; |
| 57 |
for (decltype(v.size()) i = imin + 1; i < v.size(); ++i) { |
| 58 |
if (v[i] == threshold) { |
| 59 |
// skip values precisely equal to threshold (following R's randtests package) |
| 60 |
continue; |
| 61 |
} |
| 62 |
bool above = (v[i] > threshold); |
| 63 |
if (above) { |
| 64 |
++nabove; |
| 65 |
} else { |
| 66 |
++nbelow; |
| 67 |
} |
| 68 |
if (run_up == above) { |
| 69 |
continue; |
| 70 |
} |
| 71 |
else { |
| 72 |
run_up = above; |
| 73 |
runs++; |
| 74 |
} |
| 75 |
} |
| 76 |
|
| 77 |
// If you make n an int, the subtraction is gonna be bad in the variance: |
| 78 |
Real n = nabove + nbelow; |
| 79 |
|
| 80 |
Real expected_runs = Real(1) + Real(2*nabove*nbelow)/Real(n); |
| 81 |
Real variance = 2*nabove*nbelow*(2*nabove*nbelow-n)/Real(n*n*(n-1)); |
| 82 |
|
| 83 |
// Bizarre, pathological limits: |
| 84 |
if (variance == 0) |
| 85 |
{ |
| 86 |
if (runs == expected_runs) |
| 87 |
{ |
| 88 |
Real statistic = 0; |
| 89 |
Real pvalue = 1; |
| 90 |
return std::make_pair(statistic, pvalue); |
| 91 |
} |
| 92 |
else |
| 93 |
{ |
| 94 |
return std::make_pair(std::numeric_limits<Real>::quiet_NaN(), Real(0)); |
| 95 |
} |
| 96 |
} |
| 97 |
|
| 98 |
Real sd = sqrt(variance); |
| 99 |
Real statistic = (runs - expected_runs)/sd; |
| 100 |
|
| 101 |
auto normal = boost::math::normal_distribution<Real, no_promote_policy>(0,1); |
| 102 |
Real pvalue = 2*boost::math::cdf(normal, -abs(statistic)); |
| 103 |
return std::make_pair(statistic, pvalue); |
| 104 |
} |
| 105 |
|
| 106 |
template<class RandomAccessContainer> |
| 107 |
auto runs_above_and_below_median(RandomAccessContainer const & v) |
| 108 |
{ |
| 109 |
using Real = typename RandomAccessContainer::value_type; |
| 110 |
using std::log; |
| 111 |
using std::sqrt; |
| 112 |
|
| 113 |
// We have to memcpy v because the median does a partial sort, |
| 114 |
// and that would be catastrophic for the runs test. |
| 115 |
auto w = v; |
| 116 |
Real median = boost::math::statistics::median(w); |
| 117 |
return runs_above_and_below_threshold(v, median); |
| 118 |
} |
| 119 |
|
| 120 |
} |
| 121 |
#endif |