Empirical Cumulative Distribution function

doc/distributions/dist_reference.qbk

@@ -37,6 +37,7 @@
 [include triangular.qbk]
 [include uniform.qbk]
 [include weibull.qbk]
+[include empirical_cdf.qbk]
 
 [endsect] [/section:dists Distributions]
 
@@ -138,10 +139,3 @@ opportunity to integrate the statistical tests with this framework at some later
   (See accompanying file LICENSE_1_0.txt or copy at
   http://www.boost.org/LICENSE_1_0.txt).
 ]
-
-
-
-
-
-
-

doc/distributions/empirical_cdf.qbk

@@ -0,0 +1,71 @@
+[/
+Copyright (c) 2019 Nick Thompson
+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)
+]
+
+[section:empirical_cdf Empirical Cumulative Distribution Function]
+
+[heading Synopsis]
+
+```
+#include <boost/math/distributions/empirical_cumulative_distribution_function.hpp>
+
+namespace boost{ namespace math{
+
+template <class RandomAccessContainer>
+class empirical_cumulative_distribution_function
+{
+public:
+    using Real = typename RandomAccessContainer::value_type;
+    empirical_cumulative_distribution_function(RandomAccessContainer && v);
+
+    auto operator()(Real t) const;
+};
+
+}}
+```
+
+[heading Empirical Cumulative Distribution Function]
+
+The empirical cumulative distribution function is a step function constructed from observed data which converges to the true cumulative distribution function in the limit of infinite data.
+This function is a basic building block of hypothesis testing workflows that attempt to answer the question "does my data come from a given distribution?"
+These tests require computing quadratures over some function of the empirical CDF and the supposed CDF to create a distance measurement, and hence it is occasionally useful to construct a continuous callable from the data.
+
+An example usage is demonstrated below:
+
+```
+#include <vector>
+#include <random>
+#include <boost/math/distributions/empirical_cumulative_distribution_function.hpp>
+using boost::math::empirical_cumulative_distribution_function;
+std::random_device rd;
+std::mt19937 gen{rd()};
+std::normal_distribution<double> dis(0, 1);
+size_t n = 128;
+std::vector<double> v(n);
+for (size_t i = 0; i < n; ++i) {
+  v[i] = dis(gen);
+}
+
+auto ecdf = empirical_cumulative_distribution_function(std::move(v));
+std::cout << "ecdf(0.0) = " << ecdf(0.0) << "\n";
+// should print approximately 0.5 . . .
+```
+
+The empirical distribution function operates on sorted data.
+If the data are not already sorted, the constructor sorts it for you at O(Nlog(N)) cost.
+
+Call operator complexity is O(log(N)).
+
+Works with both integer and floating point types.
+If the input data consists of integers, the output of the call operator is a double. Requires C++17.
+
+[$../graphs/empiricial_cumulative_distribution_gauss.svg]
+
+[$../graphs/empiricial_cumulative_distribution_uniform.svg]
+
+
+[endsect]
+[/section:empirical_cdf]

doc/graphs/empiricial_cumulative_distribution_gauss.svg

@@ -2,7 +2,7 @@
 <svg xmlns='http://www.w3.org/2000/svg' width='1100' height='679'>
 <style>svg { background-color: black; }
 </style>
-<text x='550' y='20' font-family='Palatino' font-size='25' fill='white'  alignment-baseline='middle' text-anchor='middle'>Empirical (blue) and continuous CDF (orange) of an 𝓝(0,1) Gaussian distribution for n = 128 samples</text>
+<text x='550' y='20' font-family='Palatino' font-size='25' fill='white'  alignment-baseline='middle' text-anchor='middle'>Empirical (blue) and continuous CDF (orange) of an 𝓝(0,1) distribution on 128 samples</text>
 <g transform='translate(25, 40)'>
 <line x1='0' y1='0' x2='0' y2='619' stroke='gray' stroke-width='1' />
 <line x1='0' y1='619' x2='1055' y2='619' stroke='gray' stroke-width='1' />

doc/graphs/empiricial_cumulative_distribution_uniform.svg

@@ -2,7 +2,7 @@
 <svg xmlns='http://www.w3.org/2000/svg' width='1100' height='679'>
 <style>svg { background-color: black; }
 </style>
-<text x='550' y='20' font-family='Palatino' font-size='25' fill='white'  alignment-baseline='middle' text-anchor='middle'>Empirical (blue) and theoretical CDF (orange) of an dice roll distribution on n = 128 samples</text>
+<text x='550' y='20' font-family='Palatino' font-size='25' fill='white'  alignment-baseline='middle' text-anchor='middle'>Empirical (blue) and theoretical CDF (orange) of the dice roll distribution on n = 128 samples</text>
 <g transform='translate(25, 40)'>
 <line x1='0' y1='0' x2='0' y2='619' stroke='gray' stroke-width='1' />
 <line x1='0' y1='619' x2='1055' y2='619' stroke='gray' stroke-width='1' />

include/boost/math/distributions/empirical_cumulative_distribution_function.hpp

@@ -6,6 +6,7 @@
 #ifndef BOOST_MATH_DISTRIBUTIONS_EMPIRICAL_CUMULATIVE_DISTRIBUTION_FUNCTION_HPP
 #define BOOST_MATH_DISTRIBUTIONS_EMPIRICAL_CUMULATIVE_DISTRIBUTION_FUNCTION_HPP
 #include <algorithm>
+#include <iterator>
 
 namespace boost { namespace math{
 
@@ -22,7 +23,8 @@ public:
     }
 
     auto operator()(Real x) const {
-       if constexpr (std::is_integral_v<Real>) {
+       if constexpr (std::is_integral_v<Real>)
+       {
          if (x < m_v[0]) {
            return double(0);
          }
@@ -32,15 +34,16 @@ public:
          auto it = std::upper_bound(m_v.begin(), m_v.end(), x);
          return static_cast<double>(std::distance(m_v.begin(), it))/static_cast<double>(m_v.size());
        }
-       else {
+       else
-       if (x < m_v[0]) {
+       {
-         return Real(0);
+         if (x < m_v[0]) {
-       }
+           return Real(0);
-       if (x >= m_v[m_v.size()-1]) {
+         }
-         return Real(1);
+         if (x >= m_v[m_v.size()-1]) {
-       }
+           return Real(1);
-       auto it = std::upper_bound(m_v.begin(), m_v.end(), x);
+         }
-       return static_cast<Real>(std::distance(m_v.begin(), it))/static_cast<Real>(m_v.size());
+         auto it = std::upper_bound(m_v.begin(), m_v.end(), x);
+         return static_cast<Real>(std::distance(m_v.begin(), it))/static_cast<Real>(m_v.size());
       }
     }
 

test/Jamfile.v2

@@ -951,6 +951,7 @@ test-suite misc :
    [ run compile_test/catmull_rom_concept_test.cpp compile_test_main   : : : [ requires cxx11_hdr_array cxx11_hdr_initializer_list ] ]
    [ run ooura_fourier_integral_test.cpp ../../test/build//boost_unit_test_framework : : :  [ requires cxx17_if_constexpr cxx17_std_apply ] ]
    [ run univariate_statistics_test.cpp ../../test/build//boost_unit_test_framework : : :  [ requires cxx17_if_constexpr cxx17_std_apply ] ]
+   [ run empirical_cumulative_distribution_test.cpp  : : :  [ requires cxx17_if_constexpr cxx17_std_apply ] ]
    [ run norms_test.cpp ../../test/build//boost_unit_test_framework : : :  [ requires cxx17_if_constexpr cxx17_std_apply ] ]
    [ run signal_statistics_test.cpp : : : [ requires cxx17_if_constexpr cxx17_std_apply ] ]
    [ run bivariate_statistics_test.cpp : : : [ requires cxx17_if_constexpr cxx17_std_apply ] ]

test/empirical_cumulative_distribution_test.cpp

@@ -16,18 +16,54 @@
 using boost::multiprecision::float128;
 #endif
 
-using boost::math::distributions::empirical_cumulative_distribution_function;
+using boost::math::empirical_cumulative_distribution_function;
+
+template<class Z>
+void test_uniform_z()
+{
+    std::vector<Z> v{6,3,4,1,2,5};
+
+    auto ecdf = empirical_cumulative_distribution_function(std::move(v));
+
+    CHECK_ULP_CLOSE(1.0/6.0, ecdf(1), 1);
+    CHECK_ULP_CLOSE(2.0/6.0, ecdf(2), 1);
+    CHECK_ULP_CLOSE(3.0/6.0, ecdf(3), 1);
+    CHECK_ULP_CLOSE(4.0/6.0, ecdf(4), 1);
+    CHECK_ULP_CLOSE(5.0/6.0, ecdf(5), 1);
+    CHECK_ULP_CLOSE(6.0/6.0, ecdf(6), 1);
+
+    // Less trivial:
+
+    v = {6,3,4,1,1,1,2,4};
+    ecdf = empirical_cumulative_distribution_function(std::move(v));
+    CHECK_ULP_CLOSE(3.0/8.0, ecdf(1), 1);
+    CHECK_ULP_CLOSE(4.0/8.0, ecdf(2), 1);
+    CHECK_ULP_CLOSE(5.0/8.0, ecdf(3), 1);
+    CHECK_ULP_CLOSE(7.0/8.0, ecdf(4), 1);
+    CHECK_ULP_CLOSE(7.0/8.0, ecdf(5), 1);
+    CHECK_ULP_CLOSE(8.0/8.0, ecdf(6), 1);
+}
 
 template<class Real>
 void test_uniform()
 {
+    size_t n = 128;
+    std::vector<Real> v(n);
+    for (size_t i = 0; i < n; ++i) {
+      v[i] = Real(i+1)/Real(n);
+    }
 
+    auto ecdf = empirical_cumulative_distribution_function(std::move(v));
+
+    for (size_t i = 0; i < n; ++i) {
+      CHECK_ULP_CLOSE(Real(i+1)/Real(n), ecdf(Real(i+1)/Real(n)), 1);
+    }
 }
 
 
 int main()
 {
-
+    test_uniform_z<int>();
-    test_uniform<float>();
+    test_uniform<double>();
     return boost::math::test::report_errors();
 }