2 Copyright 2018 Nick Thompson
4 Distributed under the Boost Software License, Version 1.0.
5 (See accompanying file LICENSE_1_0.txt or copy at
6 http://www.boost.org/LICENSE_1_0.txt).
9 [section:univariate_statistics Univariate Statistics]
14 #include <boost/math/tools/univariate_statistics.hpp>
16 namespace boost{ namespace math{ namespace tools {
18 template<class Container>
19 auto mean(Container const & c);
21 template<class ForwardIterator>
22 auto mean(ForwardIterator first, ForwardIterator last);
24 template<class Container>
25 auto variance(Container const & c);
27 template<class ForwardIterator>
28 auto variance(ForwardIterator first, ForwardIterator last);
30 template<class Container>
31 auto sample_variance(Container const & c);
33 template<class ForwardIterator>
34 auto sample_variance(ForwardIterator first, ForwardIterator last);
36 template<class Container>
37 auto skewness(Container const & c);
39 template<class ForwardIterator>
40 auto skewness(ForwardIterator first, ForwardIterator last);
42 template<class Container>
43 auto kurtosis(Container const & c);
45 template<class ForwardIterator>
46 auto kurtosis(ForwardIterator first, ForwardIterator last);
48 template<class Container>
49 auto excess_kurtosis(Container const & c);
51 template<class ForwardIterator>
52 auto excess_kurtosis(ForwardIterator first, ForwardIterator last);
54 template<class Container>
55 auto first_four_moments(Container const & c);
57 template<class ForwardIterator>
58 auto first_four_moments(ForwardIterator first, ForwardIterator last);
60 template<class Container>
61 auto median(Container & c);
63 template<class ForwardIterator>
64 auto median(ForwardIterator first, ForwardIterator last);
66 template<class RandomAccessIterator>
67 auto median_absolute_deviation(ForwardIterator first, ForwardIterator last, typename std::iterator_traits<RandomAccessIterator>::value_type center=std::numeric_limits<Real>::quiet_NaN());
69 template<class RandomAccessContainer>
70 auto median_absolute_deviation(RandomAccessContainer v, typename RandomAccessContainer::value_type center=std::numeric_limits<Real>::quiet_NaN());
72 template<class Container>
73 auto gini_coefficient(Container & c);
75 template<class ForwardIterator>
76 auto gini_coefficient(ForwardIterator first, ForwardIterator last);
78 template<class Container>
79 auto sample_gini_coefficient(Container & c);
81 template<class ForwardIterator>
82 auto sample_gini_coefficient(ForwardIterator first, ForwardIterator last);
89 The file `boost/math/tools/univariate_statistics.hpp` is a set of facilities for computing scalar values from vectors.
91 Many of these functionals have trivial naive implementations, but experienced programmers will recognize that even trivial algorithms are easy to screw up, and that numerical instabilities often lurk in corner cases.
92 We have attempted to do our "due diligence" to root out these problems-scouring the literature for numerically stable algorithms for even the simplest of functionals.
94 /Nota bene/: Some similar functionality is provided in [@https://www.boost.org/doc/libs/1_68_0/doc/html/accumulators/user_s_guide.html Boost Accumulators Framework].
95 These accumulators should be used in real-time applications; `univariate_statistics.hpp` should be used when CPU vectorization is needed.
96 As a reminder, remember that to actually /get/ vectorization, compile with `-march=native -O3` flags.
98 We now describe each functional in detail.
99 Our examples use `std::vector<double>` to hold the data, but this not required.
100 In general, you can store your data in an Eigen array, and Armadillo vector, `std::array`, and for many of the routines, a `std::forward_list`.
101 These routines are usable in float, double, long double, and Boost.Multiprecision precision, as well as their complex extensions whenever the computation is well-defined.
102 For certain operations (total variation, for example) integer inputs are supported.
106 std::vector<double> v{1,2,3,4,5};
107 double mu = boost::math::tools::mean(v.cbegin(), v.cend());
108 // Alternative syntax if you want to use entire container:
109 mu = boost::math::tools::mean(v);
111 The implementation follows [@https://doi.org/10.1137/1.9780898718027 Higham 1.6a].
112 The data is not modified and must be forward iterable.
113 Works with real and integer data.
114 If the input is an integer type, the output is a double precision float.
118 std::vector<double> v{1,2,3,4,5};
119 Real sigma_sq = boost::math::tools::variance(v.cbegin(), v.cend());
121 If you don't need to calculate on a subset of the input, then the range call is more terse:
123 std::vector<double> v{1,2,3,4,5};
124 Real sigma_sq = boost::math::tools::variance(v);
126 The implementation follows [@https://doi.org/10.1137/1.9780898718027 Higham 1.6b].
127 The input data must be forward iterable and the range `[first, last)` must contain at least two elements.
128 It is /not/ in general sensible to pass complex numbers to this routine.
129 If integers are passed as input, then the output is a double precision float.
131 `boost::math::tools::variance` returns the population variance.
132 If you want a sample variance, use
134 std::vector<double> v{1,2,3,4,5};
135 Real sn_sq = boost::math::tools::sample_variance(v);
140 Computes the skewness of a dataset:
142 std::vector<double> v{1,2,3,4,5};
143 double skewness = boost::math::tools::skewness(v);
146 The input vector is not modified, works with integral and real data.
147 If the input data is integral, the output is a double precision float.
149 For a dataset consisting of a single constant value, we take the skewness to be zero by definition.
151 The implementation follows [@https://prod.sandia.gov/techlib-noauth/access-control.cgi/2008/086212.pdf Pebay].
155 Computes the kurtosis of a dataset:
157 std::vector<double> v{1,2,3,4,5};
158 double kurtosis = boost::math::tools::kurtosis(v);
161 The implementation follows [@https://prod.sandia.gov/techlib-noauth/access-control.cgi/2008/086212.pdf Pebay].
162 The input data must be forward iterable and must consist of real or integral values.
163 If the input data is integral, the output is a double precision float.
164 Note that this is /not/ the excess kurtosis.
165 If you require the excess kurtosis, use `boost::math::tools::excess_kurtosis`.
166 This function simply subtracts 3 from the kurtosis, but it makes eminently clear our definition of kurtosis.
168 [heading First four moments]
170 Simultaneously computes the first four [@https://en.wikipedia.org/wiki/Central_moment central moments] in a single pass through the data:
172 std::vector<double> v{1,2,3,4,5};
173 auto [M1, M2, M3, M4] = boost::math::tools::first_four_moments(v);
178 Computes the median of a dataset:
180 std::vector<double> v{1,2,3,4,5};
181 double m = boost::math::tools::median(v.begin(), v.end());
183 /Nota bene: The input vector is modified./
184 The calculation of the median is a thin wrapper around the C++11 [@https://en.cppreference.com/w/cpp/algorithm/nth_element `nth_element`].
185 Therefore, all requirements of `std::nth_element` are inherited by the median calculation.
186 In particular, the container must allow random access.
188 [heading Median Absolute Deviation]
190 Computes the [@https://en.wikipedia.org/wiki/Median_absolute_deviation median absolute deviation] of a dataset:
192 std::vector<double> v{1,2,3,4,5};
193 double mad = boost::math::tools::median_absolute_deviation(v);
195 By default, the deviation from the median is used.
196 If you have some prior that the median is zero, or wish to compute the median absolute deviation from the mean,
199 // prior is that center is zero:
201 double mad = boost::math::tools::median_absolute_deviation(v, center);
203 // compute median absolute deviation from the mean:
204 double mu = boost::math::tools::mean(v);
205 double mad = boost::math::tools::median_absolute_deviation(v, mu);
207 /Nota bene:/ The input vector is modified.
208 Again the vector is passed into a call to [@https://en.cppreference.com/w/cpp/algorithm/nth_element `nth_element`].
210 [heading Gini Coefficient]
212 Compute the Gini coefficient of a dataset:
214 std::vector<double> v{1,0,0,0};
215 double gini = boost::math::tools::gini_coefficient(v);
217 double s_gini = boost::math::tools::sample_gini_coefficient(v);
219 std::vector<double> w{1,1,1,1};
220 gini = boost::math::tools::gini_coefficient(w.begin(), w.end());
221 // gini = 0, as all elements are now equal.
223 /Nota bene/: The input data is altered: in particular, it is sorted. Makes a call to `std::sort`, and as such requires random access iterators.
225 The sample Gini coefficient lies in the range [0,1], whereas the population Gini coefficient is in the range [0, 1 - 1/ /n/].
227 /Nota bene:/ There is essentially no reason to pass negative values to the Gini coefficient function.
228 However, a use case (measuring wealth inequality when some people have negative wealth) exists, so we do not throw an exception when negative values are encountered.
229 You should have /very/ good cause to pass negative values to the Gini coefficient calculator.
230 Another use case is found in signal processing, but the sorting is by magnitude and hence has a different implementation.
231 See `absolute_gini_coefficient` for details.
235 * Higham, Nicholas J. ['Accuracy and stability of numerical algorithms.] Vol. 80. Siam, 2002.
236 * Philippe P. Pébay: ["Formulas for Robust, One-Pass Parallel Computation of Covariances and Arbitrary-Order Statistical Moments.] Technical Report SAND2008-6212, Sandia National Laboratories, September 2008.
239 [/section:univariate_statistics Univariate Statistics]