| 1 | /////////////////////////////////////////////////////////////////////////////// |
| 2 | // weighted_variance.hpp |
| 3 | // |
| 4 | // Copyright 2005 Daniel Egloff, Eric Niebler. Distributed under the Boost |
| 5 | // Software License, Version 1.0. (See accompanying file |
| 6 | // LICENSE_1_0.txt or copy at http://www.boost.org/LICENSE_1_0.txt) |
| 7 | |
| 8 | #ifndef BOOST_ACCUMULATORS_STATISTICS_WEIGHTED_VARIANCE_HPP_EAN_28_10_2005 |
| 9 | #define BOOST_ACCUMULATORS_STATISTICS_WEIGHTED_VARIANCE_HPP_EAN_28_10_2005 |
| 10 | |
| 11 | #include <boost/mpl/placeholders.hpp> |
| 12 | #include <boost/accumulators/framework/accumulator_base.hpp> |
| 13 | #include <boost/accumulators/framework/extractor.hpp> |
| 14 | #include <boost/accumulators/numeric/functional.hpp> |
| 15 | #include <boost/accumulators/framework/parameters/sample.hpp> |
| 16 | #include <boost/accumulators/framework/depends_on.hpp> |
| 17 | #include <boost/accumulators/statistics_fwd.hpp> |
| 18 | #include <boost/accumulators/statistics/count.hpp> |
| 19 | #include <boost/accumulators/statistics/variance.hpp> |
| 20 | #include <boost/accumulators/statistics/weighted_sum.hpp> |
| 21 | #include <boost/accumulators/statistics/weighted_mean.hpp> |
| 22 | #include <boost/accumulators/statistics/weighted_moment.hpp> |
| 23 | |
| 24 | namespace boost { namespace accumulators |
| 25 | { |
| 26 | |
| 27 | namespace impl |
| 28 | { |
| 29 | //! Lazy calculation of variance of weighted samples. |
| 30 | /*! |
| 31 | The default implementation of the variance of weighted samples is based on the second moment |
| 32 | \f$\widehat{m}_n^{(2)}\f$ (weighted_moment<2>) and the mean\f$ \hat{\mu}_n\f$ (weighted_mean): |
| 33 | \f[ |
| 34 | \hat{\sigma}_n^2 = \widehat{m}_n^{(2)}-\hat{\mu}_n^2, |
| 35 | \f] |
| 36 | where \f$n\f$ is the number of samples. |
| 37 | */ |
| 38 | template<typename Sample, typename Weight, typename MeanFeature> |
| 39 | struct lazy_weighted_variance_impl |
| 40 | : accumulator_base |
| 41 | { |
| 42 | typedef typename numeric::functional::multiplies<Sample, Weight>::result_type weighted_sample; |
| 43 | // for boost::result_of |
| 44 | typedef typename numeric::functional::fdiv<weighted_sample, Weight>::result_type result_type; |
| 45 | |
| 46 | lazy_weighted_variance_impl(dont_care) {} |
| 47 | |
| 48 | template<typename Args> |
| 49 | result_type result(Args const &args) const |
| 50 | { |
| 51 | extractor<MeanFeature> const some_mean = {}; |
| 52 | result_type tmp = some_mean(args); |
| 53 | return accumulators::weighted_moment<2>(args) - tmp * tmp; |
| 54 | } |
| 55 | }; |
| 56 | |
| 57 | //! Iterative calculation of variance of weighted samples. |
| 58 | /*! |
| 59 | Iterative calculation of variance of weighted samples: |
| 60 | \f[ |
| 61 | \hat{\sigma}_n^2 = |
| 62 | \frac{\bar{w}_n - w_n}{\bar{w}_n}\hat{\sigma}_{n - 1}^2 |
| 63 | + \frac{w_n}{\bar{w}_n - w_n}\left(X_n - \hat{\mu}_n\right)^2 |
| 64 | ,\quad n\ge2,\quad\hat{\sigma}_0^2 = 0. |
| 65 | \f] |
| 66 | where \f$\bar{w}_n\f$ is the sum of the \f$n\f$ weights \f$w_i\f$ and \f$\hat{\mu}_n\f$ |
| 67 | the estimate of the mean of the weighted samples. Note that the sample variance is not defined for |
| 68 | \f$n <= 1\f$. |
| 69 | */ |
| 70 | template<typename Sample, typename Weight, typename MeanFeature, typename Tag> |
| 71 | struct weighted_variance_impl |
| 72 | : accumulator_base |
| 73 | { |
| 74 | typedef typename numeric::functional::multiplies<Sample, Weight>::result_type weighted_sample; |
| 75 | // for boost::result_of |
| 76 | typedef typename numeric::functional::fdiv<weighted_sample, Weight>::result_type result_type; |
| 77 | |
| 78 | template<typename Args> |
| 79 | weighted_variance_impl(Args const &args) |
| 80 | : weighted_variance(numeric::fdiv(args[sample | Sample()], numeric::one<Weight>::value)) |
| 81 | { |
| 82 | } |
| 83 | |
| 84 | template<typename Args> |
| 85 | void operator ()(Args const &args) |
| 86 | { |
| 87 | std::size_t cnt = count(args); |
| 88 | |
| 89 | if(cnt > 1) |
| 90 | { |
| 91 | extractor<MeanFeature> const some_mean = {}; |
| 92 | |
| 93 | result_type tmp = args[parameter::keyword<Tag>::get()] - some_mean(args); |
| 94 | |
| 95 | this->weighted_variance = |
| 96 | numeric::fdiv(this->weighted_variance * (sum_of_weights(args) - args[weight]), sum_of_weights(args)) |
| 97 | + numeric::fdiv(tmp * tmp * args[weight], sum_of_weights(args) - args[weight] ); |
| 98 | } |
| 99 | } |
| 100 | |
| 101 | result_type result(dont_care) const |
| 102 | { |
| 103 | return this->weighted_variance; |
| 104 | } |
| 105 | |
| 106 | // make this accumulator serializeable |
| 107 | template<class Archive> |
| 108 | void serialize(Archive & ar, const unsigned int file_version) |
| 109 | { |
| 110 | ar & weighted_variance; |
| 111 | } |
| 112 | |
| 113 | private: |
| 114 | result_type weighted_variance; |
| 115 | }; |
| 116 | |
| 117 | } // namespace impl |
| 118 | |
| 119 | /////////////////////////////////////////////////////////////////////////////// |
| 120 | // tag::weighted_variance |
| 121 | // tag::immediate_weighted_variance |
| 122 | // |
| 123 | namespace tag |
| 124 | { |
| 125 | struct lazy_weighted_variance |
| 126 | : depends_on<weighted_moment<2>, weighted_mean> |
| 127 | { |
| 128 | /// INTERNAL ONLY |
| 129 | /// |
| 130 | typedef accumulators::impl::lazy_weighted_variance_impl<mpl::_1, mpl::_2, weighted_mean> impl; |
| 131 | }; |
| 132 | |
| 133 | struct weighted_variance |
| 134 | : depends_on<count, immediate_weighted_mean> |
| 135 | { |
| 136 | /// INTERNAL ONLY |
| 137 | /// |
| 138 | typedef accumulators::impl::weighted_variance_impl<mpl::_1, mpl::_2, immediate_weighted_mean, sample> impl; |
| 139 | }; |
| 140 | } |
| 141 | |
| 142 | /////////////////////////////////////////////////////////////////////////////// |
| 143 | // extract::weighted_variance |
| 144 | // extract::immediate_weighted_variance |
| 145 | // |
| 146 | namespace extract |
| 147 | { |
| 148 | extractor<tag::lazy_weighted_variance> const = {}; |
| 149 | extractor<tag::weighted_variance> const = {}; |
| 150 | |
| 151 | BOOST_ACCUMULATORS_IGNORE_GLOBAL(lazy_weighted_variance) |
| 152 | BOOST_ACCUMULATORS_IGNORE_GLOBAL(weighted_variance) |
| 153 | } |
| 154 | |
| 155 | using extract::lazy_weighted_variance; |
| 156 | using extract::weighted_variance; |
| 157 | |
| 158 | // weighted_variance(lazy) -> lazy_weighted_variance |
| 159 | template<> |
| 160 | struct as_feature<tag::weighted_variance(lazy)> |
| 161 | { |
| 162 | typedef tag::lazy_weighted_variance type; |
| 163 | }; |
| 164 | |
| 165 | // weighted_variance(immediate) -> weighted_variance |
| 166 | template<> |
| 167 | struct as_feature<tag::weighted_variance(immediate)> |
| 168 | { |
| 169 | typedef tag::weighted_variance type; |
| 170 | }; |
| 171 | |
| 172 | //////////////////////////////////////////////////////////////////////////// |
| 173 | //// droppable_accumulator<weighted_variance_impl> |
| 174 | //// need to specialize droppable lazy weighted_variance to cache the result at the |
| 175 | //// point the accumulator is dropped. |
| 176 | ///// INTERNAL ONLY |
| 177 | ///// |
| 178 | //template<typename Sample, typename Weight, typename MeanFeature> |
| 179 | //struct droppable_accumulator<impl::weighted_variance_impl<Sample, Weight, MeanFeature> > |
| 180 | // : droppable_accumulator_base< |
| 181 | // with_cached_result<impl::weighted_variance_impl<Sample, Weight, MeanFeature> > |
| 182 | // > |
| 183 | //{ |
| 184 | // template<typename Args> |
| 185 | // droppable_accumulator(Args const &args) |
| 186 | // : droppable_accumulator::base(args) |
| 187 | // { |
| 188 | // } |
| 189 | //}; |
| 190 | |
| 191 | }} // namespace boost::accumulators |
| 192 | |
| 193 | #endif |
| 194 | |