]> git.donarmstrong.com Git - rsem.git/blobdiff - boost/random/piecewise_constant_distribution.hpp
Updated boost to v1.55.0
[rsem.git] / boost / random / piecewise_constant_distribution.hpp
diff --git a/boost/random/piecewise_constant_distribution.hpp b/boost/random/piecewise_constant_distribution.hpp
new file mode 100644 (file)
index 0000000..21e9c1c
--- /dev/null
@@ -0,0 +1,466 @@
+/* boost random/piecewise_constant_distribution.hpp header file
+ *
+ * Copyright Steven Watanabe 2011
+ * Distributed under 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)
+ *
+ * See http://www.boost.org for most recent version including documentation.
+ *
+ * $Id: piecewise_constant_distribution.hpp 85813 2013-09-21 20:17:00Z jewillco $
+ */
+
+#ifndef BOOST_RANDOM_PIECEWISE_CONSTANT_DISTRIBUTION_HPP_INCLUDED
+#define BOOST_RANDOM_PIECEWISE_CONSTANT_DISTRIBUTION_HPP_INCLUDED
+
+#include <vector>
+#include <numeric>
+#include <boost/assert.hpp>
+#include <boost/random/uniform_real.hpp>
+#include <boost/random/discrete_distribution.hpp>
+#include <boost/random/detail/config.hpp>
+#include <boost/random/detail/operators.hpp>
+#include <boost/random/detail/vector_io.hpp>
+
+#ifndef BOOST_NO_CXX11_HDR_INITIALIZER_LIST
+#include <initializer_list>
+#endif
+
+#include <boost/range/begin.hpp>
+#include <boost/range/end.hpp>
+
+namespace boost {
+namespace random {
+
+/**
+ * The class @c piecewise_constant_distribution models a \random_distribution.
+ */
+template<class RealType = double, class WeightType = double>
+class piecewise_constant_distribution {
+public:
+    typedef std::size_t input_type;
+    typedef RealType result_type;
+
+    class param_type {
+    public:
+
+        typedef piecewise_constant_distribution distribution_type;
+
+        /**
+         * Constructs a @c param_type object, representing a distribution
+         * that produces values uniformly distributed in the range [0, 1).
+         */
+        param_type()
+        {
+            _weights.push_back(WeightType(1));
+            _intervals.push_back(RealType(0));
+            _intervals.push_back(RealType(1));
+        }
+        /**
+         * Constructs a @c param_type object from two iterator ranges
+         * containing the interval boundaries and the interval weights.
+         * If there are less than two boundaries, then this is equivalent to
+         * the default constructor and creates a single interval, [0, 1).
+         *
+         * The values of the interval boundaries must be strictly
+         * increasing, and the number of weights must be one less than
+         * the number of interval boundaries.  If there are extra
+         * weights, they are ignored.
+         */
+        template<class IntervalIter, class WeightIter>
+        param_type(IntervalIter intervals_first, IntervalIter intervals_last,
+                   WeightIter weight_first)
+          : _intervals(intervals_first, intervals_last)
+        {
+            if(_intervals.size() < 2) {
+                _intervals.clear();
+                _intervals.push_back(RealType(0));
+                _intervals.push_back(RealType(1));
+                _weights.push_back(WeightType(1));
+            } else {
+                _weights.reserve(_intervals.size() - 1);
+                for(std::size_t i = 0; i < _intervals.size() - 1; ++i) {
+                    _weights.push_back(*weight_first++);
+                }
+            }
+        }
+#ifndef BOOST_NO_CXX11_HDR_INITIALIZER_LIST
+        /**
+         * Constructs a @c param_type object from an
+         * initializer_list containing the interval boundaries
+         * and a unary function specifying the weights.  Each
+         * weight is determined by calling the function at the
+         * midpoint of the corresponding interval.
+         *
+         * If the initializer_list contains less than two elements,
+         * this is equivalent to the default constructor and the
+         * distribution will produce values uniformly distributed
+         * in the range [0, 1).
+         */
+        template<class T, class F>
+        param_type(const std::initializer_list<T>& il, F f)
+          : _intervals(il.begin(), il.end())
+        {
+            if(_intervals.size() < 2) {
+                _intervals.clear();
+                _intervals.push_back(RealType(0));
+                _intervals.push_back(RealType(1));
+                _weights.push_back(WeightType(1));
+            } else {
+                _weights.reserve(_intervals.size() - 1);
+                for(std::size_t i = 0; i < _intervals.size() - 1; ++i) {
+                    RealType midpoint = (_intervals[i] + _intervals[i + 1]) / 2;
+                    _weights.push_back(f(midpoint));
+                }
+            }
+        }
+#endif
+        /**
+         * Constructs a @c param_type object from Boost.Range
+         * ranges holding the interval boundaries and the weights.  If
+         * there are less than two interval boundaries, this is equivalent
+         * to the default constructor and the distribution will produce
+         * values uniformly distributed in the range [0, 1).  The
+         * number of weights must be one less than the number of
+         * interval boundaries.
+         */
+        template<class IntervalRange, class WeightRange>
+        param_type(const IntervalRange& intervals_arg,
+                   const WeightRange& weights_arg)
+          : _intervals(boost::begin(intervals_arg), boost::end(intervals_arg)),
+            _weights(boost::begin(weights_arg), boost::end(weights_arg))
+        {
+            if(_intervals.size() < 2) {
+                _intervals.clear();
+                _intervals.push_back(RealType(0));
+                _intervals.push_back(RealType(1));
+                _weights.push_back(WeightType(1));
+            }
+        }
+
+        /**
+         * Constructs the parameters for a distribution that approximates a
+         * function.  The range of the distribution is [xmin, xmax).  This
+         * range is divided into nw equally sized intervals and the weights
+         * are found by calling the unary function f on the midpoints of the
+         * intervals.
+         */
+        template<class F>
+        param_type(std::size_t nw, RealType xmin, RealType xmax, F f)
+        {
+            std::size_t n = (nw == 0) ? 1 : nw;
+            double delta = (xmax - xmin) / n;
+            BOOST_ASSERT(delta > 0);
+            for(std::size_t k = 0; k < n; ++k) {
+                _weights.push_back(f(xmin + k*delta + delta/2));
+                _intervals.push_back(xmin + k*delta);
+            }
+            _intervals.push_back(xmax);
+        }
+
+        /**  Returns a vector containing the interval boundaries. */
+        std::vector<RealType> intervals() const { return _intervals; }
+
+        /**
+         * Returns a vector containing the probability densities
+         * over all the intervals of the distribution.
+         */
+        std::vector<RealType> densities() const
+        {
+            RealType sum = std::accumulate(_weights.begin(), _weights.end(),
+                                             static_cast<RealType>(0));
+            std::vector<RealType> result;
+            result.reserve(_weights.size());
+            for(std::size_t i = 0; i < _weights.size(); ++i) {
+                RealType width = _intervals[i + 1] - _intervals[i];
+                result.push_back(_weights[i] / (sum * width));
+            }
+            return result;
+        }
+
+        /** Writes the parameters to a @c std::ostream. */
+        BOOST_RANDOM_DETAIL_OSTREAM_OPERATOR(os, param_type, parm)
+        {
+            detail::print_vector(os, parm._intervals);
+            detail::print_vector(os, parm._weights);
+            return os;
+        }
+        
+        /** Reads the parameters from a @c std::istream. */
+        BOOST_RANDOM_DETAIL_ISTREAM_OPERATOR(is, param_type, parm)
+        {
+            std::vector<RealType> new_intervals;
+            std::vector<WeightType> new_weights;
+            detail::read_vector(is, new_intervals);
+            detail::read_vector(is, new_weights);
+            if(is) {
+                parm._intervals.swap(new_intervals);
+                parm._weights.swap(new_weights);
+            }
+            return is;
+        }
+
+        /** Returns true if the two sets of parameters are the same. */
+        BOOST_RANDOM_DETAIL_EQUALITY_OPERATOR(param_type, lhs, rhs)
+        {
+            return lhs._intervals == rhs._intervals
+                && lhs._weights == rhs._weights;
+        }
+        /** Returns true if the two sets of parameters are different. */
+        BOOST_RANDOM_DETAIL_INEQUALITY_OPERATOR(param_type)
+
+    private:
+
+        friend class piecewise_constant_distribution;
+
+        std::vector<RealType> _intervals;
+        std::vector<WeightType> _weights;
+    };
+
+    /**
+     * Creates a new @c piecewise_constant_distribution with
+     * a single interval, [0, 1).
+     */
+    piecewise_constant_distribution()
+    {
+        _intervals.push_back(RealType(0));
+        _intervals.push_back(RealType(1));
+    }
+    /**
+     * Constructs a piecewise_constant_distribution from two iterator ranges
+     * containing the interval boundaries and the interval weights.
+     * If there are less than two boundaries, then this is equivalent to
+     * the default constructor and creates a single interval, [0, 1).
+     *
+     * The values of the interval boundaries must be strictly
+     * increasing, and the number of weights must be one less than
+     * the number of interval boundaries.  If there are extra
+     * weights, they are ignored.
+     *
+     * For example,
+     *
+     * @code
+     * double intervals[] = { 0.0, 1.0, 4.0 };
+     * double weights[] = { 1.0, 1.0 };
+     * piecewise_constant_distribution<> dist(
+     *     &intervals[0], &intervals[0] + 3, &weights[0]);
+     * @endcode
+     *
+     * The distribution has a 50% chance of producing a
+     * value between 0 and 1 and a 50% chance of producing
+     * a value between 1 and 4.
+     */
+    template<class IntervalIter, class WeightIter>
+    piecewise_constant_distribution(IntervalIter first_interval,
+                                    IntervalIter last_interval,
+                                    WeightIter first_weight)
+      : _intervals(first_interval, last_interval)
+    {
+        if(_intervals.size() < 2) {
+            _intervals.clear();
+            _intervals.push_back(RealType(0));
+            _intervals.push_back(RealType(1));
+        } else {
+            std::vector<WeightType> actual_weights;
+            actual_weights.reserve(_intervals.size() - 1);
+            for(std::size_t i = 0; i < _intervals.size() - 1; ++i) {
+                actual_weights.push_back(*first_weight++);
+            }
+            typedef discrete_distribution<std::size_t, WeightType> bins_type;
+            typename bins_type::param_type bins_param(actual_weights);
+            _bins.param(bins_param);
+        }
+    }
+#ifndef BOOST_NO_CXX11_HDR_INITIALIZER_LIST
+    /**
+     * Constructs a piecewise_constant_distribution from an
+     * initializer_list containing the interval boundaries
+     * and a unary function specifying the weights.  Each
+     * weight is determined by calling the function at the
+     * midpoint of the corresponding interval.
+     *
+     * If the initializer_list contains less than two elements,
+     * this is equivalent to the default constructor and the
+     * distribution will produce values uniformly distributed
+     * in the range [0, 1).
+     */
+    template<class T, class F>
+    piecewise_constant_distribution(std::initializer_list<T> il, F f)
+      : _intervals(il.begin(), il.end())
+    {
+        if(_intervals.size() < 2) {
+            _intervals.clear();
+            _intervals.push_back(RealType(0));
+            _intervals.push_back(RealType(1));
+        } else {
+            std::vector<WeightType> actual_weights;
+            actual_weights.reserve(_intervals.size() - 1);
+            for(std::size_t i = 0; i < _intervals.size() - 1; ++i) {
+                RealType midpoint = (_intervals[i] + _intervals[i + 1]) / 2;
+                actual_weights.push_back(f(midpoint));
+            }
+            typedef discrete_distribution<std::size_t, WeightType> bins_type;
+            typename bins_type::param_type bins_param(actual_weights);
+            _bins.param(bins_param);
+        }
+    }
+#endif
+    /**
+     * Constructs a piecewise_constant_distribution from Boost.Range
+     * ranges holding the interval boundaries and the weights.  If
+     * there are less than two interval boundaries, this is equivalent
+     * to the default constructor and the distribution will produce
+     * values uniformly distributed in the range [0, 1).  The
+     * number of weights must be one less than the number of
+     * interval boundaries.
+     */
+    template<class IntervalsRange, class WeightsRange>
+    piecewise_constant_distribution(const IntervalsRange& intervals_arg,
+                                    const WeightsRange& weights_arg)
+      : _bins(weights_arg),
+        _intervals(boost::begin(intervals_arg), boost::end(intervals_arg))
+    {
+        if(_intervals.size() < 2) {
+            _intervals.clear();
+            _intervals.push_back(RealType(0));
+            _intervals.push_back(RealType(1));
+        }
+    }
+    /**
+     * Constructs a piecewise_constant_distribution that approximates a
+     * function.  The range of the distribution is [xmin, xmax).  This
+     * range is divided into nw equally sized intervals and the weights
+     * are found by calling the unary function f on the midpoints of the
+     * intervals.
+     */
+    template<class F>
+    piecewise_constant_distribution(std::size_t nw,
+                                    RealType xmin,
+                                    RealType xmax,
+                                    F f)
+      : _bins(nw, xmin, xmax, f)
+    {
+        if(nw == 0) { nw = 1; }
+        RealType delta = (xmax - xmin) / nw;
+        _intervals.reserve(nw + 1);
+        for(std::size_t i = 0; i < nw; ++i) {
+            _intervals.push_back(xmin + i * delta);
+        }
+        _intervals.push_back(xmax);
+    }
+    /**
+     * Constructs a piecewise_constant_distribution from its parameters.
+     */
+    explicit piecewise_constant_distribution(const param_type& parm)
+      : _bins(parm._weights),
+        _intervals(parm._intervals)
+    {
+    }
+
+    /**
+     * Returns a value distributed according to the parameters of the
+     * piecewist_constant_distribution.
+     */
+    template<class URNG>
+    RealType operator()(URNG& urng) const
+    {
+        std::size_t i = _bins(urng);
+        return uniform_real<RealType>(_intervals[i], _intervals[i+1])(urng);
+    }
+    
+    /**
+     * Returns a value distributed according to the parameters
+     * specified by param.
+     */
+    template<class URNG>
+    RealType operator()(URNG& urng, const param_type& parm) const
+    {
+        return piecewise_constant_distribution(parm)(urng);
+    }
+    
+    /** Returns the smallest value that the distribution can produce. */
+    result_type min BOOST_PREVENT_MACRO_SUBSTITUTION () const
+    { return _intervals.front(); }
+    /** Returns the largest value that the distribution can produce. */
+    result_type max BOOST_PREVENT_MACRO_SUBSTITUTION () const
+    { return _intervals.back(); }
+
+    /**
+     * Returns a vector containing the probability density
+     * over each interval.
+     */
+    std::vector<RealType> densities() const
+    {
+        std::vector<RealType> result(_bins.probabilities());
+        for(std::size_t i = 0; i < result.size(); ++i) {
+            result[i] /= (_intervals[i+1] - _intervals[i]);
+        }
+        return(result);
+    }
+    /**  Returns a vector containing the interval boundaries. */
+    std::vector<RealType> intervals() const { return _intervals; }
+
+    /** Returns the parameters of the distribution. */
+    param_type param() const
+    {
+        return param_type(_intervals, _bins.probabilities());
+    }
+    /** Sets the parameters of the distribution. */
+    void param(const param_type& parm)
+    {
+        std::vector<RealType> new_intervals(parm._intervals);
+        typedef discrete_distribution<std::size_t, RealType> bins_type;
+        typename bins_type::param_type bins_param(parm._weights);
+        _bins.param(bins_param);
+        _intervals.swap(new_intervals);
+    }
+    
+    /**
+     * Effects: Subsequent uses of the distribution do not depend
+     * on values produced by any engine prior to invoking reset.
+     */
+    void reset() { _bins.reset(); }
+
+    /** Writes a distribution to a @c std::ostream. */
+    BOOST_RANDOM_DETAIL_OSTREAM_OPERATOR(
+        os, piecewise_constant_distribution, pcd)
+    {
+        os << pcd.param();
+        return os;
+    }
+
+    /** Reads a distribution from a @c std::istream */
+    BOOST_RANDOM_DETAIL_ISTREAM_OPERATOR(
+        is, piecewise_constant_distribution, pcd)
+    {
+        param_type parm;
+        if(is >> parm) {
+            pcd.param(parm);
+        }
+        return is;
+    }
+
+    /**
+     * Returns true if the two distributions will return the
+     * same sequence of values, when passed equal generators.
+     */
+    BOOST_RANDOM_DETAIL_EQUALITY_OPERATOR(
+        piecewise_constant_distribution, lhs,  rhs)
+    {
+        return lhs._bins == rhs._bins && lhs._intervals == rhs._intervals;
+    }
+    /**
+     * Returns true if the two distributions may return different
+     * sequences of values, when passed equal generators.
+     */
+    BOOST_RANDOM_DETAIL_INEQUALITY_OPERATOR(piecewise_constant_distribution)
+
+private:
+    discrete_distribution<std::size_t, WeightType> _bins;
+    std::vector<RealType> _intervals;
+};
+
+}
+}
+
+#endif