glm/test/external/boost/graph/rmat_graph_generator.hpp

// Copyright 2004, 2005 The Trustees of Indiana University.

// Use, modification and distribution is 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)

//  Authors: Nick Edmonds
//           Andrew Lumsdaine
#ifndef BOOST_GRAPH_RMAT_GENERATOR_HPP
#define BOOST_GRAPH_RMAT_GENERATOR_HPP

#include <math.h>
#include <iterator>
#include <utility>
#include <vector>
#include <queue>
#include <map>
#include <boost/shared_ptr.hpp>
#include <boost/assert.hpp>
#include <boost/random/uniform_int.hpp>
#include <boost/random/uniform_01.hpp>
#include <boost/graph/graph_traits.hpp>
#include <boost/type_traits/is_base_and_derived.hpp>
#include <boost/type_traits/is_same.hpp>
#include <boost/test/floating_point_comparison.hpp>

using boost::shared_ptr;
using boost::uniform_01;

// Returns floor(log_2(n)), and -1 when n is 0
template <typename IntegerType>
inline int int_log2(IntegerType n) {
  int l = 0;
  while (n > 0) {++l; n >>= 1;}
  return l - 1;
}

struct keep_all_edges {
  template <typename T>
  bool operator()(const T&, const T&) { return true; }
};

template <typename Distribution, typename ProcessId>
struct keep_local_edges {

  keep_local_edges(const Distribution& distrib, const ProcessId& id)
    : distrib(distrib), id(id)
  { }

  template <typename T>
  bool operator()(const T& x, const T& y)
  { return distrib(x) == id || distrib(y) == id; }

private:
  const Distribution& distrib;
  const ProcessId&    id;
};

template <typename RandomGenerator, typename T>
void
generate_permutation_vector(RandomGenerator& gen, std::vector<T>& vertexPermutation, T n)
{
  using boost::uniform_int;

  vertexPermutation.resize(n);

  // Generate permutation map of vertex numbers
  uniform_int<T> rand_vertex(0, n-1);
  for (T i = 0; i < n; ++i)
    vertexPermutation[i] = i;

  // Can't use std::random_shuffle unless we create another (synchronized) PRNG
  for (T i = 0; i < n; ++i)
    std::swap(vertexPermutation[i], vertexPermutation[rand_vertex(gen)]);
}

template <typename RandomGenerator, typename T>
std::pair<T,T>
generate_edge(shared_ptr<uniform_01<RandomGenerator> > prob, T n,
              unsigned int SCALE, double a, double b, double c, double d)
{
  T u = 0, v = 0;
  T step = n/2;
  for (unsigned int j = 0; j < SCALE; ++j) {
    double p = (*prob)();

    if (p < a)
      ;
    else if (p >= a && p < a + b)
      v += step;
    else if (p >= a + b && p < a + b + c)
      u += step;
    else { // p > a + b + c && p < a + b + c + d
      u += step;
      v += step;
    }

    step /= 2;

    // 0.2 and 0.9 are hardcoded in the reference SSCA implementation.
    // The maximum change in any given value should be less than 10%
    a *= 0.9 + 0.2 * (*prob)();
    b *= 0.9 + 0.2 * (*prob)();
    c *= 0.9 + 0.2 * (*prob)();
    d *= 0.9 + 0.2 * (*prob)();

    double S = a + b + c + d;

    a /= S; b /= S; c /= S;
    // d /= S;
    // Ensure all values add up to 1, regardless of floating point errors
    d = 1. - a - b - c;
  }

  return std::make_pair(u, v);
}

namespace boost {

  /*
    Chakrabarti's R-MAT scale free generator.

    For all flavors of the R-MAT iterator a+b+c+d must equal 1 and for the
    unique_rmat_iterator 'm' << 'n^2'.  If 'm' is too close to 'n^2' the
    generator may be unable to generate sufficient unique edges

    To get a true scale free distribution {a, b, c, d : a > b, a > c, a > d}
  */

  template<typename RandomGenerator, typename Graph>
  class rmat_iterator
  {
    typedef typename graph_traits<Graph>::directed_category directed_category;
    typedef typename graph_traits<Graph>::vertices_size_type vertices_size_type;
    typedef typename graph_traits<Graph>::edges_size_type edges_size_type;

  public:
    typedef std::input_iterator_tag iterator_category;
    typedef std::pair<vertices_size_type, vertices_size_type> value_type;
    typedef const value_type& reference;
    typedef const value_type* pointer;
    typedef void difference_type;

    // No argument constructor, set to terminating condition
    rmat_iterator()
      : gen(), edge(0) { }

    // Initialize for edge generation
    rmat_iterator(RandomGenerator& gen, vertices_size_type n,
                  edges_size_type m, double a, double b, double c,
                  double d, bool permute_vertices = true)
      : gen(), n(n), a(a), b(b), c(c), d(d), edge(m),
        permute_vertices(permute_vertices),
        SCALE(int_log2(n))

    {
      this->gen.reset(new uniform_01<RandomGenerator>(gen));

      BOOST_ASSERT(boost::test_tools::check_is_close(a + b + c + d, 1., boost::test_tools::fraction_tolerance(1.e-5)));

      if (permute_vertices)
        generate_permutation_vector(gen, vertexPermutation, n);

      // TODO: Generate the entire adjacency matrix then "Clip and flip" if undirected graph

      // Generate the first edge
      vertices_size_type u, v;
      boost::tie(u, v) = generate_edge(this->gen, n, SCALE, a, b, c, d);

      if (permute_vertices)
        current = std::make_pair(vertexPermutation[u],
                                 vertexPermutation[v]);
      else
        current = std::make_pair(u, v);

      --edge;
    }

    reference operator*() const { return current; }
    pointer operator->() const { return &current; }

    rmat_iterator& operator++()
    {
      vertices_size_type u, v;
      boost::tie(u, v) = generate_edge(this->gen, n, SCALE, a, b, c, d);

      if (permute_vertices)
        current = std::make_pair(vertexPermutation[u],
                                 vertexPermutation[v]);
      else
        current = std::make_pair(u, v);

      --edge;

      return *this;
    }

    rmat_iterator operator++(int)
    {
      rmat_iterator temp(*this);
      ++(*this);
      return temp;
    }

    bool operator==(const rmat_iterator& other) const
    {
      return edge == other.edge;
    }

    bool operator!=(const rmat_iterator& other) const
    { return !(*this == other); }

  private:

    // Parameters
    shared_ptr<uniform_01<RandomGenerator> > gen;
    vertices_size_type n;
    double a, b, c, d;
    int edge;
    bool permute_vertices;
    int SCALE;

    // Internal data structures
    std::vector<vertices_size_type> vertexPermutation;
    value_type current;
  };

  // Sorted version for CSR
  template <typename T>
  struct sort_pair {
    bool operator() (const std::pair<T,T>& x, const std::pair<T,T>& y)
    {
      if (x.first == y.first)
        return x.second > y.second;
      else
        return x.first > y.first;
    }
  };

  template<typename RandomGenerator, typename Graph,
           typename EdgePredicate = keep_all_edges>
  class sorted_rmat_iterator
  {
    typedef typename graph_traits<Graph>::directed_category directed_category;
    typedef typename graph_traits<Graph>::vertices_size_type vertices_size_type;
    typedef typename graph_traits<Graph>::edges_size_type edges_size_type;

  public:
    typedef std::input_iterator_tag iterator_category;
    typedef std::pair<vertices_size_type, vertices_size_type> value_type;
    typedef const value_type& reference;
    typedef const value_type* pointer;
    typedef void difference_type;

    // No argument constructor, set to terminating condition
    sorted_rmat_iterator()
      : gen(), values(sort_pair<vertices_size_type>()), done(true)
    { }

    // Initialize for edge generation
    sorted_rmat_iterator(RandomGenerator& gen, vertices_size_type n,
                         edges_size_type m, double a, double b, double c,
                         double d, bool permute_vertices = true,
                         EdgePredicate ep = keep_all_edges())
      : gen(), permute_vertices(permute_vertices),
        values(sort_pair<vertices_size_type>()), done(false)

    {
      BOOST_ASSERT(boost::test_tools::check_is_close(a + b + c + d, 1., boost::test_tools::fraction_tolerance(1.e-5)));

      this->gen.reset(new uniform_01<RandomGenerator>(gen));

      std::vector<vertices_size_type> vertexPermutation;
      if (permute_vertices)
        generate_permutation_vector(gen, vertexPermutation, n);

      // TODO: "Clip and flip" if undirected graph
      int SCALE = int_log2(n);

      for (edges_size_type i = 0; i < m; ++i) {

        vertices_size_type u, v;
        boost::tie(u, v) = generate_edge(this->gen, n, SCALE, a, b, c, d);

        if (permute_vertices) {
          if (ep(vertexPermutation[u], vertexPermutation[v]))
            values.push(std::make_pair(vertexPermutation[u], vertexPermutation[v]));
        } else {
          if (ep(u, v))
            values.push(std::make_pair(u, v));
        }

      }

      current = values.top();
      values.pop();
    }

    reference operator*() const { return current; }
    pointer operator->() const { return &current; }

    sorted_rmat_iterator& operator++()
    {
      if (!values.empty()) {
        current = values.top();
        values.pop();
      } else
        done = true;

      return *this;
    }

    sorted_rmat_iterator operator++(int)
    {
      sorted_rmat_iterator temp(*this);
      ++(*this);
      return temp;
    }

    bool operator==(const sorted_rmat_iterator& other) const
    {
      return values.empty() && other.values.empty() && done && other.done;
    }

    bool operator!=(const sorted_rmat_iterator& other) const
    { return !(*this == other); }

  private:

    // Parameters
    shared_ptr<uniform_01<RandomGenerator> > gen;
    bool permute_vertices;

    // Internal data structures
    std::priority_queue<value_type, std::vector<value_type>, sort_pair<vertices_size_type> > values;
    value_type current;
    bool       done;
  };


  // This version is slow but guarantees unique edges
  template<typename RandomGenerator, typename Graph,
           typename EdgePredicate = keep_all_edges>
  class unique_rmat_iterator
  {
    typedef typename graph_traits<Graph>::directed_category directed_category;
    typedef typename graph_traits<Graph>::vertices_size_type vertices_size_type;
    typedef typename graph_traits<Graph>::edges_size_type edges_size_type;

  public:
    typedef std::input_iterator_tag iterator_category;
    typedef std::pair<vertices_size_type, vertices_size_type> value_type;
    typedef const value_type& reference;
    typedef const value_type* pointer;
    typedef void difference_type;

    // No argument constructor, set to terminating condition
    unique_rmat_iterator()
        : gen(), done(true)
    { }

    // Initialize for edge generation
    unique_rmat_iterator(RandomGenerator& gen, vertices_size_type n,
                         edges_size_type m, double a, double b, double c,
                         double d, bool permute_vertices = true,
                         EdgePredicate ep = keep_all_edges())
      : gen(), done(false)

    {
      BOOST_ASSERT(boost::test_tools::check_is_close(a + b + c + d, 1., boost::test_tools::fraction_tolerance(1.e-5)));

      this->gen.reset(new uniform_01<RandomGenerator>(gen));

      std::vector<vertices_size_type> vertexPermutation;
      if (permute_vertices)
        generate_permutation_vector(gen, vertexPermutation, n);

      int SCALE = int_log2(n);

      std::map<value_type, bool> edge_map;

      edges_size_type edges = 0;
      do {
        vertices_size_type u, v;
        boost::tie(u, v) = generate_edge(this->gen, n, SCALE, a, b, c, d);

        // Lowest vertex number always comes first
        // (this means we don't have to worry about i->j and j->i being in the edge list)
        if (u > v && is_same<directed_category, undirected_tag>::value)
          std::swap(u, v);

        if (edge_map.find(std::make_pair(u, v)) == edge_map.end()) {
          edge_map[std::make_pair(u, v)] = true;

          if (permute_vertices) {
            if (ep(vertexPermutation[u], vertexPermutation[v]))
              values.push_back(std::make_pair(vertexPermutation[u], vertexPermutation[v]));
          } else {
            if (ep(u, v))
              values.push_back(std::make_pair(u, v));
          }

          edges++;
        }
      } while (edges < m);
      // NGE - Asking for more than n^2 edges will result in an infinite loop here
      //       Asking for a value too close to n^2 edges may as well

      current = values.back();
      values.pop_back();
    }

    reference operator*() const { return current; }
    pointer operator->() const { return &current; }

    unique_rmat_iterator& operator++()
    {
      if (!values.empty()) {
        current = values.back();
        values.pop_back();
      } else
        done = true;

      return *this;
    }

    unique_rmat_iterator operator++(int)
    {
      unique_rmat_iterator temp(*this);
      ++(*this);
      return temp;
    }

    bool operator==(const unique_rmat_iterator& other) const
    {
      return values.empty() && other.values.empty() && done && other.done;
    }

    bool operator!=(const unique_rmat_iterator& other) const
    { return !(*this == other); }

  private:

    // Parameters
    shared_ptr<uniform_01<RandomGenerator> > gen;

    // Internal data structures
    std::vector<value_type> values;
    value_type              current;
    bool                    done;
  };

  // This version is slow but guarantees unique edges
  template<typename RandomGenerator, typename Graph,
           typename EdgePredicate = keep_all_edges>
  class sorted_unique_rmat_iterator
  {
    typedef typename graph_traits<Graph>::directed_category directed_category;
    typedef typename graph_traits<Graph>::vertices_size_type vertices_size_type;
    typedef typename graph_traits<Graph>::edges_size_type edges_size_type;

  public:
    typedef std::input_iterator_tag iterator_category;
    typedef std::pair<vertices_size_type, vertices_size_type> value_type;
    typedef const value_type& reference;
    typedef const value_type* pointer;
    typedef void difference_type;

    // No argument constructor, set to terminating condition
    sorted_unique_rmat_iterator()
      : gen(), values(sort_pair<vertices_size_type>()), done(true) { }

    // Initialize for edge generation
    sorted_unique_rmat_iterator(RandomGenerator& gen, vertices_size_type n,
                                edges_size_type m, double a, double b, double c,
                                double d, bool bidirectional = false,
                                bool permute_vertices = true,
                                EdgePredicate ep = keep_all_edges())
      : gen(), bidirectional(bidirectional),
        values(sort_pair<vertices_size_type>()), done(false)

    {
      BOOST_ASSERT(boost::test_tools::check_is_close(a + b + c + d, 1., boost::test_tools::fraction_tolerance(1.e-5)));

      this->gen.reset(new uniform_01<RandomGenerator>(gen));

      std::vector<vertices_size_type> vertexPermutation;
      if (permute_vertices)
        generate_permutation_vector(gen, vertexPermutation, n);

      int SCALE = int_log2(n);

      std::map<value_type, bool> edge_map;

      edges_size_type edges = 0;
      do {

        vertices_size_type u, v;
        boost::tie(u, v) = generate_edge(this->gen, n, SCALE, a, b, c, d);

        if (bidirectional) {
          if (edge_map.find(std::make_pair(u, v)) == edge_map.end()) {
            edge_map[std::make_pair(u, v)] = true;
            edge_map[std::make_pair(v, u)] = true;

            if (ep(u, v)) {
              if (permute_vertices) {
                values.push(std::make_pair(vertexPermutation[u], vertexPermutation[v]));
                values.push(std::make_pair(vertexPermutation[v], vertexPermutation[u]));
              } else {
                values.push(std::make_pair(u, v));
                values.push(std::make_pair(v, u));
              }
           }

            ++edges;
          }
        } else {
          // Lowest vertex number always comes first
          // (this means we don't have to worry about i->j and j->i being in the edge list)
          if (u > v && is_same<directed_category, undirected_tag>::value)
            std::swap(u, v);

          if (edge_map.find(std::make_pair(u, v)) == edge_map.end()) {
            edge_map[std::make_pair(u, v)] = true;

            if (permute_vertices) {
              if (ep(vertexPermutation[u], vertexPermutation[v]))
                values.push(std::make_pair(vertexPermutation[u], vertexPermutation[v]));
            } else {
              if (ep(u, v))
                values.push(std::make_pair(u, v));
            }

            ++edges;
          }
        }

      } while (edges < m);
      // NGE - Asking for more than n^2 edges will result in an infinite loop here
      //       Asking for a value too close to n^2 edges may as well

      current = values.top();
      values.pop();
    }

    reference operator*() const { return current; }
    pointer operator->() const { return &current; }

    sorted_unique_rmat_iterator& operator++()
    {
      if (!values.empty()) {
        current = values.top();
        values.pop();
      } else
        done = true;

      return *this;
    }

    sorted_unique_rmat_iterator operator++(int)
    {
      sorted_unique_rmat_iterator temp(*this);
      ++(*this);
      return temp;
    }

    bool operator==(const sorted_unique_rmat_iterator& other) const
    {
      return values.empty() && other.values.empty() && done && other.done;
    }

    bool operator!=(const sorted_unique_rmat_iterator& other) const
    { return !(*this == other); }

  private:

    // Parameters
    shared_ptr<uniform_01<RandomGenerator> > gen;
    bool             bidirectional;

    // Internal data structures
    std::priority_queue<value_type, std::vector<value_type>,
                        sort_pair<vertices_size_type> > values;
    value_type current;
    bool       done;
  };

} // end namespace boost

#ifdef BOOST_GRAPH_USE_MPI
#include <boost/graph/distributed/rmat_graph_generator.hpp>
#endif // BOOST_GRAPH_USE_MPI

#endif // BOOST_GRAPH_RMAT_GENERATOR_HPP