1. 技术文章
  2. What is needed to use BGL algorithms on existing data structures ( edges and vertices as vector<Object *>)?

What is needed to use BGL algorithms on existing data structures ( edges and vertices as vector<Object *>)?

由 我怕爱的太早我们不能终老 提交于 2019-12-06 04:29:19

The documentation for the Graph concepts is conveniently here: https://www.boost.org/doc/libs/1_70_0/libs/graph/doc/graph_concepts.html

So - you never told us what algorithms you intend to use.

So let me pick an examples: BFS. The docs say it requires:

A directed or undirected graph. The graph type must be a model of Vertex List Graph and Incidence Graph.

Looking at your pre-existing data structures, it looks like you only cover the Vertex List use case easily.

The edges are implemented more as an Edge List. It's not possible to emulate Incidence Graph from Edge List without runtime or storage overhead (that's mathematics, nothing to do with library or code quality).

In reality, it's pretty likely that you omitted parts of your pre-existing data-structures that are relevant to the problem, as most algorithms will be highly sub-optimal on just Vertex+Edge lists.

In practice I suppose you Edge list might be organized like a classical adjacency list (e.g. ordering by source vertex, so you CAN have a O(log(n)) lookup by source vertex).

For the example below I'm assuming this is the case. Keep in mind we're only approaching the complexity guarantees from Incidence Graph Concept:

Complexity guarantees

The source(), target(), and out_edges() functions must all be constant time. The out_degree() function must be linear in the number of out-edges.

To actually meet these requirements, you will need to have dedicated storage of out-edges per vertex

So, let'st have a go:

Mocking YourLibrary

namespace YourLibrary {
    struct myVertex {
    };

    struct myEdge {
        myVertex* _s = nullptr;
        myVertex* _t = nullptr;

        myVertex* source() const { return _s; }
        myVertex* target() const { return _t; }
    };

    using Vertices = std::vector<myVertex *>;
    using Edges = std::vector<myEdge *>;
}

Fulfilling the Graph Concepts

You wanted to keep references to the pre-existing data structures:

namespace Glue {

    struct MyGraph {
        struct EdgeOrder {
            template <typename A, typename B>
                bool operator()(A const* a, B const* b) const { return source(a) < source(b); }
            private:
            static auto source(YourLibrary::myVertex const* v) { return v; }
            static auto source(YourLibrary::myEdge const* e) { return e->source(); }
        };

        using Vertices = YourLibrary::Vertices;
        using Edges = YourLibrary::Edges;

        Vertices& _vertices;
        Edges& _edges;

        MyGraph(Vertices& vv, Edges& ee) : _vertices(vv), _edges(ee)  { }
    };
}

Now, I'll just run down the list of required trait types per concept:

namespace boost {

    template <> struct graph_traits<Glue::MyGraph> {
        // Due to Graph concept
        using vertex_descriptor      = YourLibrary::myVertex*;
        using edge_descriptor        = YourLibrary::myEdge*;
        using directed_category      = directed_tag;
        using edge_parallel_category = allow_parallel_edge_tag;
        static vertex_descriptor null_vertex() { return nullptr; }

        // Due to Vertex List concept
        struct traversal_category : vertex_list_graph_tag, incidence_graph_tag { };
        using vertex_iterator        = Glue::MyGraph::Vertices::const_iterator;
        using vertices_size_type     = std::size_t;

        // Due to Incidence Graph concept
        using out_edge_iterator = Glue::MyGraph::Edges::const_iterator;
        using degree_size_type = std::size_t;
    };

}

And finally re-open the namespace so ADL can find these functions that are required to satisfy the "valid expressions" criteria:

namespace Glue {
    // Due to Vertex List concept
    auto vertices(MyGraph const& g) {
        return std::make_pair(g._vertices.begin(), g._vertices.end());
    }

    std::size_t num_vertices(MyGraph const& g) {
        return g._vertices.size();
    }

    // Due to Incidence Graph concept
    auto source(YourLibrary::myEdge const* e, MyGraph const& g) {
        return e->source();
    }
    auto target(YourLibrary::myEdge const* e, MyGraph const& g) {
        return e->target();
    }

    auto out_edges(YourLibrary::myVertex const* v, MyGraph const& g) {
        return std::equal_range(g._edges.begin(), g._edges.end(), v, MyGraph::EdgeOrder{});;
    }
    std::size_t out_degree(YourLibrary::myVertex const* v, MyGraph const& g) {
        auto oee = std::equal_range(g._edges.begin(), g._edges.end(), v, MyGraph::EdgeOrder{});
        return std::distance(oee.first, oee.second);
    }
}

This would be roughly functionally equivalent to an adjacency_list with a setS for the vertex container.

See it Live On Coliru

Running BFS

All that's required in addition is for the arguments of the algorithm. You'd need both the color map and the vertex index map. This is completely normal and would also be required if you had e.g. adjacency_list<vecS, listS, directedS>.

I'll hide the index map inside the MyGraph wrapper and expose the color map so you can pick your preference:

Live On Coliru

#include <boost/graph/adjacency_list.hpp>
#include <boost/graph/breadth_first_search.hpp>
#include <boost/container/flat_map.hpp>
#include <algorithm>

namespace YourLibrary {
    struct myVertex {
    };

    struct myEdge {
        myVertex* _s = nullptr;
        myVertex* _t = nullptr;

        myVertex* source() const { return _s; }
        myVertex* target() const { return _t; }
    };

    using Vertices = std::vector<myVertex *>;
    using Edges = std::vector<myEdge *>;
}

namespace Glue {

    struct MyGraph {
        struct EdgeOrder {
            template <typename A, typename B>
                bool operator()(A const* a, B const* b) const { return source(a) < source(b); }
            private:
            static auto source(YourLibrary::myVertex const* v) { return v; }
            static auto source(YourLibrary::myEdge const* e) { return e->source(); }
        };

        using Vertices = YourLibrary::Vertices;
        using Edges = YourLibrary::Edges;

        using Index = boost::container::flat_map<Vertices::value_type, std::size_t>;

        Vertices& _vertices;
        Edges& _edges;
        Index _index;

        MyGraph(Vertices& vv, Edges& ee) : _vertices(vv), _edges(ee)  {
            _index.reserve(vv.size());
            std::size_t i = 0;
            for(auto v : vv) { _index[v] = i++; }
        }
    };
}

namespace boost {

    template <> struct graph_traits<Glue::MyGraph> {
        // Due to Graph concept
        using vertex_descriptor      = YourLibrary::myVertex*;
        using edge_descriptor        = YourLibrary::myEdge*;
        using directed_category      = directed_tag;
        using edge_parallel_category = allow_parallel_edge_tag;
        static vertex_descriptor null_vertex() { return nullptr; }

        // Due to Vertex List concept
        struct traversal_category : vertex_list_graph_tag, incidence_graph_tag { };
        using vertex_iterator        = Glue::MyGraph::Vertices::const_iterator;
        using vertices_size_type     = std::size_t;

        // Due to Incidence Graph concept
        using out_edge_iterator = Glue::MyGraph::Edges::const_iterator;
        using degree_size_type = std::size_t;
    };

}

namespace Glue {
    // Due to Vertex List concept
    auto vertices(MyGraph const& g) {
        return std::make_pair(g._vertices.begin(), g._vertices.end());
    }

    std::size_t num_vertices(MyGraph const& g) {
        return g._vertices.size();
    }

    // Due to Incidence Graph concept
    auto source(YourLibrary::myEdge const* e, MyGraph const& g) {
        return e->source();
    }
    auto target(YourLibrary::myEdge const* e, MyGraph const& g) {
        return e->target();
    }

    auto out_edges(YourLibrary::myVertex const* v, MyGraph const& g) {
        return std::equal_range(g._edges.begin(), g._edges.end(), v, MyGraph::EdgeOrder{});;
    }
    std::size_t out_degree(YourLibrary::myVertex const* v, MyGraph const& g) {
        auto oee = std::equal_range(g._edges.begin(), g._edges.end(), v, MyGraph::EdgeOrder{});
        return std::distance(oee.first, oee.second);
    }

    // Due to BFD requiring the index_map
    auto get(boost::vertex_index_t, MyGraph const& g) {
        return boost::make_assoc_property_map(g._index);
    }
}

int main() {
    // I hate manual memory management, so let's own some objects
    auto a = std::make_unique<YourLibrary::myVertex>();
    auto b = std::make_unique<YourLibrary::myVertex>();
    auto c = std::make_unique<YourLibrary::myVertex>();
    auto ab = std::make_unique<YourLibrary::myEdge>(YourLibrary::myEdge{a.get(), b.get()});
    auto bc = std::make_unique<YourLibrary::myEdge>(YourLibrary::myEdge{b.get(), c.get()});

    // These were given in your question:
    YourLibrary::Vertices vv { a.get(), b.get(), c.get() };
    YourLibrary::Edges ee { ab.get(), bc.get() };

    // this is the glue required to fulfill the BGL concepts:
    Glue::MyGraph g(vv, ee);

    // this is showing that you can now BFS on it
    using V = boost::graph_traits<Glue::MyGraph>::vertex_descriptor;
    V start_vertex = a.get();
    std::map<V, boost::default_color_type> color_data;

    boost::breadth_first_search(g, start_vertex,
            boost::visitor(boost::default_bfs_visitor{})
            .color_map(boost::make_assoc_property_map(color_data)));
}

Conclusion

Algorithms have requirements, and as long as you satisfy them, you can use whatever data structure you want.

In this case you MIGHT want to be really sure about the assumptions made and add this to the MyGraph constructor:

assert(std::is_sorted(_edges.begin(), _edges.end(), EdgeOrder{}));
标签
c++
boost
graph
boost-graph
易学教程内所有资源均来自网络或用户发布的内容，如有违反法律规定的内容欢迎反馈！
该文章没有解决你所遇到的问题?点击提问,说说你的问题,让更多的人一起探讨吧!