我正在寻找一种将Boost Graph Library与Boost uBLAS结合使用的智能方法。更准确地说,我需要通过使用图形邻接矩阵和包含每个顶点的一些其他顶点属性的向量之间的标量积的结果来更新每个顶点的给定顶点属性。让我给你一个(不幸的是冗长的)最小的例子来说明这个问题:
#include <boost/graph/adjacency_list.hpp>
#include <boost/graph/iteration_macros.hpp>
#include <boost/numeric/ublas/matrix.hpp>
#include <boost/numeric/ublas/io.hpp>
using namespace boost;
namespace ublas = boost::numeric::ublas;
struct Vertex { //Using bundled vertex properties
double old_potential;
double new_potential;
};
typedef adjacency_list< listS, vecS, directedS, Vertex > Graph;
int main(){
//[Prepare a graph with two vertex property maps and initialize
Graph graph;
add_edge (0, 1, graph);
add_edge (0, 3, graph);
add_edge (1, 2, graph);
auto v_old_potential = boost::get( &Vertex::old_potential, graph );
auto v_new_potential = boost::get( &Vertex::new_potential, graph );
unsigned int source_strength = 0;
BGL_FORALL_VERTICES( v, graph, Graph ) {
v_old_potential[v] = source_strength++;
v_new_potential[v] = 0;
}
//]
//[ Extracting the adjacency matrix by iteration through all edges --> uBLAS matrix
ublas::zero_matrix<int> zero_matrix( num_vertices(graph) , num_vertices(graph) );
ublas::matrix<int> adjacency_matrix( zero_matrix ); //initialize properly
BGL_FORALL_EDGES( e, graph, Graph ) {
adjacency_matrix( source(e,graph), target(e,graph) ) = 1;
adjacency_matrix( target(e,graph), source(e,graph) ) = 1;
}
//]
//[ Extracting the old potentials by iteration through all vertices --> uBLAS vector
ublas::zero_vector<double> zero_vector( num_vertices(graph) );
ublas::vector<double> old_potential_vector( zero_vector ); //initialize properly
ublas::vector<double> new_potential_vector( zero_vector ); //initialize properly
BGL_FORALL_VERTICES(v, graph, Graph) {
old_potential_vector( vertex(v,graph) ) = v_old_potential[v];
}
//]
//[ Compute new potentials = A . old potentials !
new_potential_vector = ublas::prod ( adjacency_matrix, old_potential_vector ); // new = A.old
//]
//[ Updating the property map for the new potentials with the newly computed values from above
BGL_FORALL_VERTICES(v, graph, Graph) {
v_new_potential[v] = old_potential_vector( vertex(v,graph) );
}
//]
std::cout << adjacency_matrix << std::endl; //output = [4,4]((0,1,0,1),(1,0,1,0),(0,1,0,0),(1,0,0,0))
std::cout << old_potential_vector << std::endl; //output = [4](0,1,2,3)
std::cout << new_potential_vector << std::endl; //output = [4](4,2,1,0)
}
现在,虽然我的建议是一个可能的解决方案,但我对它不太满意。问题是,(a)我将整个old_potential
属性映射复制到关联ublas::vector
以计算标量积。并且(b)我还需要遍历new_potential
属性映射以将新计算的值返回到图中。我怀疑这些操作会在我的应用程序中重复很多次,这就是为什么我希望从一开始就尽可能干净地完成这部分。
理想情况下,我希望完成所有这些来回复制,而是使用某种适配器来制作boost::property_map
工作作为ublas::vector
. prod()
使用这样的东西会很棒:
adapter(new_potential) = ublas::prod( adjacency_matrix, adapter(old_potential) );
如果有人知道如何实现此类功能或如何改进我的解决方案,我将不胜感激。