On 11/11/13 14:34, Jeremiah Willcock wrote:
OK, so you can iterate them, just not store them. One approach you might consider is to:
1. Create an iterator range containing all possible (source, target) pairs in sorted order.
2. Apply a boost::filter_iterator or similar that computes your predicate.
3. Use one of the edges_are_sorted_t constructors to do a single copy from the filtered iterator range; those constructors only make one pass over the range for a directed graph, so the fact that filter_iterator recomputes the predicate for each pass over the sequence is not a problem.
Actually, I must admit I'm quite new to boost :( You are suggesting me to take a look to boost.range, and to boost.iterator libraries, am I correct? Then, after creating these magical objects (I assume they're like functors that will be computed at runtime), the graph constructor will do the rest. Did I interpret this correctly? Since this is a N^2 problem, and I must apply a function f(vert1, vert2) in order to create or not an edge, I was in the process of parallelizing the edge construction. Note that computations are completely independent from one node to another. This would be a great thing, as you can imagine. I read that BGL does not support concurrency, so if you have any suggestion, I would greatly appreciate that :)
Yes, but this is C++03 code, so I didn't have move construction to work with. Thus, you pass in your own vectors by reference and they are given back empty.
Thanks for your pointer. However, I feel something is weird in my code:
the targets get deallocated, while sources remain the same. This is my
testing code, just for fun:
// IN THE HEADER:
// typedef boost::compressed_sparse_row_graphboost::directedS
boost_graph;
long num_verts = 1000000;
std::vectorstd::size_t orig, dest;
orig.reserve(num_verts * 16);
dest.reserve(num_verts * 16);
std::cout << "reserved " << num_verts * 16 << std::endl;
std::cout << "orig " << orig.size() << " dest " << dest.size() <<
std::endl;
std::default_random_engine generator;
std::uniform_int_distributionstd::size_t distribution(0, num_verts);
auto rng = std::bind(distribution, generator);
for (long i = 0; i < num_verts; i++)
{
std::size_t s = rng();
std::size_t e = rng() ;
orig.push_back(s);
dest.push_back(e);
//std::cout << s << " - " << e << std::endl;
}
std::cout << "orig " << orig.size() << " dest " << dest.size() <<
std::endl;
graph_ = std::unique_ptr