|
Boost Users : |
Subject: Re: [Boost-users] [boost][graph] Efficiency problems in boost::adjacency_list
From: Jeremiah Willcock (jewillco_at_[hidden])
Date: 2009-06-22 15:18:16
On Tue, 23 Jun 2009, Yinghai Lu wrote:
> hi, erveryone,
> I'm building a graph incrementally using boost::adjacency_list with bundled properties. I did the following typedef:
> typedef boost::adjacency_list<setS, vecS, bidirectionalS, EdgeClass, NodeClass> Graph;
>
> EdgeClass and NodeClass is some classes holding the properties of edges and nodes so that I can access them by using
> g[u].some_property = 1;
>
> My problem is that I'm building the graph incrementally and the graph is huge. And with the definition I'm using the
> vector to hold the vertices, which results in many memory relocation and copy and makes the program stagnate. I wanted to
> use listS/mapS for the vertex list instead. However, the bundled property seems to work only with vecS only. Otherwise,
> compilation error will be thrown. So does anyone know how to solve this problem? Thanks.
How incrementally are you building the graph? Do you have the graph data
up-front but in an unknown order? Is it possible for you to make two
passes over the input data to build the graph? Or do you need to look at
the graph being built before it is fully finished? Normally, for memory
and efficiency issues for read-only graphs, I recommend using the
compressed sparse row graph format. It can be built from edges in an
arbitrary order (in the trunk or 1.40 release branches), and supports
bundled properties (only). Would that be useful for you?
If you do need more dynamic capabilities, you would need an adjacency_list
and might need to switch to non-bundled properties (I do not know about
the issue with listS and mapS preventing bundled properties). Note that
you can have a single non-bundled property whose value type is just your
bundle structure.
-- Jeremiah Willcock
Boost-users list run by williamkempf at hotmail.com, kalb at libertysoft.com, bjorn.karlsson at readsoft.com, gregod at cs.rpi.edu, wekempf at cox.net