On Friday, 8. April 2011 16:33:23 Jeremiah Willcock wrote:
On Fri, 8 Apr 2011, Cedric Laczny wrote:
Hi,
I would like to know if there is a way to accelerate the retrieval of properties of an edge or a vertex? Actually, I define a new vertex-property (vertex_properties) and this again uses a "struct VertexProperties { ... }" to specify the details of the properties. However, when I want to get the properties of say vertex v, I need to first get the properties-map from the graph and then get the property for v. This obviously suffers from the lookup in the property_map and I would like to know if there is a way to accelerate this?
You can save the property map in a variable to avoid repeated lookups, but I suspect they are not too expensive.
I agree with you on that.
In particular, when using vecS as the vertex_container because the vertices could generally be used as indices... I would find this very convenient, especially in combination with filtered_graph when filtering only on vertices that have certain properties (e.g. a category) and one only wants to see those that have a specific character (e.g. color).
Fast lookups will probably require external properties, since those are much simpler and have fewer layers of indirection (no pointers to members, for example). Can you copy your properties to an iterator_property_map (built on a vector) before the high-performance parts of your code?
This is a nice idea. So if I am getting this right, in the case of vecS, get(vertex_index, g) should return the identity_property_map, thus resulting in linear lookup. At the same time, using the intermediate vertex_index map and an iterator_property_map would guarantee at least valid lookups in case of other vertex-containers, albeit they would then require the usual O(log n) lookup. This of course only if the graph also provides a vertex-index map. Or am I missing something here?
-- Jeremiah Willcock
Best, Cedric