On 30.06.2017 18:44, François Kawala wrote:
> I realize that I'm not familiar enough with boost to do this change.
>
> From what I get, I'll add three private members : _distance and _predecessor
> they would be initialized as follows :
>
>      _distance(get(vertex_distance, *_mg)),
>      _predecessor(get(predecessor, *_mg)),
>
> I don't know if the _distance member should be filled with zeros ?
>
> The last private member would be _reach a  std::vector<size_t>
>
> From that I'll declare two functions :
>
>      set_reached to update the reached vertices
>      reset_distance to reset the _distance, _predecessor and _reach to their
> default values.
>
> Does that sounds right ? If so, I'll guess that I'll have to make
> the do_djk_search function to call the set_reached and reset_distance functions.
>
> Am I missing something ?
>
> Sorry for this stuttering approach.

I've just pushed a commit that implements what you want; you can look inside
for the details.

In short, you can now do multiple searches as follows:

    dist_map = g.new_vp("double", numpy.inf)  # must be infinity
    pred_map = g.vertex_index.copy()          # must be identity

    for source in sources:

        # the following does not initialize dist_map and pred_map, and
        # returns an array of reached vertices

        dist_map, pred_map, reached = \
            shortest_distance(g, source, weights=w, pred_map=pred_map,
                              dist_map=dist_map, return_reached=True)

        # reset property maps for next search; this is O(len(reached))

        dist_map.a[reached] = numpy.inf
        pred_map.a[reached] = reached

Please tell me if this brings the improvements you were seeking.

Best,
Tiago




_______________________________________________
graph-tool mailing list
graph-tool@skewed.de
https://lists.skewed.de/mailman/listinfo/graph-tool