Problem with pseudo_diameter() and GraphView object ()

I get a 'Segmentation fault (core dumped)' error when I try to calculate
the pseudo_diameter() of a GraphView object <class 'graph_tool.GraphView'>.
I must convert the GraphView object to a Graph object <class
'graph_tool.Graph'> to perform this calculation? How do I make this
conversion?

My detailed problem is:

I have a graph that is fragmented in several components and I want to
calculate several measures on every component, so I must extract every
component as a graph to be able to apply the different tools on it. To do
this I use label_components() to create a vertex PropertyMap that label the
components

vprop_comps, size_comps = gt.label_components(g)

from this Property Map (vprop_comps) I create one Boolean PropertyMap for
every component, the Boolean Property Map for a component will have 1 for
every vertex that belong to this component and 0 otherwise, finally I
construct a GraphView object from the Boolean Property Map of a component,
and then apply the different tools on this object. My code is:

# for all components
for i in range(size_comps.size):
      # create a boolean property map
      vprop_component_i = g.new_vertex_property("bool")
      # explore the property map that label the components 'vprop_comps'
      for j in range(vprop_comps.a.size):
      # if vprop_comps sugest that vertex j is in component i
            if (vprop_comps.a[j] == i):
                  # put 'true' in the property of vertex j in the property
map of component i
                  vprop_component_i[g.vertex(j)] = True
      # With the boolean property map constructed for component i I can
create a
      # GraphView object representing component i:
      c_i = gt.GraphView(g, vfilt=vprop_component_i)
      # and apply the different measures on it:
      clust_c_i = gt.global_clustering(c_i)[0]
      length_c_i = gt.pseudo_diameter(c_i)[0]

Everything goes fine until I calculate the pseudo_diameter() of c_i, if I
try to, I receive a 'Segmentation fault (core dumped)' error.

Is there a more direct way to extract the components as graph objects?

attachment.html (2.33 KB)

2013/1/9 Sirius Fuenmayor <fuenmayor.sirius(a)gmail.com>

I get a 'Segmentation fault (core dumped)' error when I try to calculate
the pseudo_diameter() of a GraphView object <class 'graph_tool.GraphView'>.
I must convert the GraphView object to a Graph object <class
'graph_tool.Graph'> to perform this calculation? How do I make this
conversion?

My detailed problem is:

I have a graph that is fragmented in several components and I want to
calculate several measures on every component, so I must extract every
component as a graph to be able to apply the different tools on it. To do
this I use label_components() to create a vertex PropertyMap that label the
components

vprop_comps, size_comps = gt.label_components(g)

from this Property Map (vprop_comps) I create one Boolean PropertyMap for
every component, the Boolean Property Map for a component will have 1 for
every vertex that belong to this component and 0 otherwise, finally I
construct a GraphView object from the Boolean Property Map of a component,
and then apply the different tools on this object. My code is:

# for all components
for i in range(size_comps.size):
      # create a boolean property map
      vprop_component_i = g.new_vertex_property("bool")
      # explore the property map that label the components 'vprop_comps'
      for j in range(vprop_comps.a.size):
      # if vprop_comps sugest that vertex j is in component i
            if (vprop_comps.a[j] == i):
                  # put 'true' in the property of vertex j in the property
map of component i
                  vprop_component_i[g.vertex(j)] = True
      # With the boolean property map constructed for component i I can
create a
      # GraphView object representing component i:
      c_i = gt.GraphView(g, vfilt=vprop_component_i)
      # and apply the different measures on it:
      clust_c_i = gt.global_clustering(c_i)[0]
      length_c_i = gt.pseudo_diameter(c_i)[0]

Everything goes fine until I calculate the pseudo_diameter() of c_i, if I
try to, I receive a 'Segmentation fault (core dumped)' error.

Is there a more direct way to extract the components as graph objects?

I don't know about the segfault (I encountered the same problem with a test
graph), your approach uses too much code to select the components. I
suggest the following:

vprop_comps, size_comps = gt.label_components(g)
for i in range(len(size_comps)):
    gv = gt.GraphView(g,vfilt=lambda x: vprop_comps==i)

Hope it helps, at least in having a cleaner code.

Giuseppe

attachment.html (3.24 KB)

I get a 'Segmentation fault (core dumped)' error when I try to calculate
the pseudo_diameter() of a GraphView object <class
'graph_tool.GraphView'>. I must convert the GraphView object to a Graph
object <class 'graph_tool.Graph'> to perform this calculation? How do I
make this conversion?

The segfault is obviously a bug, I'll fix that. As a workaround, you can
convert a GraphView into a Graph instance by doing the following:

    g = Graph(g, prune=True)

Is there a more direct way to extract the components as graph objects?

You can do a little more compactly as Giuseppe showed, but as far as
actually efficiency is concerned, you cannot do much better.

Cheers,
Tiago

The bug is now fixed in the git version.

As another workaround, you can do simply:

    pseudo_diameter(g, source=g.vertex(0, use_index=False))

This will also avoid the segfault.

Cheers,
Tiago