[igraph] python-igraph data to numpy

[Andreas Geiges]

Dear Igraph users,

I am using python-igraph in python2.7 in combination with numpy.

I need to speed up the following scenario and was wondering if someone 
can give my some hints.
I have a large graph with several thousands of vertices.
One property of the vertices is a one-dimensional numpy array ( e.g. 
array([0,0,0,0]) )

If I read out the property of one vertice, I get the numpy array 
directly back, no conversion needed.
But if I read the property of an vertice sequence, I get a list of 
array, which is perfectly fine.
BUT I like to have the data as a 2D-array in numpy, thus I need to use 
numpy.asarray() which cause an overhead, which
cost me a lost of computing time form many of this operation.

Thus, I was wondering if there is a way, to directly receive a 2D-array 
back from igraph, when using a sequence.
I think this is not possible right now, but maybe it can be implemented.
I would be willing to try that, but I would need some hints, where to start.

The following code is maybe demonstrating better what I want to accomplish:

######################################
import igraph  as ig
import numpy as np
from  timeit import timeit as ti
#%%  Setup
graph = ig.Graph()

nVert = 1000
graph.add_vertices(nVert)

for vertice in graph.vs:
    vertice['property'] = np.random.random(5)
    #print vertice

#creation of a arbitray sequence
vertexSequence = graph.vs[np.arange(0,nVert,5).tolist()]

#%% reading the properties of a sequence to an numpy array


#first possiblity using asarray
%timeit numpyArray2D = 
np.asarray(vertexSequence['property'],dtype=np.float64)

#second possiblity with pre-allocated array and numpy view
numpyArray2D = np.zeros([len(vertexSequence),5])
%timeit numpyArray2D[:] = vertexSequence['property']
######################################

Any help or comments are welcome. Thanks a lot in advance!!!

Best,
Andreas

--
Dr.-Ing. Andreas Geiges

Global Climate Forum (GCF)
Neue Promenade 6, D-10178 Berlin, Germany
http://www.globalclimateforum.org