[Networkit] NetworKit Digest, Vol 16, Issue 16

Christian Staudt christian.staudt at kit.edu
Tue Mar 3 10:12:17 CET 2015


Hi Ian,

> Before I reply, a question: is there a way to load a graph from a gzip'ed file or from a python file-like-object?


So far there isn’t. Any idea on how this could be implemented?

>     I made an undirected version of G, and connected components is working fine on it.

For a directed graph, you should use StronglyConnectedComponents rather than ConnectedComponents.
In the current Dev branch version, ConnectedComponents throws an exception is G is directed, which was not the case before and may have resulted in a crash. 

You may want to switch to the Dev branch to avoid running into bugs that have already been fixed but the fix has not yet been released. With the Dev branch, does properties.overview(G) still fail?

> Is there a way I can turn on more verbose debug logging? 

Yes. There is networkit.setLogLevel. If you build NetworKit in optimized mode (e.g. python3 setup.py develop --optimize=Dbg, which is the default), you have loglevels ERROR, WARN and INFO available. For more verbose logging, you need to compile in debug mode (e.g. —python3 setup.py develop --optimize=Dbg), then you have also the levels DEBUG and TRACE.

Best,
Chris

On 03 Mar 2015, at 02:36, Ian <ian.wood at anu.edu.au> wrote:

> Before I reply, a question: is there a way to load a graph from a gzip'ed file or from a python file-like-object?
> 
> 
>> 
>> Hi Ian,
>> a segfault should never happen, so we need to do some repair work. However, it?s not clear what causes the error, there are many possibilities along the way. A couple of diagnostics questions:
>> 
>> - What is the result of properties.size(G) ?
>     (265724,17425529)
>> - Is G directed or weighted?
>     directed, not weighted
>> - Is it only ConnectedComponents that fails while other algorithms run properly?
>     I've only looked at communities (PLM) and the degree distribution, they work ok.
>     I made an undirected version of G, and connected components is working fine on it.
>> - Can you verify that the GMLReader reads your input file format correctly, e.g. with a smaller example?
> I constructed a smaller graph (with networkx) from a random selection of nodes in my big graph and some of their neighbours, saved as gml, and imported into networkit. properties.overview works fine on this graph.
> I also constructed a tiny graph with edges (i,i+1) for i in range(10), and it imports as it should.
> The GML files were made with networkx.write_gml(). 
> 
> Is there a way I can turn on more verbose debug logging? 
> 
> I don't think I mentioned that the networkit version is 3.4.1, installed with pip.
> 
> Cheers
> Ian
>> 
>> Chris
>> 
>> 
>> On 27 Feb 2015, at 06:44, Ian <ian.wood at anu.edu.au> wrote:
>> 
>>> Hi Chris,
>>> 
>>> I have a software stack running in python 2 that I don't really have time to port. There may be some modules without python 3 support too.
>>> 
>>> I tried today to install into a virtualenv running python 3.3, but I'm still using Ubuntu 12.04 which only has 3.2 in it's default repositories. It was matplotlib that gave me lots of trouble - it wasn't recognising system installed backends (perhaps because they're compiled against  python 3.2?). I'm giving up on that for now - too much time in that black hole already!  It might be a good idea to add python3 to the requirements list on the "Getting Started" page.. 
>>> 
>>> I put together a new virtual machine with Ubuntu 14.04 which has the required versions and installed networkit into a virtual env - all good.
>>> I now have a problem though. I've a directed graph with some 265k nodes and 17 million edges. 
>>> I can read it in ok, but when I do properties.overview(G), I get this:
>>> 
>>> In [5]: G = graphio.readGraph("netowrk-internal.gml",Format.GML)
>>> 
>>> In [6]: prop                                    
>>> properties  property                            
>>> 
>>> In [6]: properties.overview(G,showDegreeHistogram=False)
>>> INFO:root:[...] calculating properties          
>>> INFO:root:[...] determining degree distribution 
>>> Value 0 in data. Throwing out 0 values          
>>> Calculating best minimal value for power law fit
>>> /home/ubuntu/python-virtual-environments/networkit/lib/python3.4/site-packages/networkit/powerlaw.py:546: FutureWarning: comparison to `None` will result in an elementwise object comparison in the future.
>>>  if data!=None and not (parameter_range and self.parent_Fit):
>>> /home/ubuntu/python-virtual-environments/networkit/lib/python3.4/site-packages/networkit/powerlaw.py:1047: FutureWarning: comparison to `None` will result in an elementwise object comparison in the future.
>>>  if data==None and hasattr(self, 'parent_Fit'):
>>> /home/ubuntu/python-virtual-environments/networkit/lib/python3.4/site-packages/networkit/powerlaw.py:598: FutureWarning: comparison to `None` will result in an elementwise object comparison in the future.
>>>  if data==None and hasattr(self, 'parent_Fit'):
>>> /home/ubuntu/python-virtual-environments/networkit/lib/python3.4/site-packages/networkit/powerlaw.py:671: FutureWarning: comparison to `None` will result in an elementwise object comparison in the future.
>>>  if data==None and hasattr(self, 'parent_Fit'):
>>> /home/ubuntu/python-virtual-environments/networkit/lib/python3.4/site-packages/networkit/powerlaw.py:556: FutureWarning: comparison to `None` will result in an elementwise object comparison in the future.
>>>  if data==None and hasattr(self, 'parent_Fit'):
>>> /home/ubuntu/python-virtual-environments/networkit/lib/python3.4/site-packages/networkit/powerlaw.py:1166: FutureWarning: comparison to `None` will result in an elementwise object comparison in the future.
>>>  if data==None and hasattr(self, 'parent_Fit'):
>>> /home/ubuntu/python-virtual-environments/networkit/lib/python3.4/site-packages/networkit/powerlaw.py:725: FutureWarning: comparison to `None` will result in an elementwise object comparison in the future.
>>>  if data==None and hasattr(self, 'parent_Fit'):
>>> INFO:root:[...] detecting communities           
>>> INFO:root:[...] performing community detection: PLM
>>> <_NetworKit.PLM object at 0x7f34a977cc88>       
>>> INFO:root:[...] preparing degree histogram      
>>> INFO:root:[...] finding connected components....
>>> Segmentation fault   
>>> 
>>> It may be relevant that it's twitter data, so node id's are python int's, and many are long (ie: I think more than a C long int)?
>>> If you give me instructions, I can attempt to get more informative diagnostics..
>>> 
>>> Cheers
>>> Ian
>>> 
>>> On 20/02/2015, at 10:00 PM, <networkit-request at ira.uni-karlsruhe.de> wrote:
>>> 
>>>> Message: 3
>>>> Date: Fri, 20 Feb 2015 10:08:50 +0100
>>>> From: Christian Staudt <christian.staudt at kit.edu>
>>>> To: "NetworKit: a toolkit for high-performance network analysis"
>>>> 	<networkit at ira.uni-karlsruhe.de>
>>>> Subject: Re: [Networkit] pip install fails with "cannot import name
>>>> 	DEVNULL"
>>>> Message-ID: <66B1DEB7-7E4C-437C-B5B8-07BDE89A7996 at kit.edu>
>>>> Content-Type: text/plain; charset="windows-1252"
>>>> 
>>>> 
>>>> On 20 Feb 2015, at 03:15, Ian <ian.wood at anu.edu.au> wrote:
>>>> 
>>>>> OK, it's proving difficult to install with pip (as opposed to pip3, which I've not tried yet - will it be available to python 2.7 if I install with pip3??)
>>>> 
>>>> Hi Ian,
>>>> it seems like you are trying to install networkit for Python 2.x. However, Python 3.x is required. Support for Python 2 is not something we have currently planned, frankly I do not see the point in backporting. So what are your reasons for wanting to stick with Python 2 rather than upgrading?
>>>> 
>>>> Best,
>>>> Chris
> 
> <ATT00001.c>

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.ira.uni-karlsruhe.de/mailman/private/networkit/attachments/20150303/81ede37d/attachment-0001.html>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 495 bytes
Desc: Message signed with OpenPGP using GPGMail
URL: <https://lists.ira.uni-karlsruhe.de/mailman/private/networkit/attachments/20150303/81ede37d/attachment-0001.sig>


More information about the NetworKit mailing list