Re: [igraph] Python Interface to igraph

[Tamas Nepusz]

Dear Lorenzo,

> Now, I have found out that there is a Python interface, so it does not
> make any sense to invoke it from R.
Well, it might - the R interface currently has some features that the  
Python interface does not (related to parts that are implemented in C+ 
+ in the original library, like walktrap/spinglass community detection  
and the BLISS isomorphism - however they will be included in 0.5!),  
and the calling convention of the Python interface is not always the  
same as one would guess from the R interface.

> (1) Is this the correct way of proceeding? I would guess so, since
> Python recognizes the library as being installed, but I would like to
> be told so by someone acknowledgeable.
As for me, I use python setup.py build && python setup.py install.  
However, you can simply check whether the installation is correct by  
invoking Python and typing:

>>> from igraph.test import test
>>> test()
........................................................................................................
----------------------------------------------------------------------
Ran 104 tests in 0.695s
OK

(note that this is v0.5 from the dev tree, v0.4.5 has ~50 tests only)

> (2)Say that at some point a version 0.5 is release and I want to
> upgrade my installation of igraph. Should I just repeat what I did at
> (1)?
Yes. The new version will simply overwrite the old one in /usr/lib/ 
python.
Just a small note: until up to 0.4.5, the Python extension was  
statically linked to the C core, so there was no need to install the C  
libraries in order to use it from Python. This will change with 0.5,  
you'll have to install the C libraries as well before using igraph  
from Python. Debian packages for the C library will be made available  
soon after the release of 0.5 (maybe in a day or so). We have a Debian  
package repository for igraph releases at http://cneurocvs.rmki.kfki.hu/packages 
. The binaries are built for the i386 architecture, the source  
packages build on i386, x86_64 and amd64 architectures. So after the  
release of igraph 0.5, you'll first have to build the C library  
packages for your architecture. See http://www.debian.org/doc/manuals/apt-howto/ch-sourcehandling.en.html 
 for instructions on how to build source packages. The two lines that  
you'll have to add to /etc/apt/sources.list are as follows:
deb http://cneurocvs.rmki.kfki.hu /packages/binary/
deb-src htp://cneurocvs.rmki.kfki.hu /packages/source/

I have plans on building Debian packages for the Python interface as  
well, but it's unlikely that I'll have time to do that in the near  
future. :(

> Would that be the case now? In other words: what is the procedure to
> uninstall igraph if I need to?
The Python interface installs itself in /usr/lib/python2.4/site- 
packages/ (or .../python2.5/... of course). Either you'll find an  
igraph subdirectory here that you can safely delete, or a file called  
igraph-0.4.5-py2.4-linux-x86_64.egg (replace version numbers and  
architecture names as necessary) which can also be safely deleted.  
That's all.

> my_g=graph.adjacency(dist_mat<=d,mode="undirected")
> my_g=simplify(my_g)
> nc=no.clusters(my_g)
> cluster_comp=clusters(my_g)$csize
>
> How would they translate in Python?
First suppose that dist_mat consists of only zeros and ones. Let's  
call it dist_mat_01. You can type:

>>> from igraph import *
>>> g=Graph.Adjacency(dist_mat_01, ADJ_UNDIRECTED)
>>> g.simplify()
>>> clustering=g.clusters()
>>> nc=len(clustering)
>>> cluster_sizes=cl.sizes()

Now, unfortunately producing a dist_mat consisting of only zeros and  
ones is not so easy in Python as in R (at least the syntax is a little  
bit different). I'd do it as follows:

>>> def clamp(value, threshold):
...   if value<threshold: return 1
...   return 0
...
>>> dist_mat_01 = [map(clamp, row) for row in dist_mat]

A more powerful matrix datatype is available for Python in a package  
called NumPy (Numeric Python). You can also use that. If dist_mat is a  
NumPy matrix, then you can simply type:

>>> g=Graph.Adjacency((dist_mat <= d).tolist(), ADJ_UNDIRECTED)

(dist_mat <= d) works the same as in R, but its result is a NumPy  
matrix which igraph cannot process directly, since it expects a list  
of lists. That's why it is converted to a list first by invoking  
the .tolist() method.

Best,
--
T.