Had to use from __future__ import division.
This pull request now works both with python2.7 and python3.4.

I'm sorry if you asked for this functionality before, I must have missed it. Both @bjedwards and myself have implemented the modularity maximization (also for the directed case) before. You can find his GSoC project and my own code.

Both are dated and not fully integrated so it's great that you're issuing this pull request. I don't have the time right now to compare the code but maybe it'll be helpful. Cheers.

Thanks for the answer @Midnighter.
I never asked for that feature before, but merely implemented it straight away!
It was not too complicated :)

As code comparison goes:

• Your code does implement this matrix (line 154 of https://github.com/Midnighter/Everyday-Utilities/blob/master/meb/utils/network/algorithms.py), although it is strictly in the context of community detection.
• GSoc does not implement the modularity matrix spectrum, but the modularity matrix appears in the context of modularity optimization (line 70).

The idea behind this pull request is to implement a method that construct the modularity matrix itself, independent of any optimization methods.
Much like the Laplacian matrix, this matrix does not necessarily requires community information to be of any use.
For example, it has seen been the subject of a renewed interest, due to papers that focus on matrix spectrum calculations (e.g. http://arxiv.org/abs/1306.2507 or http://arxiv.org/abs/1208.1275).

I think that the code would fit the present model as is.
My rationale is basically that it sits with the Laplacian matrix related code, which is a very similar concept.
I might be wrong, because I'm not yet fully aware of NetworkX future development targets / philosophy.

While we're discussing this, I can think of some extra features that I could implement if there is some interest:

• Null model selection (e.g. random graph, spin models, etc.)
• Directed / weighted extensions. I'd go for the variants that appear in GSoC project.

I like this. Code is clean and it's standalone. Let's merge now, and you can submit another pull request with additional features. The directed extensions would be nice.