Return networkx graph

[amanmajid]

Potential solution to #39

[tomalrussell]

Thanks for working on this @amanmajid 😊 .. will be super useful

Apart from the notes above, could you:

• add tests for each new method
• add networkx to setup.py extras and dev-rquirements
• add a bit more to the docs - what do you think would be most helpful? Maybe a section in the README and/or demo notebook?

What do you reckon about @czor847's suggestion of a to_igraph method - could you add it too while you're looking at this related stuff?

[amanmajid]

No worries!

I've just made some changes to the code as requested. In terms of the additional bullets:

• I will add tests once you've checked and approved the revisions.
• Will do. However, should we do this since you said networkx is should be an optional dependency?
• Hmm... I think we should update the README once we have several functions interacting with other networks libraries. So for now I will just add a cell to the existing notebook.
• I could add a similar function for the to_igraph - @czor847 do you have any existing code that I can adapt?

[tomalrussell]

Great 😊

I wonder if we should create the networkx graph as a MultiDiGraph - that way it can handle multiple edges between pairs of nodes, and will retain the direction of edges, both of which are okay within the GeoDataFrame representation. It should also keep snkit in line with osmnx and momepy in terms of exchanging networkx graphs.

In terms of recording requirements - yes, I’d put it on a line in “extras” so that pip will only install networkx if you do something like ‘pip install snkit[nx]’ and in dev-requirements so we can test the optional functionality.

[tomalrussell]

Some detail-oriented comments! But also a suggested change to how we construct the networkx graph that would make a reasonably large difference to the API.

[tomalrussell]

Should we pass in a column to use as weight here?

[tomalrussell]

Just thinking about this - it would be generally useful to add all the edge attributes here, including geometry.

[amanmajid]

I've changed this part to extract nodal positions and edge weights (https://github.com/amanmajid/snkit/blob/e55bc44c4aa848fb9f1724f8db504a3c6a7ad114/src/snkit/network.py#L613-L639). I'm not sure how to go about efficiently adding all edge attributes when defining the graph?

[tomalrussell]

Similarly to edges, can we add node attributes to the graph here?

[tomalrussell]

Thanks for this, @amanmajid - closes #39 and #40, potential follow-up in #42