Add network/graph category to scientific domains tab

[jarrodmillman]

Fixes gh-320. See #313 (comment).

[rgommers]

Thanks Jarrod. Having another look around for other packages that may deserve a mention, what do you think about these:

• https://igraph.org/python/
• https://snap.stanford.edu/snappy/
• https://scikit-network.readthedocs.io/en/latest/

They all seem solid/popular/maintained, and it's not clear to me how they compare against graph-tool and PyGSP.

[mdeff]

NetworkX, graph-tool, igraph, NetworKit, snappy, scikit-network do network science: they are concerned with the analysis of networks. PyGSP, which stands for Graph Signal Processing, is more concerned with the analysis of data / features / signals supported by networks (as functions of vertices) and is closer to (spectral) graph theory.

I have the following to say about the NetSci category:

• NetworkX is definitely the most popular
• graph-tool is well polished and maintained by its author, more mature and probably a bit more popular than NetworKit
• scikit-network is too new (first commit a year ago)
• snap and their Python SWIG interface (snappy) seems mostly used internally (by Leskovec's group at Stanford)
• igraph and NetworKit are both solid complements to NetworkX and graph-tool
• igraph is however more popular (e.g., 257,455 vs 29,153 downloads from conda-forge, or dependent projects on github and libraries.io) and mature (first commit 14 vs 7 years ago).

I see two alternative categories depending on whether or not you want to take the alternative PyGSP point-of-view:

1. Network Science: NetworkX, graph-tool, igraph, NetworKit
2. Graphs and Networks: NetworkX, graph-tool, igraph, PyGSP

I'd vote for 2, but as the author of PyGSP I'm of course biased.

[rgommers]

Thanks for the thorough overview @mdeff

I'd vote for 2, but as the author of PyGSP I'm of course biased.

I'm happy to go with this option. "Graphs and Networks" does sound better to me.

[jarrodmillman]

I updated the PR to go with option 2.

I also slightly changed the icon. I committed the script that generates it:

• static/images/content_images/sc_dom_img/graph.py

Feel free to change the colors, line thickess, or whatever. Or let me know if you want me to change something.

Is there a better place to put the script? Should I take it out of the repo?

[jarrodmillman]

I just looked #369 and noticed it looks like they got rid of the colors and the grays. I'll change the icon again.

[rgommers]

Adding the script next to the icon sounds fine to me. Would it be possible to make the png size smaller though @jarrodmillman? The new icon is 10x larger than the other icons, 250 kb is a lot for a small image.

[mdeff]

Why not export as SVG? Should be small given the small number of elements. I think they're moving to SVG in gh-369.

[jarrodmillman]

I saved the icon as a SVG instead of a PNG. It is now 8.0K.

I also made it 160px by 120px. I couldn't find any specification of the preferred size, so I just randomly selected one of the existing PNGs and used its width.

[rgommers]

LGTM now, thanks @jarrodmillman!