change graph display from matplotlib to something more interactive

[mamueller]

candidates:

• igraph (library used in SoilRexp and dependency of py2cytoscape) https://igraph.org/python/doc/tutorial/tutorial.html#creating-a-graph-from-scratch

  • pro:
    • nice layouts
  • issues:
    • non interactive

• bokeh very interactive and well integrated with python and jupyter

  • issues:
  • does not support directed graphs out of the box
  • does depend on nextworkx layouts which are somewhat unusable

• ipycytoscape https://blog.jupyter.org/interactive-graph-visualization-in-jupyter-with-ipycytoscape-a8828a54ab63

  • issues:
  • labels of nx.graphs do not seem to show up but native cytospace nodes are (workaround seems possible)

• py2cytoscape
  https://py2cytoscape.readthedocs.io/en/latest/#py2cytoscape wrapper for cytoscape REST API

  • issues:
    Examples from documentation do not work (probably due to network details)

• plotly

  • issues: no support for visualizing directed graphs and multigraphs
  • lately gone commercial ?

• native networkx drawing

  • issues:
    • not interactive, sometimes not so beautiful to look at
    • no multigraph ( plotting) support
    • somewhat unpredictable graph layouts

examples: https://github.com/MPIBGC-TEE/bgc_md2/tree/iss16-vegetation-cycling-matrices/prototypes/GraphLibExamples/ModelGraphs

[mamueller]

There are some examples of the computability graphs which I have not yet linked. I am still working on it.

[mamueller]

Found a very simple d3.js based interactive graph plotter, which could be very useful to build complete compartmental models, which is the second important use case for the improved graph plotting and interaction. (besides the computability graphs)

https://bl.ocks.org/cjrd/6863459

The only thing missing is a way to add edge attributes (in our case sympy expressions for the fluxes)