Notes from GraphXD sprints

Also see the example folder for concrete code

From discussions with Nelle Varoquaux, Aric Hagberg, and Dan Schult

  • restrict to ‘small’ networks (few hundred to thousand)
  • there are many choices in plotting mapping graph attributes -> visual properties
    • (x, y) layout
    • colors
    • labels
    • edges style
    • interactivity (hover / pick)
    • edge routing
  • want to be able to update the plot in response to updating the network
  • want to make it easily extensible
  • performance?

From discussion with large group of network practitioners

  • need “seaborn for networks” - heursitics for edge style
  • also want “seaborn for network statistics”
  • look at igraph - have lots of layout engines
  • people tend to use different plots for data exploration vs publication
  • functions for different on graph size
  • input to layout engine should be flexible
  • provide way to go to the bottom!

ball and edge

  • per-vertex/edges labels
  • per-vertex/edges annotation box (maybe?)
  • per-vertex/edge artists / subplots - excited about pie charts - do this as roll-over / tool-tip
  • mark sub-graphs
  • per-node style
  • per-edge style - directed edges - un-directed edges - multi-edges - arbitrary path

adjacency matrix view

  • show adjacency matrix
  • show anything at all in the gutters (all 8 places)


  • dendograms

sankey / flow

  • this may be better interface to the existing sankey functionality in Matplotlib

initial target cases

  • re-make jarod’s slides - small (25 nodes)
  • a 1000 node random graph - no labels, transparency, color per edge
  • A tree of some sort
  • splicing graph
  • neural networks
  • some flow network