[igraph] Re: a few questions about saving and plotting a 50k nodes network

[Michael Bommarito]

It is definitely possible to plot graphs this large, though as Tamas
says, many structures become obscured and the plots are often
uninformative.

This is a graph of my Twitter 4-neighbors with |V|=130k and some large
number of edges.  It took half a day or so to render.
http://computationallegalstudies.com/2010/01/24/large-scale-130000-zoomable-visualization-of-a-twitter-network/

Here are some tricks I have developed over time:
* If you are comfortable with C and programming animations or
visualizations, invest the time to use igraph's C bindings and
libcairo.
* If you don't want to use C, use Python, not R.
* Drop edge['width'], edge['arrow_width'] and edge['arrow_size']
* Don't make edges too high contrast w.r.t. background
* Use faster layout algorithms to build a seed layout and then use the
seed layout for the next layout algorithm

I know the cost of C and libcairo is high, but the results can be
worth it!
http://www.vimeo.com/9366071

Thanks,
-Michael J. Bommarito II
University of Michigan
http://www-personal.umich.edu/~mjbommar/

On Feb 10, 8:45 am, Tamas Nepusz <address@hidden> wrote:
> Hi Yannick,
>
> > I'm currently analysing this graph (with R) :
> > [...]
> > First, is it a good move to save everything as graph attributes as above ?
>
> I don't think you'll do any harm if you do that. If it fits into the
> memory of your computer, then why not.
>
> > > plot(g.sw.gcc)
> > Erreur : impossible d'allouer un vecteur de taille 261.5 Mo
> > [...]
> > What is wrong ? Is this graph too big ?
>
> Probably yes. I have never tried plotting a graph of this size using R
> -- chances are that you would be getting an intangible mess of edges
> anyway, no matter how good the layout is. Note that it doesn't really
> matter how large your graph is when you save it, as R might be
> allocating several auxiliary data structures when plotting the graph
> itself, and these may exceed the memory limits. I don't know how
> plotting in R works and what's going on behind the scenes, but my
> impression is that this simply won't work in R (or any other general
> purpose graph drawing package).
>
> Once, when I had to plot a very large graph for illustration purposes
> (~70K vertices, ~500K edges), I came up with a Python script that drew
> the graph itself based on a pre-calculated layout using Python's
> bindings to the Cairo library.
>
> If I were you and I really needed a visualisation of the graph, I would
> try plotting the graph without the edges first. In other words, create
> an empty graph with the same number of vertices and plot that, using
> your precalculated DrL layout. If R fails to do that, maybe GraphViz can
> (http://www.graphviz.org), but I don't know how to pass a predefined
> layout to GraphViz, or whether it's even possible or not.
>
> > Thirdly, do you have any recommendations, when using closeness.estimate or
> > betweenness.estimate, in the choice of cutoffs ?
>
> I'm not aware of such recommendations in the literature.
>
> --
> Tamas
>
> _______________________________________________
> igraph-help mailing list
> address@hidden://lists.nongnu.org/mailman/listinfo/igraph-help