Hacking sfdp to properly show large graphs

graph-tool
1

Hi all,

I'm trying to visualize some large (>100k vertices) graphs using graph-tool,
but I'm having trouble getting a good picture with the sfdp layout. I'm
unsure about is whether this is due to me not knowing how to use
sfdp_layout(), or whether it's just impossible lay out large graphs
efficiently due to the topology.

Here's a sample graph of what I'm talking about. I'm uploaded it in xml
format here: https://www.dropbox.com/s/5zg4u9re9rw6s4y/test.xml.gz?dl=0

It's about 100k vertices and max degree ~ 2400 and average degree ~ 7.8.
Obviously, this will be hard to graph. Anyway, when I draw the graph using
the standard method:

pos = gt.sfdp_layout(G)
gt.graph_draw(G, pos=pos)

Then I get a graph that looks like this:

<http://main-discussion-list-for-the-graph-tool-project.982480.n3.nabble.com/file/n4026993/test.png&gt;

The vertices are tiny. Which makes sense, because otherwise with the If I
increase the vertex size by supplying vprops={'size': 2} (for example) to
gt.graph_draw(), then I just get a huge blob.

My question: is it possible to force sfdp to space out the vertices more so
that I can increase the vertex size *and* have them not all coalesce into
one giant blob?

2

Visualization of very large graphs is tricky. Note that if you have 100k
vertices, and if you lay them side by side, each occupying one pixel, you
already have a (useless) 316x316 image. If you increase this to 1000x1000,
you still do not have sufficient space to have the vertices not lying on top
of each other in a meaningful way.

But to answer your question more directly: No, graph-tool is not very smart
about node overlaps. The SFDP algorithm does not currently take into account
the size of the vertices.

Best,
Tiago