Plotting a partition of a bipartite graph takes much more time than the entire graph

[ghost]

In my code, plotting the entire bipartite graph takes much less time than plotting one of the projections alone (which, of course, have less nodes than the bipartite graph, since it is only a part of the bipartite graph). How can it be?

Here is the code:

rm(list=ls())
library(igraph)
library(tictoc)
nodes <- read.csv("./ma_nodes.csv", header=T, as.is=T)
links <- read.csv("./ma_edges.csv", header=T, as.is=T)
nodes$type <- as.logical(nodes$IsInvestor)
net <- graph_from_data_frame(d=links, vertices=nodes, directed=T)
net.bp <- bipartite_projection(net, multiplicity=F)
net.prj1 = net.bp$proj1
net.prj2 = net.bp$proj2

tic("Plotting bibpartite net")
plot(net)
toc()

tic("Plotting prj2")
plot(net.prj2)
toc()

tic("Plotting prj1")
plot(net.prj1)
toc()

Here is the result:

Plotting bibpartite net: 5.04 sec elapsed
Plotting prj2: 0.21 sec elapsed
Plotting prj1: 77.9 sec elapsed

Also note that using Sys.time() to measure time (and taking the difference end time - start time), this effect is not captured: the result would be that plotting prj1 takes less time than plotting net. But this is not true: plotting the last graph takes way more time than plotting the other twos. It is a problem of sys.time() which is not able to capture that (maybe it has to do with the problem?)

What is going on here?

I can send the files of the dataset if you want, I am just not sure how I can do that

[ntamas]

It is not the number of nodes that primarily matters when plotting a graph but the number of edges. A bipartite projection can easily have more edges than the initial graph; if you have a vertex of type A with d neighbors of type B in the bipartite graph, the corresponding subgraph in the type B bipartite projection will have d*(d-1)/2 edges. You only need a few vertices of high degree in the initial graph to see this effect.

[ghost]

I see. Is there a way to speed up the plot? I tried to install the Intel MKL libraries but nothing

[ntamas]

The Intel MKL libraries would have no effect here.

Is there a way to speed up the plot?

There are two significant time-consuming parts of a plotting call: the layout calculation and the actual drawing itself. You can try using a different layout algorithm using the layout=... argument of the plot() function. layout="circle" can be calculated fast so if plotting is still slow with layout=..., then the drawing is the bottleneck -- otherwise you might be lucky with a different layout algorithm.

[ghost]

Actually I am pre-calulating the layout in the real code, and then passing to the function the layout parameter

[gaborcsardi]

The graphics device can be slow as well, e.g. with anti-aliasing, etc.