About the overlapping problem of drawing self-loops

[ljt10086]

As for python-igraph when drawing a self-loop, it does not seem to be able to change the orientation of the self-loop by modifying edge.loop.angle like r version. After reading your code, I find that you use the draw_loop_edge function when drawing the self-loop, and use pi/4 angle, which makes it impossible to modify edge.loop.angle. Eventually, there will be self-loop edges that overlap with other edges when drawing.

[ljt10086]

[iosonofabio]

I wrote that bit. Happy to add this option, just didn't think about it at the time.

There is a big PR related to plotting in the queue. As soon as that one's send, I can quickly fix this one.

[szhorvat]

@iosonofabio This was fixed at some point in the past, but then it broke again, see #321.

[iosonofabio]

Yep, remember the discussion.

[iosonofabio]

@ljt10086 I can't find a place in the docs where we allow users to change the angle of loops, what you call loop.angle. Can you please paste a link to the relevant section in the docs?

@szhorvat I didn't realize #321 is still open. Well a good chance to fix it again.

I'll add this on top of #669 which should make it into 0.11. Unfortunately, I don't think I currently have bandwidth to also fix it in 0.10.X

[ljt10086]

In this page there is some description of loop.angle in R-igraph：https://igraph.org/r/doc/plot.common.html
And I saw that in the r version of the implementation someone solved the self-loop orientation problem by setting loop.angle:
https://stackoverflow.com/questions/52410379/r-igraph-orientation-of-edge-beginning-and-ending-from-same-vertex

[iosonofabio]

Thank you. @szhorvat @ntamas any opinion? Implement the user angle would definitely be easier than the auto detection, and consistent with R.

Btw at some point we should try to harmonize most of these arguments across our interfaces perhaps.

[szhorvat]

I am not too worried about harmonizing visualization across interfaces. Each interface has separately implemented visualization that works with the very different native visualization capabilities of the host language.

IMO loop.angle is a crutch, and the proper solution is automatic adjustment of self-loops, similar to how Mathematica does it and to how it used to work in py-igraph 0.9.

Here are a bunch of random graphs to show how Mathematica plots multi-self-loops in various positions:

[szhorvat]

Issue for R interface is at igraph/rigraph#407

[ntamas]

I agree with @szhorvat that finding the right angles for loops is not something that the user should be dealing with, but I don't know what the best algorithm would be. My idea is something like this:

1. Calculate the angles to all neighbors of the vertex.
2. Sort the angles in ascending order.
3. Find the largest difference between consecutive angles in the sorted list, taking into account "wrapping around" at the end of the list. (Of course everything should be mod 360). Take the angles at both sides of the largest interval that we've found here -- this is the interval in which the layout is likely to be the least crowded.
4. Place the loop edges spaced out equally in this interval. If there is only one loop edge to be placed, orient it towards the angle halfway between the lower and upper bounds. If there are two edges, place them at 1/3 and 2/3 and so on.

[szhorvat]

Exactly.

I'm pretty sure this is what Mathematica does, i.e. find the largest gap between non-loop edges and place the self-loops there.

The self-loops have an effective angular size, defined through Bezier curves. If this does not fit in the gap, then reduce this size.

Special case: Isolated vertices where there are no non-loop edges to set a reference for the angle. Mathematica uses a fixed horizontal angle here. An alternative solution is to orient away from the centre.

---

What we do need a parameter for (and I'm not sure if we have it) is self-loop size. Non-loop edges have a natural size: the distance between endpoints. But for loops it is sometimes useful to adjust the size.

[iosonofabio]

Let's continue this discussion on #669

I've implemented the algo as suggested by both of you and will now add a new documented argument for loop size. There's currently a discussion about arrows in self loops.

[stale]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed in 14 days if no further activity occurs. Thank you for your contributions.

[szhorvat]

I think this is fixed now (reopen if I'm mistaken).