spring layout does not converge on some graphs

[Carl Witty]

Try the following in 4.5.2.alpha0 (or after applying #9532):

sage: G = graphs.PetersenGraph()
sage: set_random_seed(0); G.plot(layout='spring', iterations=10000)
sage: set_random_seed(0); G.plot(layout='spring', iterations=10001)

I get very different-looking graphs. (If you go back and try with iterations=10000 again, you get the same graph again, showing that #9532 did make it reproducible, at least.)

Maybe some constants need tweaking?

I think this may be causing the problem reported in the first comment on #9584.

[Leif Leonhardy]

Replying to cwitty:

I think this may be causing the problem reported in the first comment on #9584.

Carl, should we fix that one here (numeric noise causing doctest failure)?

[Carl Witty]

The story so far: #9532 tried to make spring layout reproducible (and added tests to see that it was reproducible), but it wasn't enough and so the layout is actually not reproducible across platforms. Before #9532, spring layout was totally non-reproducible.

So in my opinion, the correct thing to do for the next release is just to remove the failing doctest. Spring layout wasn't reproducible before, so nobody can be depending on it being reproducible; and it isn't now, so there's no point in a test that verifies that it is reproducible. The patch that removes the doctest should not go on this ticket (removing the doctest is not part of fixing layout convergence). When this ticket is fixed, the doctest (or a similar one) should be added, to show that spring layout then does become reproducible.

[Leif Leonhardy]

Ok, I'll open a new ticket that simply adds a # random to that doctest.

[Leif Leonhardy]

Replying to leif:

Ok, I'll open a new ticket that simply adds a # random to that doctest.

Oh, I completely forgot to add this here: The related doctest error in Sage 4.5.2.alpha0 is now #9594, which already has positive review. (The comment in the patch links back to this ticket, which I found more appropriate, since the # random tag can hopefully be removed again once we have reproducible spring layout.)

[Jori Mäntysalo]

I think that this one has just forgotten to be closed. Frédéric, please click positive_review if you agree.

[Frédéric Chapoton]

ok, let us close that

[Erik Bray]

Determined to be invalid/duplicate/wontfix (closing as "wontfix" as a catch-all resolution).