[JOSS REVIEW] sp.visualize closes the plot for more than one sample?

[j-faria]

When running

y = sp.sample_ylm(nsamples=2).eval()
y.shape # (2, 256)

sp.visualize(y)

the plot shows up but closes right after.
There are a few plt.close() in starry_process.visualize.visualize which are likely the cause.

(This is part of joss-reviews/issues/3071)

[rodluger]

Thanks for raising this. It works for me when I do

python -c "from starry_process import StarryProcess; sp = StarryProcess(); y = sp.sample_ylm(nsamples=2).eval(); sp.visualize(y)"

I see a flickering animation switching between the two samples, and the plot remains open until I close it.

Are you in a Jupyter notebook?

[j-faria]

Are you in a Jupyter notebook?

No, in the IPython console. It's fine if you don't want to support that, I was just surprised with the behaviour.

[rodluger]

Oh I definitely want to support that! But I still can't reproduce the issue, even inside an IPython session.

What do you get when you run

get_ipython()

in your session?

[j-faria]

Ok, I tracked it down to the --pylab option to ipython (I know, considered evil!).
Without it, the plot shows and stays flickering.

Regardless of that option, I get this

get_ipython()
<IPython.terminal.interactiveshell.TerminalInteractiveShell at 0x7f9a8f0d2ed0>

[j-faria]

Just to add that with ipython --matplotlib it also doesn't work.

[rodluger]

Interesting. Seems to be an issue with FuncAnimation. I'll keep digging -- thanks for pointing this out.

This is related:
https://stackoverflow.com/questions/19466825/ipython-pylab-matplotlib-animations-hang

[j-faria]

It works if you return the animation from inside visualize and then from sp.visualize

[rodluger]

Ah, I think I got it. The --matplotlib and --pylab settings enable interactive mode, in which plt.show() is not blocking, so plt.close() immediately closes the animation. (I still want to call plt.close() in non-interactive mode, otherwise the plot stays open and calling visualize multiple times gobbles up RAM).

The fix is to simply check if plt.isinteractive() when calling plt.close(). Let me know if the latest commit fixed the issue fore you.