Interactivity does not work in Jupyter Lab

[lschr]

Everything works fine in Jupyter Notebook, but not in Jupyter Lab (version 0.34.7).

• Styling of the buttons does not seem to work
• The buttons have no functionality; e.g. pressing the panning button does not enable panning

On the Javascript console I can the following error whenever the mouse pointer enters the area of the figure, which may or may not be related:

[feanor12]

I had a similar problem when I was using a notebook which had output cells with plots generated by (%matplotlib notebook)

I fixed it by:

1. delete all output cells
2. save notebook
3. reload page
4. restart kernel
5. run with %matplotlib widget

I think that %matplotlib notebook injects some kind of javascript that kills something inside jupyter lab.

[lschr]

@feanor12 Thanks, it looks like you are right. After restarting and clearing all output, the buttons work (except for the "power" button, which has already been reported in #61) and also the mouse pointer coordinates are shown properly in the bottom right corner.

The buttons still look Windows 95-ish, though (not sure if this is a bug).

[feanor12]

So maybe this is the cause for #62 as well.

[MerlinPendragon]

@feanor12 Yes!

[pdeitel]

Is this supposed to work with Matplotlib FuncAnimations too? All I get is a blank output and the FuncAnimation itself--which works just fine in IPython in a Terminal window--does not call the update function I defined for each frame. Thanks in advance for any suggestions.

EDIT: FuncAnimation works if the frames argument is an integer. It does not work if it's a generator function, which is what I am trying to use.

[tacaswell]

@pdeitel that is extremely odd. I am a bit worried that your generator is raising someplace and the exceptions are getting snarfed incorrectly. Internally, the animation module homgonizes all the input to generators (an integer gets turned into range).

[tacaswell]

I think that %matplotlib notebook injects some kind of javascript that kills something inside jupyter lab.

Yes, all of the js to make the plot interactive is inserted by the notebook backend (which is why it does not work in jlab as they have closed that convenient but large security hole).

[goiosunsw]

Additionally to @feanor12's recipe I had to close all tabs on the right hand side (open notebooks). Probably their javascript was still interfering with ipympl

[mischki]

Same problem and clearing outputs, restarting etc doesn't work. Any ideas? I've tried this in Chrome, Safari..

[pdeitel]

@pdeitel that is extremely odd. I am a bit worried that your generator is raising someplace and the exceptions are getting snarfed incorrectly. Internally, the animation module homgonizes all the input to generators (an integer gets turned into range).

I have not gone back to try this again recently, but I will be sometime in the next two weeks. I have all exceptions dumping text so I can see the problems that occur, so there should not be any getting "snarfed".

[amueller]

Seems related so adding here:
If I try

% matplotlib widget

in a notebook that had %matplotlib notebook before, I get

Warning: Cannot change to a different GUI toolkit: widget. Using notebook instead.

I understand this to mean "use % matplotlib notebook instead" which seems bad advice. Maybe that's not what it means?
I know I'm doing it wrong, but I think the warning is misleading.

[dopplershift]

It means once you’ve run %matplotlib notebook, you can’t later switch to %matplotlib widget (because the two will use different event loops). You need to only ever use one or the other while the kernel is running.

Open to suggestions to language that would make that more clear, but unfortunately that’s exactly what the existing message says to me (but then again I’ve been doing this for way too long).

[amueller]

Hm that's not what I understood from the message, and you can't use notebook at all in jupyter lab so the second sentence is confusing.
I would just say "Cannot change to different GUI toolkit widget from toolkit notebook" or "after already activating toolkit notebook"?
Or just your explanation "Can't change to toolkit widget since you already ran %matplotlib notebook"

[amueller]

I think I might have misread "using" as "use" on my first read which probably added to the confusion, so feel free to ignore me.

[thomasaarholt]

The warning should probably explain why it cannot change backend, or at least how the user can use a different one.

My suggestion below is verbose, but I think 50% of basic users have probably encountered this error message and then become very confused or worried.

How about changing

Warning: Cannot change to a different GUI toolkit: widget. Using notebook instead.

to

Warning: Cannot change GUI toolkit from notebook to widget. 
Only the inline and *one* interactive toolkit can be used together. 
Restart kernel to use a different toolkit.