MNT: Migrate VTK Widgets to GUI API

[GuillaumeFavelier]

This PR moves the VTK widgets into a dedicated Qt interface for the standalone solution and into Jupyter Widgets for notebooks.

Here is the latest sketch (it may change during development):

For the approach I chose, I'll just build the dock and move one widget at a time. So both old and new ways will live in parallel for a while. I expect working on refactoring and coverage at end. For now, it's a work in progress.

It's an item of #7162

[GuillaumeFavelier]

The previous interface relies on sliders heavily but I forgot that QSliders use integer by default and I don't seem to make the pyvistaqt.DoubleSlider work.

[larsoner]

Don't use sliders in Qt, use QDoubleSpinBox es

[larsoner]

(for time you probably do want a silder -- I would actually oversample the .times by a factor of 10 or so to give it a float-like behavior)

[GuillaumeFavelier]

That's the easy way out of this for sure 😁

[GuillaumeFavelier]

In the previous interface, each renderer had an orientation widget. On the sketch, there is only one widget for that. The simple solution is to duplicate those. A complicated one would be to command the click on a renderer to update the widget.

[larsoner]

How about a SpinBox representing the raveled plotter number to control? Then at a later date we can connect some on_click action to update this SpinBox with whichever one was clicked, then you have interactive mouse-based control and direct indexing control.

[GuillaumeFavelier]

How about a SpinBox representing the raveled plotter number to control?

This works too!

[GuillaumeFavelier]

For now, this is how it looks:

The time controller is not a slider and the orientation widget is not migrated yet. Also, I find the clim/scale controller hard to use.

[GuillaumeFavelier]

Now everything is migrated. I must admit, it is a cleaner 3D viewer:

The time widget is still not a slider tough.

[GuillaumeFavelier]

Once the case of the time widget is closed, I think it will be a good time to double-check the UI/UX and fix the (hopefully) minor details. Then I'll move on to test/coverage. I'm wondering how LinkViewer will behave with this interface 🤔

And finally same treatment for the notebook backend.

[GuillaumeFavelier]

I had modified the time callback back at the time because VTK sliders handle float values but QSlider support time index natively:

[GuillaumeFavelier]

It's now ready for UI/UX reviews! Meanwhile I'll work on fixing the tests.

[larsoner]

@GuillaumeFavelier sizing issue should be fixed, take a look at the code and _qt.py and _notebook.py etc. and make sure I didn't do something bad in terms of the hierarchy.

The sizing fix was QScrollArea + scroll.setMinimumSize(scroll.sizeHint().width, 0), i.e., make sure it always occupies its minimum width, then our _ensure_minimum_sizes and event processing takes care of the rest (i.e., resizing the window for us to accommodate the brain + dock). I also fixed the movie generation, no need to toggle the dock when capturing a movie anymore!

[agramfort]

screenshot now only takes a copy of the brain (no time series). I am saying why not but is it a conscious choice? for the me the scale control for clim does not behave smoothly. I can show you by sharing my screening my screen on monday

…

[larsoner]

screenshot now only takes a copy of the brain (no time series).

This behavior should change depending on whether you do time_viewer=True or (default) False

[agramfort]

with time viewer True:

with time viewer False:

using this code

import os
import mne
from mne.datasets import sample
data_path = sample.data_path()
sample_dir = os.path.join(data_path, 'MEG', 'sample')
subjects_dir = os.path.join(data_path, 'subjects')
fname_stc = os.path.join(sample_dir, 'sample_audvis-meg')
stc = mne.read_source_estimate(fname_stc, subject='sample')
initial_time = 0.13
time_viewer = False
# time_viewer = True
brain = stc.plot(subjects_dir=subjects_dir, initial_time=initial_time,
                 clim=dict(kind='value', lims=[3, 6, 9]),
                 hemi='split',
                 views=['lat', 'med'],
                 time_viewer=time_viewer
                 )

[GuillaumeFavelier]

How do you take the screenshot @agramfort ?

[GuillaumeFavelier]

I pushed a commit to fix:

AttributeError: module 'mne.viz.backends._qt' has no attribute '_close_3d_figure'
AttributeError: module 'mne.viz.backends._qt' has no attribute '_set_3d_view'

[GuillaumeFavelier]

one you're living in "Qt land" and in the other you're living in "ipywidgets land"

I admit this separation is simpler 🏹 and I like _abstract.py too

[GuillaumeFavelier]

Locally by using Brain in ipython, I obtain the size I request for the screenshot but during testing with pytest, test_brain_time_viewer still fails with:

===================================================================================================== FAILURES =====================================================================================================
_________________________________________________________________________________________ test_brain_time_viewer[pyvista] __________________________________________________________________________________________
mne/viz/_brain/tests/test_brain.py:506: in test_brain_time_viewer
    assert_allclose(img.shape, want_shape)
E   AssertionError: 
E   Not equal to tolerance rtol=1e-07, atol=0
E   
E   Mismatched elements: 1 / 3 (33.3%)
E   Max absolute difference: 282.
E   Max relative difference: 0.94
E    x: array([300, 582,   3])
E    y: array([300., 300.,   3.])

Reproduced in https://github.com/mne-tools/mne-python/runs/1989903394#step:13:3644 for example

[GuillaumeFavelier]

I pushed a commit to fix the screenshot taken from the interface. I'm surprised it was not reported sooner.

[hoechenberger]

I'm surprised it was not reported sooner.

Sorry 😢

[larsoner]

Locally by using Brain in ipython, I obtain the size I request for the screenshot but during testing with pytest, test_brain_time_viewer still fails with:

This is because we toggle the interface off earlier in the test, which makes the 3D area take up the entire window width, which I'd consider expected behavior (though in the future we could consider shrinking the window). Toggling the interface back on before doing the screenshot fixes it locally, will push a fix.

[GuillaumeFavelier]

I pushed a commit to fix AttributeError from _set_3d_title and _take_3d_screenshot as well.

[larsoner]

Okay, let's get this in and iterate further on the interface as necessary. Thanks a ton @GuillaumeFavelier , this is great!

[hoechenberger]

@GuillaumeFavelier You're a wizard 🧙🏻‍♂️ Thank you!