Mayavi problems on MAC

[bloyl]

I'm having problems with the macos and mayavi visulaization in mne 0.23 (but probably earlier versions as well).

MAC hardware

macOS Catalina 
Version 10.15.7
  Model Name:	MacBook Pro
  Model Identifier:	MacBookPro15,2
  Processor Name:	Quad-Core Intel Core i7
  Processor Speed:	2.8 GHz

 Intel Iris Plus Graphics 655:
    Chipset Model:	Intel Iris Plus Graphics 655
    Type:	GPU

I created a fresh environment for mne 0.23

Make env

 conda env remove -n mne-stable
 curl --remote-name https://raw.githubusercontent.com/mne-tools/mne-python/master/environment.yml
 conda env create -n mne-stable --file environment.yml
 conda activate mne-stable

mne sys_info

(mne-stable) bloyl@507b9dedf1fe ~ % mne sys_info
Platform:      macOS-10.15.7-x86_64-i386-64bit
Python:        3.9.2 | packaged by conda-forge | (default, Feb 21 2021, 05:02:20)  [Clang 11.0.1 ]
Executable:    /Users/bloyl/anaconda3/envs/mne-stable/bin/python3.9
CPU:           i386: 8 cores
Memory:        16.0 GB

mne:           0.23.0
numpy:         1.20.2 {blas=NO_ATLAS_INFO, lapack=lapack}
scipy:         1.6.3
matplotlib:    3.4.1 {backend=MacOSX}

sklearn:       0.24.1
numba:         0.53.1
nibabel:       3.2.1
nilearn:       0.7.1
dipy:          1.4.0
cupy:          Not found
pandas:        1.2.4
mayavi:        4.7.2
pyvista:       0.29.1 {pyvistaqt=0.3.0, OpenGL 4.1 INTEL-14.7.8 via Intel(R) Iris(TM) Plus Graphics 655}
vtk:           9.0.1
PyQt5:         5.12.3

The pyvista backend seems to work, however the mayavi backend just shows grey background plots.

To reproduce:

(mne-stable) bloyl@507b9dedf1fe ~ % export MNE_3D_BACKEND=mayavi; ipython
Python 3.9.2 | packaged by conda-forge | (default, Feb 21 2021, 05:02:20)
Type 'copyright', 'credits' or 'license' for more information
IPython 7.22.0 -- An enhanced Interactive Python. Type '?' for help.

In [1]: import mne
   ...: from mne.viz import plot_alignment, set_3d_view
   ...:
   ...: print(__doc__)
   ...:
   ...: data_path = mne.datasets.sample.data_path()
   ...: subjects_dir = data_path + '/subjects'
   ...: trans = mne.read_trans(data_path + '/MEG/sample/sample_audvis_raw-trans.fif')
   ...: raw = mne.io.read_raw_fif(data_path + '/MEG/sample/sample_audvis_raw.fif')
   ...: # Plot electrode locations on scalp
   ...: fig = plot_alignment(raw.info, trans, subject='sample', dig=False,
   ...:                      eeg=['original', 'projected'], meg=[],
   ...:                      coord_frame='head', subjects_dir=subjects_dir)
   ...: # Set viewing angle
   ...: set_3d_view(figure=fig, azimuth=135, elevation=80)
Automatically created module for IPython interactive environment
Opening raw data file /Users/bloyl/mne_data/MNE-sample-data/MEG/sample/sample_audvis_raw.fif...
    Read a total of 3 projection items:
        PCA-v1 (1 x 102)  idle
        PCA-v2 (1 x 102)  idle
        PCA-v3 (1 x 102)  idle
    Range : 25800 ... 192599 =     42.956 ...   320.670 secs
Ready.
Using outer_skin.surf for head surface.
Using mayavi 3d backend.

qt.qpa.drawing: Layer-backing can not be explicitly controlled on 10.14 when built against the 10.14 SDK

In [2]:

it will generate a grey window ('TVTK Scene') with no mayavi control bar. I don't think the QT warning is the issue as it also displays with the pyvista backend.

Interestingly if I press cntl-d to quit ipython. the figure shows up while ipython asks Do you really want to exit ([y]/n)?
If I press no, the plot remains but the window is non-responsive.

I should note that mayavi plotting functions (https://docs.enthought.com/mayavi/mayavi/auto/mlab_helper_functions.html#mayavi.mlab.contour3d) seem to work fine.

The reason this is an issue is that mne coreg is still mayavi based and displays the same problem.

[larsoner]

I can't replicate on my macOS system with Intel Iris graphics.

Does python -ic "from mayavi import mlab; mlab.test_plot3d()" work where it displays and you can interact?

What about MPLBACKEND=Qt5Agg python -ic "import matplotlib.pyplot as plt; plt.ion(); plt.plot(0, 0)" where you can interact and zoom?

[bloyl]

Both
python -ic "from mayavi import mlab; mlab.test_plot3d()"
and
MPLBACKEND=Qt5Agg python -ic "import matplotlib.pyplot as plt; plt.ion(); plt.plot(0, 0)"
work fine.

note 2d plotting (matplotlib) is working fine. I was able to run this example without issue (https://mne.tools/stable/auto_examples/visualization/channel_epochs_image.html#sphx-glr-auto-examples-visualization-channel-epochs-image-py)

similarly 3d plotting (pyvista) also seems to be fine. its only the mayavi backend

[larsoner]

Both ... and ... work fine... similarly 3d plotting (pyvista) also seems to be fine. its only the mayavi backend

Okay but it's not a purely Mayavi problem because the mlab.test_plot3d() worked, in the sense that it plotted something like this that you could rotate with the mouse, etc., right?

In other words, Mayavi itself is fine, but somehow our use of Mayavi / the mayavi backend is broken.

Next can you try:

$ python -ic "import mne; from surfer import Brain; brain = Brain('fsaverage', 'lh', 'inflated', subjects_dir=mne.datasets.sample.data_path() + '/subjects')"

You might need to pip install pysurfer.

It just occurred to me that this might be an antialiasing issue since it's broken on most macs. Can you see if setting MNE_3D_OPTION_ANTIALIAS=false as an env var helps?

Your sys.platform is 'darwin', right?

[bloyl]

MNE_3D_OPTION_ANTIALIAS=false solves the problem with mne coreg which was the real show stopped for me.

However there is still something strange going on.

the following behavior isn't changed by MNE_3D_OPTION_ANTIALIAS=false and to confirm sys.platform is 'darwin'

python -ic "import mne; from surfer import Brain; brain = Brain('fsaverage', 'lh', 'inflated', subjects_dir=mne.datasets.sample.data_path() + '/subjects')"
works just fine.

but running the same thing in ipython does this:

mavi_ipython.mov

note similar behavior occurs if I run

export MNE_3D_BACKEND=mayavi; ipython -i ~/Projects/mne-python/mne-python-devel/examples/visualization/plot_eeg_on_scalp.py

which works with python calls.

;---
Since mne coreg works (with MNE_3D_OPTION_ANTIALIAS=false) and I generally use the pyvista backend feel free to close if no-one else can replicate.

[larsoner]

MNE_3D_OPTION_ANTIALIAS=false solves the problem with mne coreg which was the real show stopped for me.

Great! We should add this to the 3D debugging docs. @GuillaumeFavelier want to take a stab?

but running the same thing in ipython does this:

This looks like an ipython<->mayavi bug. I wonder if it would be fixed by doing %gui qt or %matplotlib qt

[bloyl]

MNE_3D_OPTION_ANTIALIAS is on the advanced setup page.

but I only tested in the ipython session where it didn't have an effect. so never tested it against mne coreg

Its up to you 2 if you feel the current doc is sufficient.

[larsoner]

@bloyl I would argue that if it tripped you up, it's probably not sufficient :)

And looking at it, this is listed in the Off-screen rendering with MESA section, not the Troubleshooting section. I think it should live there as well, and probably also specifically mention macOS as possibly needing this parameter.