VIZ: EEG sensor locations wrong in inset plots

[drammock]

EEG sensor plot insets seem to be broken in some cases. Here are a couple examples. First, from plot_compare_evokeds:

Similar problem with `evoked.plot(spatial_colors=True):

Compare those to the sensor positions w/r/t the head on mne.channels.make_eeg_layout(raw.info).plot() (raw.plot_sensors() gives similar result):

[larsoner]

I don't see this as a blocker for 0.18, which we want to get out tomorrow if possible. So moving the milestone to 0.19.

[massich]

Is this related to #5190 ?

[drammock]

@massich I don't know the layout plotting code well enough to say whether they're related. But these examples should make a good test case for whatever fix comes in for #5190 (or vice versa)

[jona-sassenhagen]

Isn't the issue simply that layouts are calculated using both MEG and EEG channels, and EEG channels (in this specific montage) don't go as deep?

In which case this might be working as it should be.

[drammock]

@jona-sassenhagen to me the issue is that the same set of EEG sensors nearly fills the head with raw.plot_sensors() or mne.channels.make_eeg_layout(raw.info).plot() but appear clustered at the crown of the head in the insets in evoked.plot() or plot_compare_evokeds() calls. IMO they should look the same.

[jona-sassenhagen]

Not sure how to deal with that. One issue here is that we make no absolute guarantees here. Consider outlines='skirt'.

Arguably, the behaviour should match, but maybe the behaviour derived from calculating all layouts simultaneously is the better one.

[larsoner]

Arguably, the behaviour should match, but maybe the behaviour derived from calculating all layouts simultaneously is the better one.

@jona-sassenhagen @drammock can you check my proposal here:

#3987 (comment)

I argue that these plots should always show the sensors reasonably aligned to a ~80cm head. Then it doesn't matter what subselection, etc. you use. For MEG you align using dev_head_t and for EEG you're already in head coords so should be fine. It should be more true to the data and head alignment, and avoid these rescaling issues altogether.

Also related: #5472, #5471

[drammock]

adding another test case for #5472. Would be really nice to be able to do plots like this in the tutorials.

import os
import mne
sample_data_folder = mne.datasets.sample.data_path()
sample_data_raw_file = os.path.join(sample_data_folder, 'MEG', 'sample',
                                    'sample_audvis_raw.fif')
raw = mne.io.read_raw_fif(sample_data_raw_file)
raw.crop(tmax=60).load_data()
raw.plot_sensors(ch_type='eeg', show_names=True)
midline_ch_names = ['EEG {:03}'.format(n) for n in (2, 12, 30, 48, 58, 60)]
midline = raw.copy().pick_channels(midline_ch_names)
midline.plot_sensors(show_names=True)

[agramfort]

I agree we should really fix this :-/

…

[larsoner]

I started looking at one point and it was a bit of a nightmare. But I'll see if I can look again soon

[massich]

so did I when I started digitization I expect plot_sensors to get simplified with the refactorings.