Magnetometer positions look odd in PSD plot

[hoechenberger]

Steps to reproduce

Let's first visualize the magnetometer channel positions via Raw.plot_sensors():

import mne
from mne.datasets import sample


data_path = sample.data_path()
fname = data_path + '/MEG/sample/sample_audvis_raw.fif'
raw = mne.io.read_raw_fif(fname)
raw.plot_sensors(ch_type='mag')

This produces:

Looking good. Now let's plot PSDs of the data.

raw.plot_psd()

Result cropped so it only shows the sensor positions (top to bottom: mag, grad, eeg):

The magnetometer sensor layout looks incorrect, and doesn't match the one produced by plot_sensors(). Am I missing something here?

cc @agramfort

[agramfort]

to me yes we have a bug. any idea @larsoner ?

…

[larsoner]

Agreed it's a bug, no idea the cause. Someone will have to dig into the viz code to figure out why this is happening...

[agramfort]

this allows to clearly show the bug:

import mne
import os

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, preload=True)
raw.plot_psd(picks=['grad'])
raw.plot_psd(picks=['mag'])
raw.plot_psd(picks=['meg'])

[agramfort]

and the bad news is that the bug is in the 0.20 stable release :( :(

[larsoner]

I'll git bisect and see how hard it is to fix

[agramfort]

thanks

[larsoner]

Looks like it was 9e78506 / #7486, cc @drammock -- I'll see if I can figure out what the problem is