[BUG] Coordinate Frame not Saved

[alexrockhill]

I'm not sure if I'm missing something obvious but it looks like the coordinate frame doesn't survive the round trip to and from disk. I'm not sure why that is.

import numpy as np
import mne

for ch_type in ('eeg', 'seeg', 'ecog', 'dbs'):
    raw = mne.io.RawArray(
        np.random.random((10, 1000)),
        mne.create_info([f'ch{i}' for i in range(1, 11)],  1000, [ch_type] * 10))
    montage = mne.channels.make_dig_montage(
        {ch: np.random.random((3,)) for ch in raw.ch_names}, lpa=[1, 0, 0], nasion=[0, 0.1, 0], rpa=[-1, 0, 0],
        coord_frame='mri')
    raw.set_montage(montage)
    raw.save('tmp-raw.fif', overwrite=True, verbose=False)
    print(ch_type)
    print('Before: ', raw.info['chs'][0]['coord_frame'])
    info = mne.io.read_info('tmp-raw.fif')
    print('After : ', info['chs'][0]['coord_frame'])

Note, it appears this is specific to seeg, ecog and dbs, it works for eeg.

[alexrockhill]

To update, it looks like this was on purpose but I think it would be better to assume that seeg, ecog or dbs channel positions are in head coordinate frames because if they are set using inst.set_montage(montage) they will be converted. To my knowledge, there are no other public APIs to set the channel positions so it seems like they should be assumed to be in head if read from a disk.

[alexrockhill]

@larsoner, this is in #9585, just wanted to make sure you were aware and thought this was reasonable

[larsoner]

Yeah I think it's just an oversight that we never added these channel types to the inferred-coord-frame list

[alexrockhill]

Great, just making sure it wasn't for a reason I wasn't thinking of