find_bad_channels_maxwell breaks if during some chunk all channels are flat

Hello everyone!

I faced with a problem using mne.preprocessing.find_bad_channels_maxwell(). At some point it returns an error:

        Interval 103:  510.000 -  514.999
            Flat ( 2): MEG1243 MEG1532
        Interval 104:  515.000 -  519.999
Traceback (most recent call last):
  File "<input>", line 172, in <module>
  File "<decorator-gen-415>", line 24, in find_bad_channels_maxwell
  File "E:\Egor_Levchenko\Software\Miniconda\envs\MCCA-MEG\lib\site-packages\mne\preprocessing\maxwell.py", line 2190, in find_bad_channels_maxwell
    delta = chunk_raw.get_data(these_picks)
  File "<decorator-gen-170>", line 22, in get_data
  File "E:\Egor_Levchenko\Software\Miniconda\envs\MCCA-MEG\lib\site-packages\mne\io\base.py", line 819, in get_data
    picks = _picks_to_idx(self.info, picks, 'all', exclude=())
  File "E:\Egor_Levchenko\Software\Miniconda\envs\MCCA-MEG\lib\site-packages\mne\io\pick.py", line 1058, in _picks_to_idx
    '(%r)' % (orig_picks,))
ValueError: No appropriate channels found for the given picks ([])

I checked the code and my data. I think the problem is that my data is corrupted a little bit and I have all flat channels for around 20 seconds. After detecting the flat channels the code takes only good_meg_picks which are not in chunk_flats. As a result, these_picks appears to be empty if all channels are in chunk_flats:

https://github.com/mne-tools/mne-python/blob/8b9616dfbc377ce468a6d52b43f7a973eac47cab/mne/preprocessing/maxwell.py#L2167-L2175 https://github.com/mne-tools/mne-python/blob/8b9616dfbc377ce468a6d52b43f7a973eac47cab/mne/preprocessing/maxwell.py#L2187

You can see a snapshot of raw data below:

I think this is more a problem of my data, but I guess it should be nice to ignore “all flat” chunks with a warning or something like that. Correct me please if I am wrong and thank you for any comments!

System properties:

OS: Windows Server 2012 R2
python=3.7.7
mne=0.22.0

[larsoner]

mne=0.22.0

I tried running it, but it is strange that flat segments are not fully detected:

I cannot reproduce on 0.22.0:

import mne
raw = mne.io.read_raw_fif('sub-1011_ses-20190430_task-vid2_meg.fif').load_data()
annot, bads = mne.preprocessing.annotate_flat(raw, picks='meg', bad_percent=100, verbose=True)
raw.set_annotations(annot)
raw.plot()

outputs:

Marking 8.85% of time points (5 segments) and 0/306 channels bad

Either way, though, I recommend updating to the latest release 0.23 as 0.22 is no longer supported. There I get the same thing:

Can you try on 0.23? If you have the same problem with the code snippet above I’ll boot into Windows to see if I can replicate there.

[larsoner]

Looks like there are likely bugs with annotate_flat and find_bad_channels_maxwell, I’ll look into it!