BUG: Averaging/plotting EpochArray/EvokedArray from ICA.get_sources leads to errors

[kdoelling1919]

I'm using ICA.get_sources() to output an EpochArray: epochs_ic. If I tryevoked_ic=epochs_ic.average() I get the following error:

/Users/keithd/mne-python/mne/io/pick.pyc in pick_info(info, sel, copy)
    387         return info
    388     elif len(sel) == 0:
--> 389         raise ValueError('No channels match the selection.')
    390 
    391     info['chs'] = [info['chs'][k] for k in sel]

ValueError: No channels match the selection.

I can deal with this issue by explicitly giving the picks: something like evoked_ic = epochs_ic.average(picks=np.arange(ncomps))

Then if I try to do a butterfly plot of the evoked data I get a different error

/Users/keithd/anaconda2/lib/python2.7/site-packages/matplotlib/gridspec.pyc in __getitem__(self, key)
    152                     k1 += nrows
    153                 if k1 >= nrows or k1 < 0 :
--> 154                     raise IndexError("index out of range")
    155                 row1, row2 = k1, k1+1
    156 

IndexError: index out of range

That error I can't fix with the previous trick. Below is some code that reproduces both errors with sample data. I think both errors are likely related to the way ICA channels are picked in /mne/io/picks.py. It seems as though the ICA channel type is not listed at all as an option in _PICK_TYPES_DATA_DICT. I think I could fix this myself if others agree it is indeed a bug and I'm not missing something obvious. Let me know what you think!

_Sample Code to reproduce error:_

#import mne and sample data
import sys
import mne
from mne.datasets import sample  # noqa
data_path = sample.data_path()
raw_fname = data_path + '/MEG/sample/sample_audvis_filt-0-40_raw.fif'
raw = mne.io.read_raw_fif(raw_fname, add_eeg_ref=False)

# get the events
events = mne.find_events(raw, stim_channel='STI 014')
event_id = dict(aud_l=1, aud_r=2)  # event trigger and conditions
tmin = -0.5  # start of each epoch (500ms before the trigger)
tmax = 0.5  # end of each epoch (500ms after the trigger)
# pick just the meg channels
picks = mne.pick_types(raw.info, meg=True, eeg=False, eog=False, stim=False,
                       exclude='bads')
# make epochs
epochs = mne.Epochs(raw, events, event_id, tmin, tmax, proj=True, picks=picks,
                    preload=True,add_eeg_ref=False)
# run ICA on the epochs
ncomps = 24
ica=mne.preprocessing.ica.ICA(max_pca_components=ncomps)
ica.fit(epochs)
# put sources into epoch array
src_data=ica.get_sources(epochs)
# try to plot evoked array
evoked_ic = src_data.average(); # get error here, can be avoided by including picks=np.arange(ncomps)
evoked_ic.plot() #then I can get an error here. This is not fixable by the same addition

[jaeilepp]

Fix in #3649
This happened because the ICA components are assigned as misc channels. Previously misc channels could not be plotted in evoked.

[kdoelling1919]

Thanks @jaeilepp !! This fixes the error with evoked_ic.plot()

It doesn't fix however the error generated from evoked_ic = src_data.average()

While changing it to evoked_ic=src_data.average(picks=np.arange(ncomps)) gets around the error, it seems like we should fix that bug as well in the first place. I guess, it should be a pretty similar issue.

[jaeilepp]

The documentation of average, says that only data channels are picked by default. User can also give picks to include misc channels. I'm not sure if we want to change that.

[jona-sassenhagen]

Maybe we should amend the warning on epoching if the object is detected to be ICA sources. I.e., the warning could tell the user about @kdoelling1919 's trick or something like that.

[kdoelling1919]

I'm happy to do this, if others agree it would be useful.

[agramfort]

+1 to improve docstrings