[MRG] Require good and bad channels when creating a SpectrumArray object

[mscheltienne]

Failure reported today:

import numpy as np
from mne.datasets import sample
from mne.io import read_raw
from mne.time_frequency import SpectrumArray
from numpy.testing import assert_allclose

fname = sample.data_path() / "MEG" / "sample" / "sample_audvis_raw.fif"
raw = read_raw(fname, preload=False).pick("eeg").load_data()

spectrum1 = raw.copy().crop(0, 10).compute_psd()
spectrum2 = raw.copy().crop(10, 20).compute_psd()

# make sure you have the same freqs!
assert_allclose(spectrum1.freqs, spectrum2.freqs)

# array of shape (n_channels, n_freqs)
data1 = spectrum1.get_data()
data2 = spectrum2.get_data()
data = np.array([data1, data2])  # array of shape (n_spectrum, n_channels, n_freqs)
data = np.average(data, axis=0)

# recompute the Spectrum object
spectrum = SpectrumArray(data, spectrum1.info, spectrum1.freqs)
spectrum.plot()

Raises IndexError: index 59 is out of bounds for axis 0 with size 59 in spectrum.plot, i.e. the plotting functions expects the good and bad channels. And indeed, spectrum1 or spectrum2 include the bads, i.e. the underlying array as shape (60, ...).

Now, if instead you include the bad channels:

data1 = spectrum1.get_data(exclude=())
data2 = spectrum2.get_data(exclude=())

It will also raise but this time in the creation of the SpectrumArray: ValueError: The number of channels in data(60) must match the number of good data channels ininfo (59).

There is a clear inconsistency in what SpectrumArray expects and BaseSpectrum.plot expects. I think the cleaner fix is to have SpectrumArray require the good and bad channels; matching the behavior obtained by Spectrum when running inst.compute_psd(...).

[larsoner]

This one looks real:

FAILED mne/time_frequency/tests/test_spectrum.py::test_spectrum_array_errors - AssertionError: Regex pattern did not match.
 Regex: 'number of channels.*good data channels'
 Input: 'The number of channels in `data` (4) must match the number of good + bads data channels in `info` (5).'

[drammock]

just a few typos. LGTM!