DOC: clean up ECG ICA docstrs

[sappelhoff]

btw: Does anyone know why the automatic ECG channel creation only works for MEG channels? It would be really nice/helpful to also support this for EEG channels.

With references like this one, it seems possible: https://doi.org/10.1016/j.cmpb.2019.105092

[agramfort]

The average of all magnetometers is a good proxy for an ECG channel. It does not well for EEG or planar gradiometers

[sappelhoff]

Test failures are unrelated

FAILED mne/decoding/tests/test_receptive_field.py::test_rf_sklearn_compliance[<ReceptiveField|tmin,tmax:(-1.000,2.000),estimator:<class'sklearn.linear_model._ridge.Ridge'>,fit:False>-check_estimators_unfitted] - AssertionError: Estimator should raise a NotFittedError when calling predict before fit. Either call check_is_fitted(self) at the beginning of predict or set tags.requires_fit=False on estimator tags to disable this check.

• check_is_fitted: https://scikit-learn.org/dev/modules/generated/sklearn.utils.validation.check_is_fitted.html
• Estimator Tags: https://scikit-learn.org/dev/developers/develop.html#estimator-tags

[larsoner]

Even though @agramfort said:

It does not [work] well for EEG or planar gradiometers

It looks like we do allow it for planar gradiometers, so I'll go ahead and merge since the changes more accurately reflect what we currently do. Maybe in some follow-up PR we could consider removing the grad support, or at least making it opt-in instead of default.

[larsoner]

Thanks @sappelhoff !