plot_compare_evokeds should have the same defaults for `combine`

Currently the function mne.viz.plot_compare_evokeds creates two different plots depending on the number of selected channels:

• If one channel is selected (or present in the evoked), it will plot the ERP.
• If two or more channels are selected (or present), it will plot the GFP.

I think this behaviour is potentially misleading, and I would expect the function to have the same defaults for different channels, i.e. mean of the evokeds. As you can see in #10465 , this created quite a bit of confusion for me (probably mainly due to me being new to ERP analysis and not knowing what GFP meant).

I see this has previously already spiked a bit of discussion (e.g. #3393) and was introduced in #6166. By now, the gfp=... parameter has been removed in favor of the combine=None parameter, which defaults to a different value depending on the present/chosen channels.

I would suggest either

• make combine have the same default for all cases
• make combine an arg, i.e. make users forcefully choose a combination mode

Additionally, the docstring should reflect the importance of the combine parameter.

What are your thoughts on this?

[ktavabi]

On a tangent, I am thinking of simplifying the nave proportions algebra in evoked. comment for the results from plot_compare_evokdes to make a more informative legend. Here is an, e.g., output screenshot

The text in evoked.comment renders unusable image file formats unless figure legends are dealt with on the fly. If that’s the intended usage, then disregard this. Otherwise, I am curious if a fix is a suitable task to piggyback here?

Resulting .png

[agramfort]

@skjerns the docstring at https://mne.tools/stable/generated/mne.viz.plot_compare_evokeds.html says:

so maybe the API in counterintuitive but I think the docstring is explicit no?