Mne-python: Set all types to False in pick_types

[cbrnr]

Would it make sense to set all arguments in pick_types to False by default? In other words, currently everything is False except for meg. I know where this is coming from, but for people working with EEG, NIRS, or other non-MEG signals, we always have to include meg=False in addition to e.g. eeg=True.

[agramfort]

deprecation cycle will be a huge pain. I feel fully guilty for this but
it's trade off.
If you have only EEG data it's not a pb though no?

>

[larsoner]

+1 for doing it. I wish we had done it 5 (10?) years ago but it still seems worth it in the long run. I don't think actually it will be too terrible for users (many probably will have set it explicitly anyway), but even forcing them to add True/False (when MEG channels are present) will be more explicit in their uses anyway. A potential benefit (aside from consistency) is that it could actually help them uncover some silent bugs in their code.

[drammock]

So then raw.pick_types() with all default arguments would be an error? Or a raw object with zero channels? :)

[larsoner]

Yeah, I suppose so. Kind of a weird side effect. But if you don't ask to pick anything it's probably better that it throws an error than it silently pick one channel type...

[drammock]

if you don't ask to pick anything it's probably better that it throws an error than it silently pick one channel type...

Agreed. +1 for embarking on the painful deprecation effort.

[hoechenberger]

+1 from my end for doing this, the current behavior always felt weird to me (an EEG user).

[hoechenberger]

@agramfort

If you have only EEG data it's not a pb though no?

Valid comment though. But since simultaneous recordings with different methods are gaining in popularity, I believe addressing this issue will become more and more important in the future.

[cbrnr]

Alright, so it looks like we agree that this must be done. I'll submit a PR.

[dengemann]

I think this is a good idea. +1 from my side.

[agramfort]

before fully diving into this let's have a clear plan as it's a lot of work.

In Master:

If you have EEG/MEG:
raw.pick_types(meg=False, eeg=True) -> EEG only
raw.pick_types() -> MEG only

If you have EEG:
raw.pick_types() -> Error?

If you have MEG:
raw.pick_types() -> MEG only

what behavior do you expect in the next release and the next next one?

[cbrnr]

The next release (0.21) should behave identically to 0.20, but a DeprecationWarning should be raised (something like "The defaults of pick_types will change in the next release. All signal types will be set to False by default (currently, meg=True by default).").

Release 0.22 should implement the new behavior. Your examples will be:

If you have EEG/MEG:

raw.pick_types(meg=False, eeg=True)  # EEG only, will work as before
raw.pick_types()  # error (no signals picked)
raw.pick_types(eeg=True)  # new option to pick EEG only

If you have EEG:

raw.pick_types()  # error (no signals picked)
raw.pick_types(eeg=True)  # need to select EEG

If you have MEG:

raw.pick_types()  # error (no signals picked)
raw.pick_types(meg=True)  # need to select MEG

[hoechenberger]

I've converted @agramfort's list to a table, which we can potentially amend as the discussion progresses.

| Type | pick_types() call | master behavior | proposed new behavior |
|-----------|---------------------------------|-----------------|-----------------------|
| EEG & MEG | pick_types(meg=False, eeg=True) | EEG only | |
| EEG & MEG | pick_types() | MEG only | |
| EEG | pick_types() | Error? | |
| MEG | pick_types() | MEG only | |

[cbrnr]

What is the current behavior when I pick types that don't exist? Does it raise an error? Or pick nothing (i.e. return a zero-channel object)? Or is this a no-op?

[hoechenberger]

Updated table based on @cbrnr's and @agramfort's comments – hope I copied everything correctly.

| Type | pick_types() call | master behavior | proposed new behavior |
|-----------|-----------------------------------|-------------------|-----------------------|
| EEG & MEG | pick_types(meg=False, eeg=True) | EEG only | EEG only |
| EEG & MEG | pick_types() | MEG only | error |
| EEG & MEG | pick_types(eeg=True) | EEG & MEG | EEG only |
| EEG | pick_types() | Error? | error |
| EEG | pick_types(eeg=True) | EEG & MEG | EEG only |
| MEG | pick_types() | MEG only | error |
| MEG | pick_types(meg=True) | MEG only | MEG only |

[agramfort]

to be tested. I would really start by writing tests here.

>