FIX: Allow eyetrack channels to be used with plot_compare_evoked

[scott-huberty]

Per this discussion post:

https://mne.discourse.group/t/mne-viz-plot-compare-evokeds-for-eyetracking/7816

It seems that eyetrack channels are not an accepted channel type for plot_compare_evokeds. Off the top of my head I can't think of a good reason for why they shouldn't be accepted, so I've gone ahead and implemented a quick fix. I wouldn't mind a moment to really think about this or discuss it with others, but I'm defending my PhD tomorrow so this thinking won't happen for me until next week 😃

A MWE, which is just a proof of concept (in this example the evokeds aren't informative at all). (requires mne 1.6):

import mne
fpath = mne.datasets.eyelink.data_path() /  "freeviewing"
fname = fpath / "sub-01_task-freeview_eyetrack.asc"
raw = mne.io.read_raw_eyelink(fname)

mne.preprocessing.eyetracking.interpolate_blinks(raw, interpolate_gaze=True)
raw.annotations.rename({"dvns": "natural", "dvss": "scrambled"})  # more intuitive
event_ids = {"natural": 1, "scrambled": 2}
events, event_dict = mne.events_from_annotations(raw, event_id=event_ids)

epochs = mne.Epochs(raw, events =events, event_id=event_dict, tmin=0, tmax=20, baseline=None)

evokeds = dict()

evokeds["natural"] = epochs["natural"].average()
evokeds["scrambled"] = epochs["scrambled"].average()

mne.viz.plot_compare_evokeds(evokeds)

[mscheltienne]

The real question to me is: should eyegaze and pupil channels be considered as data channels?

I'm defending my PhD tomorrow so this thinking won't happen for me until next week

No hurry, and best wishes!

[mscheltienne]

For the pupil size, that opens another question when comparing evoked/signals from different subjects/recordings. As it's in AU, and the camera distance/illumination/lens can differ + the subject eye size/shape differ, is it actually comparable?

[drammock]

For the pupil size, that opens another question when comparing evoked/signals from different subjects/recordings. As it's in AU, and the camera distance/illumination/lens can differ + the subject eye size/shape differ, is it actually comparable?

zscoring within subject is a typical answer to this worry. Though researchers also do try to control distance to camera/screen (e.g. with a chinrest and forehead bar)

[larsoner]

The real question to me is: should eyegaze and pupil channels be considered as data channels?

I would say no, they are not "ephys data from the brain" which is what we mean by "data"

https://mne.tools/dev/documentation/glossary.html#term-data-channels

[larsoner]

Can you ad a tiny test that it plots? So add to mne/viz/tests/test_evoked.py that creates an Evoked from eyelink data and calls plot_compare_evokeds. Doesn't matter if the comparison is meaningful.

Also on main did you get a reasonable error /ike that no acceptable channels types were found, or was it cryptic? If it was cryptic we should also improve it to say "no plottable channel types found, plot_compare_evokeds doesn't support plotting channel types found in data: {inst.get_channel_types(unique=True)} or similar.

[scott-huberty]

Can you ad a tiny test that it plots? So add to mne/viz/tests/test_evoked.py that creates an Evoked from eyelink data and calls plot_compare_evokeds.

Sure can do.

Also on main did you get a reasonable error /ike that no acceptable channels types were found, or was it cryptic?

Actually it was an assert error.. I can amend it use an if not clause and raise a more informative error

[larsoner]

Actually it was an assert error.. I can amend it use an if not clause and raise a more informative error

That would be great. I would use inst.set_channel_types(...) to change the channel types to something that can't be plotted. From the looks of it EOG might work for now. Or if you want to pick something more esoteric that we probably never care to plot with plot_compare_evokeds you could go with ias which is internal active shielding information for Neuromag/MEGIN systems.

[larsoner]

... and actually you even could use this set_channel_types approach to pretend the EEG or MEG sensor data are eyetracking pupi/gaze data for the purposes of the test that should now pass that currently fails on main. Reading actual data would be more lines of code but also more relevant. Up to you :)

[scott-huberty]

Update. I added a more informative error and a few more lines to test_plot_compare_evokeds, to test that plotting eye-tracking data works, and that an informative error is thrown if the evoked instance only contains channel types that are not accepted by plot_compare_evokeds (i.e. an evoked instance with only EOG channels`).

[larsoner]

Can you add a line to doc/changes/devel.rst and merge in latest main (or rebase) and then we can merge? Thanks @scott-huberty !

[scott-huberty]

Oh yes thanks for reminding me!

[larsoner]

Thanks @scott-huberty !