Vizualization framework for channel specific annotations

[scott-huberty]

Closes #200
Closes #196

(EDIT) starts to close mne-tools/mne-python#8946 . I think we will need to implement the same funcitionality for the matplotlib backend before closing that issue.

[scott-huberty]

This almost works:

Of course, the sample audvis dataset always seems to present the most complex scenario (floating point sfreq, etc) 🙂

If I run this:

import mne
from mne.datasets import sample
mne.viz.set_browser_backend("qt")

data_path = sample.data_path()
meg_path = data_path / "MEG" / "sample"
raw_fname = meg_path / "sample_audvis_raw.fif"
raw = mne.io.read_raw_fif(raw_fname, preload=True)
raw.pick("meg").filter(1, 30)

ch_names = ["MEG 0133", "MEG 0142", "MEG 0143", "MEG 0121"]
onsets = [1, 4.25, 7.25]
durs = [2, 1.75, 1.25]
descs = ["Specific_channels_test", 'BAD_test', 'Good_test']
chs = [ch_names, [], []]
annots = mne.Annotations(onset=onsets, duration=durs, description=descs, ch_names=chs)

raw.set_annotations(annots)
raw.plot(theme="light")

I get this plot

"MEG 0143" is in ch_names, but it is NOT highlighted in the plot. ["MEG 0212", "MEG 0222"] are NOT in the ch_names but they ARE highlighted.

@drammock I think this has to do with our "dirty hack", and that these MEG channels are not ordered in the plot by their number (i.e. "MEG 0133", "MEG 0143", "MEG 0142").

[drammock]

ah, right you are. Our dirty hack will not work. Maybe check if the channel order is separately specified / kept track of somewhere in fig.mne)?

[larsoner]

Aesthetically pleasing despite the indexing bug!

For the indexing I would look at how we figure out when you click on a channel name Y label which channel is actually bad. Indexing correspondence between channel names and what is currently shown should be used there I think

[larsoner]

@scott-huberty did you test interactivity? Using this code:

import mne
raw = mne.io.read_raw_fif(mne.datasets.sample.data_path() / "MEG" / "sample" / "sample_audvis_raw.fif")
raw.annotations.append(raw.first_time + 1, 2, "CH_SPEC", [["MEG 0113"]])
raw.plot()

pressing "a" on the keyboard to enter annot mode and shrinking the annot I see the ch-specific one does not shrink:

Can you replicate?

[scott-huberty]

I did not test that, and can replicate 😅

Looking into it

[larsoner]

Code looks good other than the interactivity issue!

[larsoner]

FYI @scott-huberty I changed your wording in the top-comment from "supersedes" to "closes" because the latter does GitHub closing magic

[scott-huberty]

Going to stop for the day soon, but FYI the fix that we thought we had for this (updating the xData of the FillBetween items) is not actually working (the FillBetween item does not redraw on the plot to reflect its new xData). to be continued...

[larsoner]

xData was just a list, right? There is probably a .update or something that you need to call. But simpler would probably be too call https://pyqtgraph.readthedocs.io/en/latest/api_reference/graphicsItems/plotcurveitem.html#pyqtgraph.PlotCurveItem.setData on the two curves that define the fill between. If that doesn't fix it, calling that then https://pyqtgraph.readthedocs.io/en/latest/api_reference/graphicsItems/fillbetweenitem.html#pyqtgraph.FillBetweenItem.setCurves should do it. It's likely that these set* functions update some attributes under the hood that are not updated when you just assign a new value in the xData list.

[scott-huberty]

Exactly! The setData is what we'll need.I just need to put it in the right place in the codebase. (I tried putting it in the _stop_changed method, but stop_changed isn't being triggered when I adjust an annotation)

[scott-huberty]

the FillBetweenItems (i.e. shaded rectangles) of channels specific annotations will now stay in sync with the annotation region IF you use the start/stop spinboxes to change the annotation region start/stop time. For some reason, dragging or extending an annotation directly in the plotting area triggers a different code path.

I had to rebase with main after #203, and force push my changes to this remote branch. Before doing that I made a local copy of the state of this remote branch, in case I realize that i've done something terribly wrong..

[larsoner]

in case I realize that i've done something terribly wrong..

At the end of the day almost all that matters for our squash-and-merge is the diff on GitHub, so just check to make sure that looks okay and you should be good

[larsoner]

For some reason, dragging or extending an annotation directly in the plotting area triggers a different code path.

This I think is where the logger.debug can/should help us. I'd add logger.debugs in all the click event handlers (now if you want!) and it should become clear which one is triggered by dragging the edge of an annot

[scott-huberty]

Alright that should do it... I'll leave this in your court @larsoner @drammock - enjoy the dev retreat!

[larsoner]

I'll commit these and merge, thanks @scott-huberty !

[larsoner]

Shouldn't nest in tests