ENH: multiple colors for ica.plot_sources(evokeds)

[marsipu]

Reference issue

For #9417

What does this implement/fix?

This is supposed to make the differentiation of multiple ICA-components displayed for evokeds easier by displaying them in different colors.

Questions

Are there more ICA-Categories than EOG/ECG which would be displayed simultaneously to consider?
In this draft I accounted for different categories by assigning a different colormap for each one.

[hoechenberger]

Good start! I would differentiate categories by line style and components by color 💪

[marsipu]

Good start! I would differentiate categories by line style and components by color 💪

Thank you for the idea. How big is the possible number of different categories I should account for?

[hoechenberger]

So far I've only seen two in real life (ECG and EOG), not sure if we even support anything else right now? 😅

[hoechenberger]

Thanks @marsipu! Can you touch one of the ICA tutorials so we can actually see the effects in the generated report?

https://mne.tools/stable/auto_tutorials/preprocessing/40_artifact_correction_ica.html?highlight=ica#using-a-simulated-channel-to-select-ica-components might be a good place!

By "touching" I really mean, simply changing a tiny little thing, e.g. improving phrasing somewhere, to enforce re-rendering of the tutorial

[marsipu]

Where do I find the generated report?

[drammock]

[marsipu]

Thank you! I also created this gist to test multiple component-categories.

[hoechenberger]

It's working! 🥳

https://29347-1301584-gh.circle-artifacts.com/0/dev/auto_tutorials/preprocessing/40_artifact_correction_ica.html#using-a-simulated-channel-to-select-ica-components

I'm not a big fan of the color map though, can we try to see what it looks like e.g. with tab10?

[agramfort]

anyone feel free to MRG if happy. thx @marsipu

[drammock]

I'm not a big fan of the color map though, can we try to see what it looks like e.g. with tab10?

I tend to agree with @hoechenberger here, that we should stick with MPL default color choices (which in this case means tab10) unless we have a good reason not to. I also found the gist example a little... overwhelming; it seemed to me difficult to keep track of which ones were ECG and which were EOG, etc. I think that particular visualization would be better if color (rather than linestyle) were held fixed within a component type (so, ECG component 1 is blue solid, ECG 2 is blue dashed; EOG1 is orange solid, EOG2 is orange dashed; etc). However, I acknowledge that this is a nitpick verging on bikeshedding, especially since I don't think it's all that common to view multiple component types at once (?). So in that light maybe it is better to [hold linestyle fixed and vary color] within a given component type, since the most common use case is to show 1-3 components of the same kind (EOG, ECG, etc) and nothing else. But having talked myself out of the suggestion to switch how color/style are varied, I still think tab10 is a better choice than rainbow.

[marsipu]

@drammock thank you for the feedback. I agree that especially the example from the gist doesn't proof the point of differentiating the components more easily. I also think that showing multiple component-types together is probably more rare than multiple components from the same type. At the moment, I just use the 4 named matplotlib linestyles, so displaying more than 4 different components with the same color but different linestyles would then lead to the original problem that components could not be differentiated (these could of course expanded with custom linestyles, but that would maybe not help much in terms of differentiation). So as you suggest I would keep it the way it is currently implemented and change the colormap to tab10 (for excluded_components>10 this would result in equal colors again, but so many excluded components are probably unlikely, right?).

[marsipu]

Colormap tab10:

[hoechenberger]

This looks fantastic!

[drammock]

in it goes! Thanks @marsipu

[marsipu]

Thank you:). Somehow #9417 wasn't closed

[hoechenberger]

Thanks @marsipu