let me know when I should look!

Codecov Report

Merging #253 (1d6e639) into master (87566cf) will decrease coverage by 0.00%.
The diff coverage is 0.00%.

@@            Coverage Diff             @@
##           master     #253      +/-   ##
==========================================
- Coverage   52.93%   52.92%   -0.01%     
==========================================
  Files          23       23              
  Lines        2747     2749       +2     
==========================================
+ Hits         1454     1455       +1     
- Misses       1293     1294       +1     

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 87566cf...1d6e639. Read the comment docs.

@jasmainak any idea why I seem to produce an MNE vertex time course (shown below) that looks different from the one on our documentation page?

any idea why I seem to produce an MNE vertex time course (shown below) that looks different from the one on our documentation page?

different MNE version :-) I don't think we can really guarantee exactly the same result. But it might be worthwhile to check if the source localization looks correct by doing stc.plot and then picking a vertex manually using the pyvista backend. I wanted to do this myself but for some reason the renderer is broken on my mac ...

different MNE version :-) I don't think we can really guarantee exactly the same result. But it might be worthwhile to check if the source localization looks correct by doing stc.plot and then picking a vertex manually using the pyvista backend. I wanted to do this myself but for some reason the renderer is broken on my mac ...

After we resolve this, it might be a good idea to plot the dipole in source space for this example to show how we interpret orientation in the HNN framework. Is there an efficient way to do this that doesn't require a fancy graphics backend?

Is there an efficient way to do this that doesn't require a fancy graphics backend?

MNE does have a 'matplotlib' backend for plotting stc. It might be pretty slow though ... and neglected. OpenGL makes it fast. Speed comes at a cost ... always :-P

see here the plot: https://mne.tools/stable/auto_tutorials/source-modeling/plot_visualize_stc.html#volume-source-estimates just above "Volume Source Estimates"

see here the plot: https://mne.tools/stable/auto_tutorials/source-modeling/plot_visualize_stc.html#volume-source-estimates just above "Volume Source Estimates"

Can the matplotlib backend render a vector inverse solution?

Can the matplotlib backend render a vector inverse solution?

honestly don't know, you just have to try ...

This is fantastic stuff @rythorpe !! You should rebase to get @ntolley 's mod to the spike raster. I tried to simplify the plotting sections, but I see why you've done what you have. This is an advanced demo, so more explicit python is ok imho. So cool!

Could be some temporary issue. The MPIBackend might be fragile with respect to timings etc. I would actually stick to JoblibBackend for the example if we can. So we don't get stuck with MPI issues in most examples except the one meant to demo it. I restarted CircleCI.

Did you check if you can even use multiple cores on Circle? If yes, and if MPI helps make the examples faster, we should consider having a function to set the backend globally and then set it in CircleCI so that the user does not see it in the examples and they can just run it without MPI. But anyhow, this is for 0.2

If it fails again, try to do what I suggest above. But if you really want to stick to MPI, you can make the sphinx output verbose here and then make a push so you see all the outputs not just what is coming from the error log.

@rythorpe if you do add a static image, please directly link to @cjayb 's image above or post a comment with the image and then link to it. That way we don't pollute the git history with images. It's a bit of a hack but will do the job for now

You can press 'i' to toggle the interface and get a cleaner image

@rythorpe if you do add a static image, please directly link to @cjayb 's image above or post a comment with the image and then link to it. That way we don't pollute the git history with images. It's a bit of a hack but will do the job for now

What do you think of 47e0181? I couldn't figure out how to get a formal caption to show up with an in-line figure in a sphinx-generated document, so I added more text to the paragraph prior. Also, I'm thinking that @cjayb's figure will do nicely since it looks very similar to what the user will produce should they run the code (unlike a cleaned-up version of the figure that I might create).

Alternatively, we could add these figures (lateral and top view, respectively) if anyone prefers:

I think it looks good. Do you not want to plot on the inflated surface? I can't see the label otherwise ...

I think it looks good. Do you not want to plot on the inflated surface? I can't see the label otherwise ...

I'm hesitant to plot this on an inflated surface because the whole point of this exercise is to prompt users to observe how real cortical structure can influence the orientation of a dipolar inverse solution.

but you don't lose that information in inflated surface. Sulci and gyri are shown in different colors ...

but you don't lose that information in inflated surface. Sulci and gyri are shown in different colors ...

This might be a conversation better had over video call. In theory, you're correct. In practice, however, I'm afraid that new users won't think twice about merely plotting a distributed inverse solution on an inflated brain, selecting the most responsive vertex (or mean or principle component vertex) and orienting the time course in HNN independent of known somatotopic organization and corresponding landmarks.

humm ... okay, let's discuss over video. I'll merge this PR when the CI comes green.

I also noticed that you used smoothing=0. I know smoothing can be confusing but it's how folks have been plotting MNE solutions for years. If you see patchy dot-like activity at a certain location, it's not because the activity there is truly 0 but rather that point may not have been sampled in the source space.

I think I'm with @rythorpe on not using the inflated surface. However, I just realised my screenshot is actually showing the anterior bank of whatever sulcus is posterior to the post-central one! Is that where the peak vertex is located? It could very well be, although physiologically the source must be on the "next sulcus over", i.e., the posterior bank of the Rolandic fissure, as we describe in the text...

This is getting complicated and potentially confusing! We shouldn't draw too much attention to this, it's supposed to be helpful.

I think I'm with @rythorpe on not using the inflated surface. However, I just realised my screenshot is actually showing the anterior bank of whatever sulcus is posterior to the post-central one! Is that where the peak vertex is located? It could very well be, although physiologically the source must be on the "next sulcus over", i.e., the posterior bank of the Rolandic fissure, as we describe in the text...

This is getting complicated and potentially confusing! We shouldn't draw too much attention to this, it's supposed to be helpful.

@cjayb since our method extracts the primary principle component time course from the post-central gyrus label, it's hard to tell exactly where the source we model is located. From the literature, you're right that it should be somewhere in area 3b/1 of the hand representation (i.e., the omega-shaped 'ripple' in the CS). I know it's late for you, but I invited you to @jasmainak and my quick zoom call later today if you wish to join.

How about these two images showing the label of interest and MNE activation, respectively.

I'm learning that formatting with reStructured text can be frustrating. I've updated the code to produce the figures shown in my previous comment and updated the corresponding links, though I'm open to suggestions for improving the formatting.

Thank you @rythorpe ! It took longer than expected but final product looks great!

Looks amazing, really great touch with the figures!!