Minimap hexels can have false-colour halos

[caiw]

E.g.

The reason is this:

I've determined that this is a bug in mne (either design or implementation).

Rather than taking the supplied Colormap and applying it to the data values, it instead takes 256 sample values and uses the Colormap to create a lookup table (create_lut()), and as far as I can see it then uses the LUT to interpolate between colours. So no amount of clever work with the Colormap itself will fix this bug. The problem simply arises when there are lots of different transforms in the show_transforms argument.

So I'm marking this as a partial fix, and I'll create a new issue to track the halos, and put this text in there too.

In other words, I believe this is a bug in MNE, and we can't fix it without MNE fixing it or without overriding their code (which I don't think is worth it).

[caiw]

@neukym We could spend time creating a MRE and submit it as a bug to MNE, to see if they'll fix it? I feel like quantising data values to 256 levels and then interpolating between them is "wrong", as (like in our case) it displays values which are not present in the source data. And it shouldn't be necessary - there's a perfectly good Colormap right there in the same scope. But maybe it's for performance reasons.

Depends how high priority this bug is.

[neukym]

I think this is important, but low priority. Let's fix the other plotting issues first as we have a bit of a backlog.

[caiw]

For future reference, I tried a few tricks like automatically interleaving transparency into the colormap to thwart the interpolation (f4d45ba, b9178f1, acb18da), but none of them bypassed this bug because of the LUT thing.