Allow random cmap creation for map.get_color_map

[AhmetNSimsek]

import siibra
from nilearn import plotting
mp = siibra.get_map("deep bundles", "mni152")
cmap = mp.get_colormap(allow_random_colors=True)  # keyword has to be specified
plotting.view_img(mp.fetch(), cmap=cmap, symmetric_cmap=False).open_in_browser()

Uses the number of regions to determine the random seed for reproducibility.

[codecov]

Codecov Report

Attention: Patch coverage is 10.00000% with 9 lines in your changes missing coverage. Please review.

Project coverage is 45.34%. Comparing base (bc7c1d5) to head (f88a7d5).
Report is 9 commits behind head on main.

Files with missing lines	Patch %	Lines
siibra/volumes/parcellationmap.py	10.00%	9 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #641      +/-   ##
==========================================
- Coverage   45.41%   45.34%   -0.07%     
==========================================
  Files          75       75              
  Lines        7212     7229      +17     
==========================================
+ Hits         3275     3278       +3     
- Misses       3937     3951      +14     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[dickscheid]

I would not return random colors if a colormap is absent. I think it is sufficient if siibra fails, but with a readable Exception which catches this case. An alternative would be to return a commonly used map, so that at least the result is always the same.

[AhmetNSimsek]

I would not return random colors if a colormap is absent. I think it is sufficient if siibra fails, but with a readable Exception which catches this case. An alternative would be to return a commonly used map, so that at least the result is always the same.

@dickscheid I disagree. When plotting, the user can use a default colormap themselves. This is an instance method which implicitly suggests there is a preconfigured colormap for this specific ParcellationMap.

Using a random seed, as the pr does, ensures reproducibility too. And we inform the user it is a non native colormap. Additionally, there are maps that are not fully covered. This PR also ensures a reproducible way to color them instead of setting to them the background.

[xgui3783]

I think the flag should be renamed to fill_uncolored or fill_unset or fill_undefined, and defaults to False (which it already does)

This indicates the action, fill color map, and the target, areas without a color.

This flag also implies that this action is not random, but deterministic. (though it may change from version to version of siibra-python)

[dickscheid]

@AhmetNSimsek I agree with you that a default colormap is not needed, but I also think that a random colormap is not useful either. I definitely prefer to raise a well defined exception if we don't have a parcellaion-specific colormap to provide.

[AhmetNSimsek]

@AhmetNSimsek I agree with you that a default colormap is not needed, but I also think that a random colormap is not useful either. I definitely prefer to raise a well defined exception if we don't have a parcellaion-specific colormap to provide.

This is already done, please see 400520d.
However, this does not address the issue that if a map has only partial coverage, some regions have no color values but others do, the regions without colors will have the color of the background. I don't think this is correct.

Also, a reproducible random map is better than a standard colormap, IMO, since the odds that neighbor regions have similar colors are very slim. Not the mention, the user has to explicitly set it to True, in order to obtain such a colormap.

[dickscheid]

@AhmetNSimsek I agree with you that a default colormap is not needed, but I also think that a random colormap is not useful either. I definitely prefer to raise a well defined exception if we don't have a parcellaion-specific colormap to provide.

This is already done, please see 400520d. However, this does not address the issue that if a map has only partial coverage, some regions have no color values but others do, the regions without colors will have the color of the background. I don't think this is correct.

Also, a reproducible random map is better than a standard colormap, IMO, since the odds that neighbor regions have similar colors are very slim. Not the mention, the user has to explicitly set it to True, in order to obtain such a colormap.

I was not aware that this should actually solve the missing region problem. Agreed then. Regarding optimal contrast of neighboring region colors, it is a complicated optimization problem so let's hope the random solution is sufficient.