Transition to new import convention for sckit-image?

[mkcor]

How could the content be improved?

Hi all,

Since I just merged scikit-image/scikit-image#7024 by @lagru, I thought I should propagate here the recommendation that we use import skimage as ski throughout the lesson.

See: https://github.com/scikit-image/scikit-image/blob/main/CONTRIBUTING.rst#stylistic-guidelines

[tobyhodges]

Thank you for taking time to keep the lesson up-to-date with the latest changes in scikit-image, @mkcor.

I am concerned that this would introduce inconsistency in the way we are handling imports and might increase cognitive load. In most cases, we are importing some submodule of skimage rather than the whole thing (by my count, 14 of 19 scikit-image-related imports in the lesson import skimage.<something>). Aliasing skimage to ski would not be possible in these cases, and I think it would be a bit confusing for learners to see ski used in some cases but skimage used elsewhere.

So the question perhaps becomes: do we want to switch over to importing all of skimage instead of only specific submodules? I think @datacarpentry/image-processing-curriculum-maintainers talked about this once before, though I cannot find the notes now, and concluded that we are modeling better practice by importing specific submodules as needed.

[mkcor]

Hi @tobyhodges,

I think https://github.com/orgs/datacarpentry/teams/image-processing-curriculum-maintainers talked about this once before, though I cannot find the notes now, and concluded that we are modeling better practice by importing specific submodules as needed.

For educational materials, I'm also in favour of importing specific submodules. So let's forget about import skimage as ski. Do we want import skimage.filters or from skimage import filters though? I used to use the latter in scikit-image tutorials (e.g., from skimage import color, filters).

Currently, file episodes/06-blurring.md contains:

import skimage                                                                 
import skimage.filters

and

from skimage.filters import gaussian

elsewhere.

I'm happy to update #294 with import fixes if we agree on a convention for importing submodules. Thanks!

[tobyhodges]

I believe we prefer the import skimage.filters form because it is more explicit, making it easier for learners to identify where a particular function or submodule has come from.

I should really expand our lesson's CONTRIBUTING.md to mention some of these stylistic things. E.g. also #291 and #288

[mkcor]

Thank you, @tobyhodges! Done f477193.