ENH Generalized rot90

[erensezener]

This PR generalizes the rot90 function to rotate an arbitrary plane (defined by axes) of a multidimensional array. Our implementation depends on the generalized flip function introduced in PR #7346.

This PR fixes #6506.

This pull request is part of a BCCN Berlin student project by @denisalevi @ge00rg @erensezener

[shoyer]

This also seems like a good idea, but like for flip, please also run this by the mailing list (could be in the same post).

[madphysicist]

You should A) add a check for len(axes) == 2 (and a test that it raises appropriately), B) change this to check that neither axis is >= ndim. Checking for ndim >= 2 is no longer necessary.

Also, if both axes are the same you may want to either raise or return the input array. Either way, you should document how you do it.

[madphysicist]

As @mhvk noted for your previous PR, you should also add support for negative axes.

[denisalevi]

added A) check for len(axes) in 1a52123

added B) removed check for ndim >= 2 in e5b5b96 and added checking for ndim >=2 and a test for it in aa0d4bf

added raise for same axes and a test for it in 83147ee

[denisalevi]

oh and I added negative axes support in ndim check here b277c31

and tests here 2d56768

[charris]

Updated.

[charris]

Needs mention in doc/release/1.12.0-notes.rst. Also squash the 24 commits into one or two. You can use git rebase -i HEAD~24 for that followed by a force push git push --force origin <your branch>.

[charris]

Need .. versionadded:: 1.12.0

[charris]

Why is flip here? Isn't that another PR?

[erensezener]

rot90 uses flip so it is based on the other PR. After squashing the commits on the flip PR, we will rebase the rot90 branch on the flip branch.

[madphysicist]

@shoyer, @seberg, @charris. I just wanted to bring up my comment on the flip PR #7347 (comment) again since I don't think there's been a response one way or the other. Is this the right file for these functions given that they do not pertain to 2D arrays? Would function_base be more appropriate?

[shoyer]

@madphysicist Yes, I agree -- it would probably make sense to move the generalized versions of these functions out of twodim_base.py.

[charris]

@erensezener It would be good to move the functions/tests and squash the commits. Same for #7346. Maybe you should do them together as on PR?

[charris]

You should add this to the documentation like was done for flip.

[erensezener]

@charris Do you mean the release note? It already exists.

[charris]

I was thinking of the official documentation, but I see the function already exists and is there.

Should probably document that the return is a view whenever possible, which AFAICT, is always the case unless wrapping a subtype does something special. I assume flip is the same.

[madphysicist]

Looking at #7421 as an example, it may be a good idea to add a note in the highlights section.

[erensezener]

@charris how about:
Returns
-------
y : ndarray
A view of m, which is rotated by 90 degrees in the plane specified by axes.

Also, should we add a note in the highlights section as @madphysicist suggests?

[jaimefrio]

This will return the array as is if k = 0, which can lead to subtle bugs, see #5258, #5260 and #5468 for some context. Could you please make it always return a view of the input, i.e. return m[:]? It may be worth mentioning it in the documentation, e.g. in the "Returns" section make it "rotated view of the array."

[erensezener]

@jaimefrio implemented your suggestions

[erensezener]

@charris this is ready for merge as far as I can see.

[charris]

Thanks @erensezener .