BUG: np.cross() warns about arrays of vectors when used with two simple 2-d vectors #26620

megies · 2024-06-05T10:47:06Z

Describe the issue:

np.cross() warns about arrays of vectors when used with simply two 2-d vectors (this use case is even in the "Examples" section of np.cross() docstring). I believe, that check whether a warning should be shown is missing a check for number of dimensions of the input.

numpy/numpy/_core/numeric.py

Lines 1645 to 1653 in 42e9aa2

    
           if a.shape[-1] not in (2, 3) or b.shape[-1] not in (2, 3): 
        
               raise ValueError(msg) 
        
           if a.shape[-1] == 2 or b.shape[-1] == 2: 
        
               # Deprecated in NumPy 2.0, 2023-09-26 
        
               warnings.warn( 
        
                   "Arrays of 2-dimensional vectors are deprecated. Use arrays of " 
        
                   "3-dimensional vectors instead. (deprecated in NumPy 2.0)", 
        
                   DeprecationWarning, stacklevel=2 
        
               )

I feel like there should be a change, either

change the warning message to something like '2-dimensional vectors are deprecated. Use 3-dimensional vectors instead.' without the 'Arrays of..' parts, or..
the check whether the warning should be shown or not should also check if a.ndim > 1 or b.ndim > 1

Tbh, I'm a bit confused what that warning is implying as recommended usage anyway. Is the user supposed to add an all-zero third dimension for 2-d vectors and then calculate cross product in 3-d and then extract the valid part from it? That seems kinda odd-looking, I guess if that really is the case it looks like this prompts to look at #13718 again potentially?

So instead of..

x = [[1, 2], [1, 2]]
y = [[4, 5], [4, 5]]
result = np.cross(x, y)

user should do something like..?

x = [[1, 2, 0], [1, 2, 0]]
y = [[4, 5, 0], [4, 5, 0]]
result = np.cross(x, y)[:, -1]

Reproduce the code example:

import numpy as np

x = [1, 2]
y = [4, 5]
# the next line shows a warning, even though it's even from the docstring examples section
np.cross(x, y)

Error message:

DeprecationWarning: Arrays of 2-dimensional vectors are deprecated. Use arrays of 3-dimensional vectors instead. (deprecated in NumPy 2.0)

Python and NumPy Versions:

2.0.0rc2
3.10.14 | packaged by conda-forge | (main, Mar 20 2024, 12:45:18) [GCC 12.3.0]

Runtime Environment:

[{'numpy_version': '2.0.0rc2',
  'python': '3.10.14 | packaged by conda-forge | (main, Mar 20 2024, 12:45:18) '
            '[GCC 12.3.0]',
  'uname': uname_result(system='Linux', node='mother', release='5.10.0-25-amd64', version='#1 SMP Debian 5.10.191-1 (2023-08-16)', machine='x86_64')},
 {'simd_extensions': {'baseline': ['SSE', 'SSE2', 'SSE3'],
                      'found': ['SSSE3',
                                'SSE41',
                                'POPCNT',
                                'SSE42',
                                'AVX',
                                'F16C',
                                'FMA3',
                                'AVX2',
                                'AVX512F',
                                'AVX512CD',
                                'AVX512_SKX',
                                'AVX512_CLX',
                                'AVX512_CNL',
                                'AVX512_ICL'],
                      'not_found': ['AVX512_KNL', 'AVX512_KNM', 'AVX512_SPR']}},
 {'architecture': 'SkylakeX',
  'filepath': '/home/megies/miniconda3/envs/np2/lib/libopenblasp-r0.3.27.so',
  'internal_api': 'openblas',
  'num_threads': 8,
  'prefix': 'libopenblas',
  'threading_layer': 'pthreads',
  'user_api': 'blas',
  'version': '0.3.27'}]

Context for the issue:

In our testsuite, we also have a test runner with pytest -W error to make sure our code is not causing deprecation warnings in order to keep our code base work with upcoming releases of dependencies. And that's where this deprecation warning came up.

The text was updated successfully, but these errors were encountered:

seberg · 2024-06-05T13:59:02Z

Ref #24818, does anyone know what the argument/reasoning for this was?

Avoid variable number of result dimensions?
Deprecated cross([0, 1], [1, 2, 3]) from working?

seberg · 2024-06-05T18:45:03Z

OK, since I missed it there. The issue is the first thing and the reasoning is in #13718 which suggested a cross2d, though.

seberg · 2024-06-05T18:56:13Z

We had a bit of a discussing, we should definitely tell people to use this (I think it is right):

cross2d = arr1[..., 0] * arr2[..., 1] - arr1[..., 1] * arr2[..., 0]

And maybe even document it in the text. The original issue had cross2d, and if a lot of people use the 2d version, then it might be nice to have it as np.linalg.cross2d. If that was clear now, I might prefer to not do the deprecation as quickly, because you cannot use cross2d until it exists (and is established).

This PR stems from the community meeting on Jun 5, 2024. This implements a new function cross2d that accepts (a stack of) 2-element vectors, in the same manner that cross currently does. Both np.cross2d and np.linalg.cross2d are added. The np.linalg.cross2d is API compatible. This PR closes numpy#13718 and closes numpy#26620. Units tests for cross2d are included. A new test_ValueError is added for cross. Updated doc links for both funtions. Added examples for cross2d. Cleaned up whitespace on both functions.

This PR stems from the community meeting on Jun 5, 2024. This implements a new function cross2d that accepts (a stack of) 2-element vectors, in the same manner that cross currently does. Both np.cross2d and np.linalg.cross2d are added. The function np.linalg.cross2d is Array API compatible. This PR closes numpy#13718 and closes numpy#26620. Units tests for cross2d are included. A new test_ValueError is added for cross. Updated doc links for both funtions and linked to .rst. Added examples for cross2d. Cleaned up whitespace on both functions. Adjusted examples in cross to show how to avoid deprecation warning. Adds an example to compare cross2d with (floating point) det. Includes a release note.

rgommers · 2024-06-12T14:43:48Z

The original issue had cross2d, and if a lot of people use the 2d version, then it might be nice to have it as np.linalg.cross2d. If that was clear now, I might prefer to not do the deprecation as quickly, because you cannot use cross2d until it exists (and is established).

It looks like linalg.cross2d is moving ahead, although the need is still slightly on the thin side. I'm -1 on changing our mind on the deprecation at the very last moment, since that has downsides and since clearly this isn't used a lot - it took ~6 months after gh-25145 got merged and almost 3 months after 2.0.0b1 was out for someone to notice.

Let's verify the one-liner above as a workaround (it seems right), and add that to the docstring as a compat note, and backport it to the 2.0.x branch. That should be easy and safe to do in the next few days.

This addresses a part of numpygh-26620. The one-liner was verified and is likely to be used in a future `np.linalg.cross2d` function implementation, in numpygh-26640. [skip actions] [skip azp] [skip cirrus]

rgommers · 2024-06-14T12:46:17Z

Once gh-26692 is backported to the 2.0.x branch, the milestone on this can be removed.

This addresses a part of numpygh-26620. The one-liner was verified and is likely to be used in a future `np.linalg.cross2d` function implementation, in numpygh-26640. [skip actions] [skip azp] [skip cirrus]

Daniel63656 · 2024-07-13T08:33:05Z

Having the same issue. Why wouldn't cross() be abstract enough to handle 2d and 3d without the user having to differentiate

leycec · 2024-09-13T06:25:58Z

lol. So, numpy.linalg.cross2d() never happened. Now numpy.linalg.cross() emits unhelpful deprecation warnings for the previously valid two-dimensional use case that fail to actually inform the end user of how to resolve the deprecation. If a use case requires two dimensions, telling end users to "just use three dimensions instead" isn't really a useful recommendation. In short, this is basically useless:

DeprecationWarning: Arrays of 2-dimensional vectors are deprecated.
Use arrays of 3-dimensional vectors instead. (deprecated in NumPy 2.0)

That's a hard UX fail. Simply hand-waving that away as "fine for certain definitions of fine" because users failed to test against NumPy pre-releases like 2.0.0b1 isn't a reasonable justification, either. Of course users failed to test against NumPy pre-releases. They're pre-releases! Without compelling reasons to install third-party pre-releases, users (...waitforit) do not install third-party pre-releases. Why would they? It's needless work for no real-world gain and significant pain.

tl;dr

In short, the existing DeprecationWarning:

(A) Is less than useless. 2D code doesn't magically become 3D code just because you tell it to become 3D code. Yes, that's exactly what the existing DeprecationWarning is telling everybody. </facepalm>
(B) Should be entirely rewritten to reflect the trivial one-liner identified above by @seberg. Something resembling the following would actually be useful: e.g.,

DeprecationWarning: np.cross() no longer supports 2-dimensional vectors.
Replace all np.cross() calls over 2-dimensional vectors with one-liners:
    cross2d = arr1[..., 0] * arr2[..., 1] - arr1[..., 1] * arr2[..., 0]

I shake my head, fam. I shake it hard. 😮‍💨

* Address numpy 2.0 warning: numpy/numpy#26620 * Address issue with codecov uploading: codecov/codecov-action#1274

This addresses a part of numpygh-26620. The one-liner was verified and is likely to be used in a future `np.linalg.cross2d` function implementation, in numpygh-26640. [skip actions] [skip azp] [skip cirrus]

burnpanck · 2025-01-22T08:53:15Z

Pity that np.linalg.cross2d never happened. While the oneliner is trivial to write when the inputs are arrays, if the inputs are expressions, the one-liner becomes a three-liner! Its of course doable, but having a true one-liner for the "2d cross product" around is a great nice-to-have for any 2D geometry task.

cslotboom · 2025-02-05T00:17:45Z

This has also affected me. Converting all my 2D vectors to 3D vectors with an empty row just to cross multiply feels like a pretty clunky work around, I'd strongly prefer not to do that if it's in the cards.

megies added the 00 - Bug label Jun 5, 2024

seberg added this to the 2.0.0 release milestone Jun 5, 2024

seberg added the 04 - Documentation label Jun 5, 2024

ngoldbaum added the Numpy 2.0 API Changes label Jun 5, 2024

bmwoodruff mentioned this issue Jun 7, 2024

ENH: Adds cross2d to linalg (with Array API compatibility) #26640

Closed

This was referenced Jun 10, 2024

Add NumPy 2.0 support vispy/vispy#2599

Merged

More numpy 2.0 related fixes napari/napari#6974

Merged

mattip assigned rgommers Jun 12, 2024

rgommers mentioned this issue Jun 14, 2024

DOC: document workaround for deprecation of dim-2 inputs to cross #26692

Merged

charris mentioned this issue Jun 14, 2024

DOC: document workaround for deprecation of dim-2 inputs to cross #26694

Merged

charris removed this from the 2.0.0 release milestone Jun 15, 2024

RobPasMue mentioned this issue Sep 10, 2024

fix: handle properly np.cross() - 2d ops deprecated in Numpy 2.X ansys/pyansys-geometry#1419

Merged

9 tasks

stefmolin added a commit to stefmolin/data-morph that referenced this issue Oct 5, 2024

Address numpy 2.0 warning and fix codecov v4 setup (#227)

e4424fd

* Address numpy 2.0 warning: numpy/numpy#26620 * Address issue with codecov uploading: codecov/codecov-action#1274

schuenke mentioned this issue Oct 18, 2024

Enable numpy version > 2.0 PTB-MR/mrpro#444

Closed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Uh oh!

BUG: np.cross() warns about arrays of vectors when used with two simple 2-d vectors #26620

BUG: np.cross() warns about arrays of vectors when used with two simple 2-d vectors #26620

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

BUG: np.cross() warns about arrays of vectors when used with two simple 2-d vectors #26620

BUG: np.cross() warns about arrays of vectors when used with two simple 2-d vectors #26620

Comments

Uh oh!

Describe the issue:

Reproduce the code example:

Error message:

Python and NumPy Versions:

Runtime Environment:

Context for the issue:

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

tl;dr

Uh oh!

Uh oh!

Uh oh!