Make unit tests compatible with NumPy 2.x

[Marc-Jindra]

Due Diligence

• General:
  • title of the PR is suitable to appear in the Release Notes
• Implementation:
  • unit tests: all split configurations tested
  • unit tests: multiple dtypes tested
  • benchmarks: created for new functionality
  • benchmarks: performance improved or maintained
  • documentation updated where needed

Description

Fixed DeprecationWarnings from numpy 2.x

• row_stack to vstack
• passing axes to np.fft.fftn when s is not None
• __array__ implementation doesn't accept a copy keyword (probably related to Tensor.__array__)
• __array_wrap__ must accept context and return_scalar arguments
• Arrays of 2-dimensional vectors are deprecated (np.cross())

Issue/s resolved: #1552

Changes proposed:

• unittests use vstack instead of row_stack if the numpy version is >= 2.0.0
• unittests now pass axes to np.fft.fftn when s is not None
• changed setup.py to allow numpy versions >= 2.0.0

Type of change

Memory requirements

Performance

Does this change modify the behaviour of other functions? If so, which?

yes / no

[Marc-Jindra]

Numpy 2 unittest Warnings

row_stack (solved)

DeprecationWarning: `row_stack` alias is deprecated. Use `np.vstack` directly.

self explanatory.

__array__ (I think it’s related to torch)

DeprecationWarning: __array__ implementation doesn't accept a copy keyword, so passing copy=False failed. __array__ must implement 'dtype' and 'copy' keyword arguments. To learn more, see the migration guide https://numpy.org/devdocs/numpy_2_0_migration_guide.html#adapting-to-changes-in-the-copy-keyword

Here is the link

Triggered in: test_save (heat.core.tests.test_io.TestIO.test_save)

Or: /heat/heat/core/io.py:441/439 by the following code: var[merged_slices] = data.larray.cpu()

Apparently, it's related to PyTorch, as the Tensor.__array__ method does not accept a copy keyword: def __array__(self, dtype=None).

__array_wrap__

DeprecationWarning: __array_wrap__ must accept context and return_scalar arguments (positionally) in the future. (Deprecated in NumPy 2.0)

Triggered in:

• test_fit_iris_unsplit (heat.cluster.tests.test_kmedians.TestKMedians.test_fit_iris_unsplit)
• test_exp2 (heat.core.tests.test_exponential.TestExponential.test_exp2)
• test_sort (heat.core.tests.test_manipulations.TestManipulations.test_sort)
• test_randn (heat.core.tests.test_random.TestRandom_Batchparallel.test_randn)
• test_RobustScaler (heat.preprocessing.tests.test_preprocessing.TestRobustScaler.test_RobustScaler)

Or:

• numpy/_core/fromnumeric.py:48
• heat/heat/core/tests/test_exponential.py:99

test_exponential Line 99: tmp = np.exp2(torch.arange(elements, dtype=torch.float64))

I couldn't find anything in the NumPy documentation, but there should be a __array_wrap__ method in an ndarray subclass to which context=None and return_scalar=None must be added. I couldn’t find such a method in heat.

np.cross()

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

Calling np.cross with two-dimensional arrays is no longer supported. As a solution, NumPy suggests writing your own function:

def cross2d(x, y):
    return x[..., 0] * y[..., 1] - x[..., 1] * y[..., 0]

However, this is not the desired solution.

axes (solved)

DeprecationWarning: `axes` should not be `None` if `s` is not `None` (Deprecated in NumPy 2.0). In a future version of NumPy, this will raise an error and `s[i]` will correspond to the size along the transformed axis specified by `axes[i]`. To retain current behaviour, pass a sequence [0, ..., k-1] to `axes` for an array of dimension k.

Now, when s is passed to np.fft.fftn, axes must also be provided. To maintain the previous default behavior, axes must be passed for k elements in s, which correspond to the last k dimensions of the array (starting from 0).

Example for a 3-dimensional array: np.fft.fftn(x.numpy(), s=(6, 6), axes=(1, 2))

[ClaudiaComito]

Sample stack trace from Marc:

...........................s...................s.s.s..................................../home/marc/Heat/heat/heat/core/tests/test_exponential.py:99: DeprecationWarning: __array_wrap__ must accept context and return_scalar arguments (positionally) in the future. (Deprecated NumPy 2.0)
  tmp = np.exp2(torch.arange(elements, dtype=torch.float64))
......................../home/marc/Heat/envs/numpy2env/lib/python3.12/site-packages/torch/functional.py:539: UserWarning: torch.meshgrid: in an upcoming release, it will be required to pass the indexing argument. (Triggered internally at /pytorch/aten/src/ATen/native/TensorShape.cpp:3637.)
  return _VF.meshgrid(tensors, **kwargs)  # type: ignore[attr-defined]
............ss....../home/marc/Heat/heat/heat/core/io.py:441: DeprecationWarning: __array__ implementation doesn't accept a copy keyword, so passing copy=False failed. __array__ must implement 'dtype' and 'copy' keyword arguments. To learn more, see the migration guide https://numpy.org/devdocs/numpy_2_0_migration_guide.html#adapting-to-changes-in-the-copy-keyword
  var[file_slices] = data.larray.cpu()
/home/marc/Heat/heat/heat/core/io.py:439: DeprecationWarning: __array__ implementation doesn't accept a copy keyword, so passing copy=False failed. __array__ must implement 'dtype' and 'copy' keyword arguments. To learn more, see the migration guide https://numpy.org/devdocs/numpy_2_0_migration_guide.html#adapting-to-changes-in-the-copy-keyword
  var[merged_slices] = data.larray.cpu()
..........................................E./home/marc/Heat/heat/heat/core/tests/test_manipulations.py:2690: DeprecationWarning: `row_stack` alias is deprecated. Use `np.vstack` directly.
  np_rstack = np.row_stack((a, b))
/home/marc/Heat/heat/heat/core/tests/test_manipulations.py:2698: DeprecationWarning: `row_stack` alias is deprecated. Use `np.vstack` directly.
  np_rstack = np.row_stack((a, b, c))
/home/marc/Heat/heat/heat/core/tests/test_manipulations.py:2705: DeprecationWarning: `row_stack` alias is deprecated. Use `np.vstack` directly.
  np_rstack = np.row_stack((a, b, c))
/home/marc/Heat/heat/heat/core/tests/test_manipulations.py:2716: DeprecationWarning: `row_stack` alias is deprecated. Use `np.vstack` directly.
  np_rstack = np.row_stack((d, e))
../home/marc/Heat/envs/numpy2env/lib/python3.12/site-packages/numpy/_core/fromnumeric.py:48: DeprecationWarning: __array_wrap__ must accept context and return_scalar arguments (positionally) in the future. (Deprecated NumPy 2.0)
  return conv.wrap(result, to_scalar=False)