Numpy warning cleanup

[j-c-c]

Noticing some new Numpy warnings:

tests/test_averager2d.py: 2388 warnings
  /Users/carmichael/Work/ASPIRE-python.unpin_confuse/src/aspire/classification/reddy_chatterji.py:42: DeprecationWarning: numpy.fix is deprecated. Use numpy.trunc instead, which is faster and follows the Array API standard.
    midpoints = np.array([np.fix(axis_size / 2) for axis_size in shape])

tests/test_CtfEstimate.py: 5 warnings
tests/test_commonline_matrix.py: 1 warning
tests/test_downsample.py: 3 warnings
tests/test_filters.py: 2 warnings
tests/test_coordinate_source.py: 79 warnings
tests/test_image.py: 1 warning
tests/test_class_src.py: 3 warnings
tests/test_load_images.py: 6093 warnings
tests/test_micrograph_simulation.py: 1 warning
tests/test_oriented_source.py: 15 warnings
tests/test_array_image_source.py: 1031 warnings
tests/test_rotation.py: 2 warnings
tests/test_simulation.py: 29 warnings
tests/test_starfile_stack.py: 45 warnings
tests/test_volume.py: 21 warnings
tests/test_relion_interop.py: 20 warnings
tests/test_relion_source.py: 31 warnings
tests/test_orient_sync_voting.py: 2 warnings
  /Users/carmichael/opt/miniconda3/envs/unpin_confuse/lib/python3.12/site-packages/numpy/_core/records.py:462: DeprecationWarning: Setting the dtype on a NumPy array has been deprecated in NumPy 2.5.
  Instead of changing the dtype on an array x, create a new array with x.view(new_dtype)
    ret = object.__setattr__(self, attr, val)

tests/test_batched_covar2d.py: 4 warnings
tests/test_bispectrum.py: 4 warnings
tests/test_class2D.py: 6 warnings
tests/test_class_src.py: 2 warnings
tests/test_covar2d.py: 7 warnings
tests/test_covar2d_denoiser.py: 3 warnings
  /Users/carmichael/opt/miniconda3/envs/unpin_confuse/lib/python3.12/site-packages/numpy/lib/_index_tricks_impl.py:923: ComplexWarning: Casting complex values to real discards the imaginary part
    a.flat[:end:step] = val

tests/test_batched_covar2d.py: 4 warnings
tests/test_bispectrum.py: 4 warnings
tests/test_class2D.py: 6 warnings
tests/test_class_src.py: 2 warnings
tests/test_covar2d.py: 7 warnings
tests/test_covar2d_denoiser.py: 3 warnings
  /Users/carmichael/Work/ASPIRE-python.unpin_confuse/src/aspire/operators/blk_diag_matrix.py:133: ComplexWarning: Casting complex values to real discards the imaginary part
    self.data[key] = value.astype(self.dtype, copy=False)

-- Docs: https://docs.pytest.org/en/stable/how-to/capture-warnings.html

[j-c-c]

The first set of warnings is an easy fix of replacing np.fix() with np.trunc() as recommended in the deprecation warning.

The second set of warnings is related to issue #1392, for which we are awaiting a resolution in mrcfile.

The final two sets of warnings are both related to the same np.linalg.eig(make_symmat(covar)) call in the src/aspire/covariance/covar2d.py method shrink_covar(), which returns complex valued eigenvalues/eigenvectors (with 0 complex part) and then packs them into real valued arrays later on (once within the method and once using the returned complex valued shrink_covar). This is a new return behavior of eig as of numpy 2.5, see here.

One solution would be to use np.linalg.eigh(make_symmat(covar)) since make_symmat(covar) enforces a symmetric matrix. This would return real-valued eigenvalues and avoid the warning.

[j-c-c]

Resolved by #1401

[garrettwrong]

Still halting on warnings in CI. See develop.

[j-c-c]

Ok, looks like cvxpy uses an empty dummy array to infer the shape of an output. The uninitialized values of the empty array can nondeterministically produce invalid values during the operation (see cvxpy/cvxpy#3235). A fix for this bug is currently in their master branch, but not yet in the current release.

We could temporarily filter the warning for this code block until the next cvxpy release. Or I could rewrite the SDP constraints to avoid the cvxpy.trace code path entirely.

[garrettwrong]

If it is fixed upstream, filtering that area of code should be fine for a while.