numpy 2

[casperdcl]

• bump numpy>=2 (fixes support numpy v2 #1915)
  • Fix NaN usage in show_geometry function #2246 (fixes show_geometry fails with numpy2 #2242)
• bump conda-forge::astra-toolbox -> astra-toolbox::astra-toolbox>=2
• bump tigre Update to support TIGRE v3 #2317
  • fix failing tests
• bump ccpi-regulariser>=25 Unittest fail with regularisation toolkit 25.0.0 #2288
  • Remove pinning to numpy versions CCPi-Regularisation-Toolkit#215
• bump ccpi::tomophantom -> httomo::tomophantom>=3.0.3 Tomophantom plugin #832, updates to TomoPhantom plugin to use v3.0 #2287
  • setup conda correctly dkazanc/TomoPhantom#130

unsure about

• CIL/Wrappers/Python/cil/optimisation/algorithms/FISTA.py

  Line 234 in 4b038a7

  return self.f(x) + self.g(x)

• cil.plugins.tigre.Atb

• CIL/Wrappers/Python/cil/optimisation/algorithms/PDHG.py

  Line 516 in 4b038a7

  d1 = f_eval_d + g_eval_d

• CIL/Wrappers/Python/cil/optimisation/algorithms/APGD.py

  Line 285 in 4b038a7

  return self.f(x) + self.g(x)

upstream problems

• test tomophantom with numpy>=2
• test tigre with numpy>=2
• test astra-toolbox CUDA with numpy>=2
• test_CIL_CVXPy -> cvxpy -> scipy:

test_cil_vs_cvxpy_totalvariation_anisotropic (test_CIL_CVXPy.Test_CIL_vs_CVXPy.test_cil_vs_cvxpy_totalvariation_anisotropic) ... 
...
$CONDA_PREFIX/lib/python3.12/site-packages/scipy/sparse/_dia.py:140: UserWarning: result dtype changed due to the removal of value-based promotion from NumPy. Changed from int64 to int32.
  nnz += min(M+k,N)
...
$CONDA_PREFIX/lib/python3.12/site-packages/cvxpy/utilities/coeff_extractor.py:296: UserWarning: result dtype changed due to the removal of value-based promotion from NumPy. Changed from int32 to int64.
  P.row += offset

[paskino]

I'm assuming that all the filters in this file return the same type as the input.

[casperdcl]

is this necessary?

[gfardell]

see below, and discord

[gfardell]

If run with numpy 2.4 we have a lot of TIGRE test failures due to numpy functionality that has been marked as deprecated, then removed in 2.4. This may be something we can address in our tigre wrappers, or it may need a contribution directly to Tigre to fix. I'll investigate more next week.

test_adjoint_weights (test_PluginsTigre_General.TestMechanics_tigre.test_adjoint_weights)
cil/plugins/tigre/ProjectionOperator.py:216:
DeprecationWarning: Conversion of an array with ndim > 0 to a scalar is deprecated, 
and will error in future. Ensure you extract a single element from your array before 
performing this operation. (Deprecated NumPy 1.25.)

For now I've set the numpy version to <2.4 in the developer build environment, but we should discuss this futher before releasing.

The error it causes with 2.4 is:

  File "C:\Users\...\CIL\Wrappers\Python\cil\plugins\tigre\ProjectionOperator.py", line 178, in __call_Ax
    return Ax(data, self.tigre_geom, self.tigre_geom.angles,
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "Python/tigre/utilities/cuda_interface/_Ax.pyx", line 41, in _Ax._Ax_ext
  File "Python/tigre/utilities/cuda_interface/_types.pxd", line 158, in tigre.utilities.cuda_interface._types.convert_to_c_geometry
TypeError: only 0-dimensional arrays can be converted to Python scalars

FYI @AnderBiguri

[gfardell]

Full test matrix doesn't caputre this as cuda tests are only run with one set of dependencies.
https://github.com/TomographicImaging/CIL/actions/runs/26634706380

[AnderBiguri]

@gfardell yup, a lot of numpy issues with the new casting. I think I have tackled quite a lot of them in master already, but have not done a decent unit-test run... I would appreciate help if you find issues, I think in 99.99% of the cases its just changing a line for casting.

[lauramurgatroyd]

I tested the CIL-Demos on this branch with numpy v2.3.5 and the other requirements listed on this PR (updated tigre etc)
and they all worked!
The only changes in print statements I noticed were: