ProjectionOperator Bug Fix (Issue #1990)

[M-A-Demir]

Description

Update in response to issue #1990: ProjectionOperator device input is not case-sensitive. Added case sensitivity and unit tests for ProjectionOperator inputs. Updated environment .yml files and env creation script to include new unittest-parametrize library used in tests. Updated developer guide to include example of parametrized test usage. Added name to NOTICE.txt.

Issue: #1990

Example Usage

ProjectionOperator(ig, ag, device="GPU")
device name passed is now case insensitive

❤️ Thanks for your contribution!

[M-A-Demir]

PS: In CIL/scripts/create_local_env_for_cil_development.sh, I couldn't install the pip module (unittest-parametrize) directly with the 'conda create' command , so I have changed it to use the requirements_test.yml file if $test_deps != 0.

I tried an alternative, which is to have a separate list of pip dependencies, use conda create, then activate the conda environment and perform pip install pip_dependencies, but ran into an error saying to run conda init before I can do the installations.

in GitHub Actions, Tests using unittest-parametrize produce a TypeError: TestAstraProjectors.test_ProjectionOperator_2Ddata() missing 3 required positional arguments: 'device', 'no_error_raised', and 'err_type'

Thought this may be due to an ASTRA dependency, so added @unittest.skipUnless(has_astra and has_nvidia, "Requires ASTRA GPU") to test_ProjectionOperator_3Ddata, and @unittest.skipUnless(has_astra, "Requires ASTRA") to test_ProjectionOperator_2Ddata - Only the 3D test skips, 2D test does not skip and produces same TypeError.

No issues running tests in local env:

• python 3.10.15
• numpy 1.24.4
• unittest-parametrize 1.6.0

[M-A-Demir]

UPDATE: Fixed the TypeError issue on test - the version I pushed did not have the TestAstraProjectors class inherit from unittest_parametrize.ParametrizedTestCase, which is required to run the parametrized tests

[gfardell]

It's a really nice addition thanks @M-A-Demir!

In addition to the small comments, could you test this on windows. @hrobarts should be able to help you out.

[gfardell]

What happens if you pass "GPU" and don't have an nvidia gpu or the astra-gpu library?

I'm guessing it doesn't return very cleanly, but I wonder if it's something we could catch?

[M-A-Demir]

@gfardell The output if you pass "gpu"/"GPU" without the NVidia GPU is:
`Warning: Unable to get available GPU memory. Defaulting to 1GB.

Error: maxBlockDimension getDevice: CUDA error 35: CUDA driver version is insufficient for CUDA runtime version.

Warning: Error querying device

Floating point exception (core dumped)`

Which don't seem to be python-raised errors, so I'm not sure I can use that in the tests I have.
I can add some checks for NVidia GPU and astra-gpu in the ProjectionOperator class