(feature): Approximate vorticity fields using velocity data. Include …

[greinerth]

…github ci/cd stuff and code quality ensuring stuff

Features:

• included modules for computing vorticity given velocity fields (vorticity.py, velocity2vorticity.py). Vorticity fields are currently missing in the 3dcfd datasets.
• included noxfile.py for automatic testing, building of API doc, and code-linting.

Changes:

• updated pyproject.toml (hatchling build) and added additional optional dependencies (dev, test)
• bash file cleanup
• cleanup of python code that can be done without domain-specific knowledge

CICD:

• included .pre-commit-config.yaml to ensure code quality in the pipeline and locally
• included github workflows (precommit.yml, python-package.yml, python-publish.yml)
• included "tests" folder with test_vorticity.py for future unit testing with pytest.

Help required:

• The employed linters (ruff, pylint) complain a great deal: Unused variables, missing typedefs for input/output parameters etc. The types are mostly unknown and are hard to infer. Thus the pipeline definitely fails.
• Packages: Some of the packages got updates (e.g. jax). This needs to be tested before everything can be updated.
• Extend python support (Python3.11 Python3.12)

[kmario23]

Hey @greinerth, appreciate the PR! Can you please have a look at the recent commits and resolve conflicts?

[greinerth]

Hey @greinerth, appreciate the PR! Can you please have a look at the recent commits and resolve conflicts?

I'll have a look and resolve the conflicts within the next days.

Cheers,
Gerhard

[greinerth]

Done ;)

[kmario23]

@greinerth I've had time to look into it. There are issues with relative imports of vorticity.compute_spectral_vorticity_jnp in data_gen/velocity2vorticity.py. vorticity.py should be in the same tree level, which is now in src.

As it exists now, it's postprocessing. So, perhaps we can move these modules to a directory, say postprocess in data_gen. And, also add some info about how to run this script. E.g.,

export PYTHONPATH="/path/to/PDEBench"
CUDA_VISIBLE_DEVICES="0" python -m data_gen.velocity2vorticity --data /pdebench_data/3D/Train/3D_CFD_Turb_M1.0_Eta1e-08_Zeta1e-08_periodic_Train.hdf5

Later, we could try integrating this as part of the JAX solver itself and serialize both fields.

[greinerth]

Hi @kmario23, good catch! The relative import issue is now fixed.
The conversion is now added as an executable script (c.f. pyproject.toml)
After installing pdebench via pip it can now be executed as follows:

velocity2vorticity --data /pdebench_data/3D/Train/3D_CFD_Turb_M1.0_Eta1e-08_Zeta1e-08_periodic_Train.hdf5

Let me know if additional adaptions are required.

Cheers,
Gerhard

[kmario23]

@greinerth LGTM.

[kmario23]

Hey @greinerth, I had a few hours today, and I tried merging the PR and fixing some Ruff linter errors, but there's still a lot to fix. We especially need to add type annotations. I'll ask the authors about this.

It seems the noxfile.py is missing. Can you add that in the meantime, please?

[greinerth]

I am currently out of office. I'll add it asap (next thursday).

Cheers,
Gerhard

[kmario23]

Perhaps we should consider using a logger (e.g., python's logging) since Ruff complains about the print() statements.