heat
heat copied to clipboard
helmholtz-analytics
Drop support for old versions of dependencies, add support for newer ones
According to the drop schedule in https://numpy.org/neps/nep-0029-deprecation_policy.html we should have dropped
- NumPy 1.22 on January 01, 2024
- Python 3.8 on April 14, 2023 moreover, we should drop
- Python 3.9 on April 05, 2024 (realistic goal for next release?)
Regarding PyTorch, the last releases were according to https://github.com/pytorch/pytorch/releases?page=1: 2.2 (last week), 2.1 (Oct 04, 2023), 2.0 (Mar 15, 2023), 1.13 (Oct 28, 2022), 1.12 (Jun 28, 2022), 1.11 (Mar 10, 2022), 1.10 (Oct 21, 2021), 1.9 (Jun 15, 2021), ...
This means that according to NEP29 we should drop at least PyTorch 1.10 or (and if we want to release in April) even 1.11.
Summary: I'd suggest to drop NumPy <= 1.22, Python <= 3.9, and PyTorch <= 1.11 for the next release (if we target April).
Also, it would be a good idea to go through the code and look for try/excepts's that have been introduced to ensure backward compatibility with versions we dont support any longer
@mtar @ClaudiaComito @JuanPedroGHM @Markus-Goetz @krajsek do you agree with aiming for a next release in April and thus dropping support of previous versions as suggested above?
![github-actions[bot] avatar](https://avatars.githubusercontent.com/in/15368?v=4 "Published by a pub.dev verified publisher"){kind=small}
Actually, we should go through the whole list of dependencies in setup.py and the ci.yaml files.
| Dependency | currently | suggested for next release |
|---|---|---|
| Python | >= 3.8 | >= 3.10 |
| mpi4py | >= 3.0.0 | |
| numpy | >= 1.20.0 | >= 1.23 |
| torch | >= 1.8.0 < 2.1.3 | >= 1.12 |
| scipy | >= 0.14.0 | |
| pillow | >= 6.0.0 | |
| torchvision | >= 0.8.0 | |
| docutils | >= 0.16 | |
| h5py | >= 2.8.0 | |
| netCD4 | >= 1.5.6 | |
| pre-commit | >= 1.18.3 | |
| scikit-learn | >= 0.24.0 | |
| matplotlib | >= 3.1.0 | |
| perun | >= 0.2.0 |