add os.Path support for load/save torch models

[turbotimon]

Checklist before merging this PR:

• Mentioned all issues that this PR fixes or addresses.
• Summarized the updates of this PR under Summary.
• Added an entry under Unreleased in the Changelog.

Summary

Currently, the save/load methdos for TorchForecastingModel do not support os.PathLike object for the path argument. Other model-families (conformal, ensemble, forecasting) already do so. This PR changes that, improving the usability as well as make the interface more similar.

No additional test are needed. As the path argument is transformed to a pathlib.Path object at the beginning it is guaranteed to work with str as well as any os.PathLike objects

The other Model-family also support file-like objects (IObuffer). But since pl.Trainer.save_checkpoint does not supports it, i didn't add this support as well. Torch.save would support it, however.

[codecov]

Codecov Report

❌ Patch coverage is 66.66667% with 1 line in your changes missing coverage. Please review.
✅ Project coverage is 95.48%. Comparing base (2c666f8) to head (8487fba).
⚠️ Report is 60 commits behind head on master.

Files with missing lines	Patch %	Lines
...arts/models/forecasting/torch_forecasting_model.py	66.66%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #2947      +/-   ##
==========================================
- Coverage   95.52%   95.48%   -0.05%     
==========================================
  Files         146      150       +4     
  Lines       15703    16201     +498     
==========================================
+ Hits        15000    15469     +469     
- Misses        703      732      +29     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[dennisbader]

Thanks a lot for this PR @turbotimon 🚀

Adding support for Path-like objects sounds great. In this case there are some more places where we should allow this, to provide a unified usage over all methods.

[dennisbader]

Would be great to add this to all other save / load methods:

• model construction param: work_dir
• to_onnx()
• load_weights()
• work_dir in load_from_checkpoint()
• work_dir in load_weights_from_checkpoint()

[turbotimon]

yes that makes sense. Will try to do so as soon as i find time

[daidahao]

@turbotimon

Would you like to continue the work on this? I think both supporting os.PathLike and file-like object in save()/load() would be useful for use cases, like pathlib or in-memory operations.

While TorchForecastingModel stores two files (pt and pt.ckpt), we could still support file-like objects by bundling both in a compressed format. I am thinking using tarfile or zipfile as solutions. What do you think?

Thanks to @dennisbader for leading me here.

[turbotimon]

@daidahao thanks for asking and sorry for the late reply

Would you like to continue the work on this?

Actually, yes, but I'm pretty busy right now… So I don't mind if someone else takes care of it; otherwise, I'll take another look at it as soon as I have some free time.

While TorchForecastingModel stores two files (pt and pt.ckpt), we could still support file-like objects by bundling both in a compressed format. I am thinking using tarfile or zipfile as solutions. What do you think?

I do not think that is related to "file-like object", it's just a representative of a path, not a real file. And .ckpt gets added by convention. Bundling can, however, be advantageous for other reasons, but I don't have a strong opinion on that.