implementing changes for exposing Lightning Tuner's LR finder

[solalatus]

Fixes #1503.

Summary

As proposed in the linked issue here, this PR implements a lightweight setting for the fit() methods of the TorchForecastingModels to just do a "setup" run, and return gracefully without being interrupted and without running the main training loop. This enabled the usage of Lightning Tuner to find optimal learning rates. For this a new method is exposed, that wraps Lightning's method and provides sane defaults.

Other Information

I have not done unittests for this functionality, since I am basically exposing some already present functionality. None the less I included a Notebook in the examples here that demonstrates the usage of the feature.

Updates (from @dennisbader):

• decouple fit_from_dataset and _train setup from fitting the model. like this we can setup the model without having to fit the model, or having to add a parameter setup_only
• adds unit tests

[review-notebook-app]

Check out this pull request on 

See visual diffs & provide feedback on Jupyter Notebooks.

---

Powered by ReviewNB

[dennisbader]

Hi @solalatus and thanks for this PR! Looks good already :)

I made a suggestion for adding a new private method that handles setting up the model for training, to avoid adapting the fit API and users having to call fit() with setup_only=True before lr_find.

[codecov-commenter]

Codecov Report

Attention: Patch coverage is 97.05882% with 1 line in your changes missing coverage. Please review.

Project coverage is 94.07%. Comparing base (de94ef4) to head (e7ff956).
Report is 289 commits behind head on master.

Files with missing lines	Patch %	Lines
...arts/models/forecasting/torch_forecasting_model.py	97.05%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #1609      +/-   ##
==========================================
- Coverage   94.14%   94.07%   -0.08%     
==========================================
  Files         125      125              
  Lines       11364    11370       +6     
==========================================
- Hits        10699    10696       -3     
- Misses        665      674       +9     

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

---

🚨 Try these New Features:

• Flaky Tests Detection - Detect and resolve failed and flaky tests

[dennisbader]

Hi @solalatus, just wanted to check in how it's going with this PR.
We plan to release a new version in the next 2 weeks, and would like to include this one :) (no rush though!)

[solalatus]

@dennisbader thanks for checking. Sadly I am completely overwhelmed, so will not be able to make progress on this. Can anyone please take over and push it through the "finish line"?

[dennisbader]

Yes, I can take over from here. Thanks for the work so far!

[madtoinou]

LGTM, I find this approach quite elegant!

[dennisbader]

Looks good now :) Thanks @solalatus for the contribution!

[solalatus]

@dennisbader thanks for pushing it through! Cheers!