Bugfix: Lr finder and hparams compatibility

[SkafteNicki]

What does this PR do?

Fixes #1983
Solves compatibility problem between the (new) hparams interface and the learning rate finder

[codecov]

Codecov Report

Merging #2821 into master will not change coverage.
The diff coverage is n/a.

@@          Coverage Diff           @@
##           master   #2821   +/-   ##
======================================
  Coverage      89%     89%           
======================================
  Files          79      79           
  Lines        7302    7302           
======================================
  Hits         6514    6514           
  Misses        788     788           

[awaelchli]

If setting the attribute failed before, it means that we have no tests for the LRfinder that involves setting the learning rate. I expected such a test case with this fix. What do you think?

[SkafteNicki]

@awaelchli I am pretty sure that there should be test for this, but I will definitely check up on it and add if missing (fix if broken)

[pep8speaks]

Hello @SkafteNicki! Thanks for updating this PR.

There are currently no PEP 8 issues detected in this Pull Request. Cheers! 🍻

Comment last updated at 2020-08-06 11:31:14 UTC

[SkafteNicki]

@awaelchli you were right, the test did not check this since the base testing model model=EvalModelTemplate() both have an model.learning_rate attribute and a model.hparams.learning_rate attribute with the latter never being checked. I have changed this now.

[awaelchli]

solid fix! Thanks @SkafteNicki
I imagine a similar problem exists in the batch size finder, so your helper functions added here could be useful there.

[williamFalcon]

looks like GPU tests are failing?
@awaelchli, @SkafteNicki

[awaelchli]

@williamFalcon It looks like these failing tests come from master

[Borda]

if has been rebased (master is fixed) but it is still failing, @SkafteNicki mind check it?

@nateraw any idea why this fails?

FAILED tests/core/test_datamodules.py::test_full_loop_ddp_spawn

[mergify]

Great job! =)