Loop Refactor 3/N - Evaluation Loop

[awaelchli]

What does this PR do?

Introduces the evaluation loop under the new interface introduced in #7871

Three new classes:

• DataLoaderLoop: This loop runs over a list of dataloaders
• EvaluationDataLoaderLoop: A subclass of DataLoaderLoop running over a list of evaluation dataloaders (can be test or validation)
• EvaluationEpochLoop: Runs a single epoch of validation or test.

In the next PR, we will introduce also the PredictionDataLoaderLoop and PredictionEpochLoop.

Before submitting

• Was this discussed/approved via a GitHub issue? (not for typos and docs)
• Did you read the contributor guideline, Pull Request section?
• Did you make sure your PR does only one thing, instead of bundling different changes together?
• Did you make sure to update the documentation with your changes? (if necessary)
• Did you write any new necessary tests? (not for typos and docs)
• Did you verify new and existing tests pass locally with your changes?
• Did you update the CHANGELOG? (not for typos, docs, test updates, or internal minor changes/refactorings)

PR review

Anyone in the community is free to review the PR once the tests have passed.
Before you start reviewing make sure you have read Review guidelines. In short, see the following bullet-list:

• Is this pull request ready for review? (if not, please submit in draft mode)
• Check that all items from Before submitting are resolved
• Make sure the title is self-explanatory and the description concisely explains the PR
• Add labels and milestones (and optionally projects) to the PR so it can be classified

Did you have fun?

Make sure you had fun coding 🙃

[pep8speaks]

Hello @awaelchli! Thanks for updating this PR.

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

Comment last updated at 2021-06-18 11:00:30 UTC

[codecov]

Codecov Report

Merging #7990 (6545bc9) into master (3fece17) will decrease coverage by 0%.
The diff coverage is 96%.

@@          Coverage Diff           @@
##           master   #7990   +/-   ##
======================================
- Coverage      92%     91%   -0%     
======================================
  Files         207     210    +3     
  Lines       13479   13558   +79     
======================================
+ Hits        12346   12392   +46     
- Misses       1133    1166   +33     

[ethanwharris]

LGTM

[tchaton]

Looks awesome !

[Borda]

Suggested change

	if self.trainer.testing:
	self.trainer.lightning_module._current_fx_name = "test_step"
	with self.trainer.profiler.profile("test_step"):
	output = self.trainer.accelerator.test_step(step_kwargs)
	else:
	self.trainer.lightning_module._current_fx_name = "validation_step"
	with self.trainer.profiler.profile("validation_step"):
	output = self.trainer.accelerator.validation_step(step_kwargs)
	name_step = "test_step" if self.trainer.testing else "validation_step"
	self.trainer.lightning_module._current_fx_name =name_step
	with self.trainer.profiler.profile(name_step):
	if self.trainer.testing:
	output = self.trainer.accelerator.test_step(step_kwargs)
	else:
	output = self.trainer.accelerator.validation_step(step_kwargs)

[Borda]

Suggested change

	if output is not None:
	if isinstance(output, ResultCollection):
	output = output.detach()
	if self.trainer.move_metrics_to_cpu:
	output = output.cpu()
	elif isinstance(output, dict):
	output = recursive_detach(output, to_cpu=self.trainer.move_metrics_to_cpu)
	elif isinstance(output, Tensor) and output.is_cuda and self.trainer.move_metrics_to_cpu:
	output = output.cpu()
	outputs.append(output)
	return outputs
	if output is None:
	return outputs
	if isinstance(output, ResultCollection):
	output = output.detach()
	if self.trainer.move_metrics_to_cpu:
	output = output.cpu()
	elif isinstance(output, dict):
	output = recursive_detach(output, to_cpu=self.trainer.move_metrics_to_cpu)
	elif isinstance(output, Tensor) and output.is_cuda and self.trainer.move_metrics_to_cpu:
	output = output.cpu()
	outputs.append(output)
	return outputs