Refactor: Imports

[SkafteNicki]

What does this PR do?

Fixes #1654

The following is kept in root:

• All regression + classification metrics
• Metric and MetricCollection
• All simple aggregation metrics: SumMetric, MeanMetric etc.

Rest should be imported from domain.

Before submitting

• Was this discussed/agreed via a Github issue? (no need for typos and docs improvements)
• Did you read the contributor guideline, Pull Request section?
• Did you make sure to update the docs?
• Did you write any new necessary tests?

PR review

Anyone in the community is free to review the PR once the tests have passed.
If we didn't discuss your PR in Github issues there's a high chance it will not be merged.

Did you have fun?

Make sure you had fun coding 🙃

[codecov]

Codecov Report

Merging #1681 (f78c894) into master (9d3a7cc) will decrease coverage by 28%.
The diff coverage is n/a.

❗ Current head f78c894 differs from pull request most recent head 5b85077. Consider uploading reports for the commit 5b85077 to get more accurate results

Additional details and impacted files

@@           Coverage Diff            @@
##           master   #1681     +/-   ##
========================================
- Coverage      89%     61%    -28%     
========================================
  Files         228     228             
  Lines       12796   12796             
========================================
- Hits        11363    7796   -3567     
- Misses       1433    5000   +3567     

[lantiga]

TorchMetrics has extremely widespread use. I know we are heading towards 1.0 etc, but I would be very conservative here.

I'm in favor of pushing users to import from sub-packages, but I don't think we should wholesale remove root imports just yet. This has a 100% chance of breaking a lot of people's code, and these people don't necessarily care about 0.11 vs 1.0.

I would keep root imports in place and just produce deprecation warnings until 2.0.

[Borda]

I would keep root imports in place and just produce deprecation warnings until 2.0

ok, this would effectively mean we would create a wrapper class for each of them so also confusing for developers that they won't import Accuracy but AccuracyDeprecatedImport

source torchmetrics/image/tv.py

class TotalVariation(Metric):
  ... # the actual implementation

class TotalVariationDeprecatedImport(TotalVariation):

	def __init__(...):
		raise DeprecationWarning(...)
		super().__init__(...)

and for imports: torchmetrics/image/__init__.py

from torchmetrics.image.tv import TotalVariation

and torchmetrics/__init__.py

from torchmetrics.image.tv import TotalVariationDeprecatedImport as TotalVariation

[Borda]

lets keep it as updating docstrings