[feat] 1/2 Add trainer.predict

[tchaton]

What does this PR do?

This PR:

By calling Trainer.predict(model, dataloaders), it will call forward function and gather predictions.

Trainer

• predict

LightningModule

• predict -> calls forward

Accelerators

• predict

Uniformize use of RunningState across LoggerConnector, DDPWrapper, DP.

Add tests for DDP, DP, DDP_SPAWN, 1 GPU, DDP_CPU, DDP_SHARDED.

Fixes # (issue) <- this links related issue to this PR

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
• Check that target branch and milestone match!

Did you have fun?

Make sure you had fun coding 🙃

[pep8speaks]

Hello @tchaton! Thanks for updating this PR.

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

Comment last updated at 2021-01-27 15:21:22 UTC

[SeanNaren]

Logic wise everything looks clear to me! Just a few nits that are not super critical :)

As mentioned, it would be nice to display predict with just dataloader rather than test_dataloader to differentiate that this works on any dataloader (I can pass my train dataloader for example)

[Borda]

could somehow make the logic that test_step is using predcit_step?

[s-rog]

Does predict() deal with ddp data padding from distributed samplers?
If not, I suppose we'll need a disclaimer that we're waiting on a pytorch fix

[Borda]

LGTM, just check the new _predition attribute