Fix shuffle for distributed sampler

[rohitgr7]

What does this PR do?

Fixes #2703

Before submitting

• Was this discussed/approved via a Github issue? (no need for typos and docs improvements)
• 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? Otherwise, we ask you to create a separate PR for every change.
• Did you make sure to update the documentation with your changes?
• Did you write any new necessary tests?
• Did you verify new and existing tests pass locally with your changes?
• If you made a notable change (that affects users), did you update the CHANGELOG?

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 🙃

[awaelchli]

could you also add a test for this? maybe mark it as wip in the meantime

[codecov]

Codecov Report

Merging #2789 into master will increase coverage by 0%.
The diff coverage is 100%.

@@          Coverage Diff           @@
##           master   #2789   +/-   ##
======================================
  Coverage      91%     91%           
======================================
  Files          76      76           
  Lines        6787    6788    +1     
======================================
+ Hits         6150    6151    +1     
  Misses        637     637           

[rohitgr7]

@awaelchli need a little help here with the test. Not sure why is it giving this error.

train_sampler = trainer.train_dataloader.sampler
AttributeError: 'NoneType' object has no attribute 'sampler'

Can't test it locally.

[awaelchli]

Yes, the train_dataloader is None because the test that runs in the main process never sees the trainer state getting updated. When using ddp, the fit and test is running in subprocesses. You can create a callback and put your assertions in e.g. on_train_start and on_test_start methods. Also, remember to skip ddp related tests on the windows platform.

[rohitgr7]

Oh! okay. Thanks :)

[awaelchli]

@rohitgr7 I drafted it here in the latest commit. Need to wait for CI since I also can't test distributed things locally.

[rohitgr7]

Nice! It passed.

[williamFalcon]

What happens in the case where the user does not want to shuffle train data?

[rohitgr7]

Add this sampler manually and pass replace_ddp_sampler=False. Maybe we should mention this thing in docs.

[williamFalcon]

yes. Mind updating that?