Deprecate {train/val/test}_transforms, dims, and size from the DataModule

[ananthsub]

🚀 Feature

Deprecate these properties off the LightningDataModule interface:
https://github.com/PyTorchLightning/pytorch-lightning/blob/963c26764682fa4cf64c93c5a7572ae0040e9c32/pytorch_lightning/core/datamodule.py#L94-L147

Motivation

We are auditing the Lightning components and APIs to assess opportunities for improvements:

• https://docs.google.com/document/d/1xHU7-iQSpp9KJTjI3As2EM0mfNHHr37WZYpDpwLkivA/edit#
• Review Lightning architecture & API #7740 (comment)

train_transforms, val_transforms, test_transforms, dims, and size are entirely optional to use, yet they’re on the DataModule interface. The Trainer does not rely on these, nor is the user forced to implement them. In reality, these are internal implementation details of individual datamodules. As a result, we ought to deprecate these off the DataModule interface.

Of course, users retain the ability to implement these properties in their LightningModules if they find these abstractions helpful.

Pitch

We can follow a similar approach as what was done for #7301

• Mark the properties as deprecated in v1.5. Warn if users are passing non-None values to the DataModule constructor and if they read/set their corresponding properties
• Remove the properties in v1.7

Alternatives

Keep as is

Additional context

---

If you enjoy Lightning, check out our other projects! ⚡

• Metrics: Machine learning metrics for distributed, scalable PyTorch applications.

• Flash: The fastest way to get a Lightning baseline! A collection of tasks for fast prototyping, baselining, finetuning and solving problems with deep learning

• Bolts: Pretrained SOTA Deep Learning models, callbacks and more for research and production with PyTorch Lightning and PyTorch

• Lightning Transformers: Flexible interface for high performance research using SOTA Transformers leveraging Pytorch Lightning, Transformers, and Hydra.

[Tshimanga]

@ananthsub I've submitted a PR for review that adds the v1.5 deprecation warnings. I didn't find any tests that needed to be (re)moved. By the way, is there a reason why rank_zero_deprecation isn't a decorator?

[ananthsub]

By the way, is there a reason why rank_zero_deprecation isn't a decorator?

rank_zero_deprecation can be used within functions. It doesn't always have to wrap a function to print a warning. Something like rank_zero_only is a decorator as it applies to the function as a whole.

[ajdillhoff]

If these are being deprecated, what will be the best practice for modifying transforms on datasets coming from pl_bolts?

https://lightning-bolts.readthedocs.io/en/latest/datamodules/vision.html

[akihironitta]

Hi @ajdillhoff, I believe it's still free for users to define the methods. They're just deprecated and will be removed from the base class. If you have a further issue/question, please use GitHub Discussions: https://github.com/PyTorchLightning/pytorch-lightning/discussions