trainer.validated/tested/predicted_ckpt_path: mark private or deprecate

[jjenniferdai]

Proposed refactor / Motivation

As detailed in this comment: #10573 (comment):

• These properties are intended as read-only, they do not support any actual functionality (they do not and should not set the load path).

• The main reason to keep them is to track the exact ckpt_path used in the case of automatic "best" path resolution (https://github.com/PyTorchLightning/pytorch-lightning/blob/master/pytorch_lightning/trainer/trainer.py#L1363).

Pitch

To clarify/emphasize this, we can either:

1. Set to private. Also just need one attribute (e.g. trainer._ckpt_path or trainer._last_restored_ckpt_path).

2. Can we deprecate these attributes entirely? checkpoint_connector gives us info on the exact path here: https://github.com/PyTorchLightning/pytorch-lightning/blob/master/pytorch_lightning/trainer/connectors/checkpoint_connector.py#L75

Additional context

---

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

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

• Lite: enables pure PyTorch users to scale their existing code on any kind of device while retaining full control over their own loops and optimization logic.

• Flash: The fastest way to get a Lightning baseline! A collection of tasks for fast prototyping, baselining, fine-tuning, 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.

cc @justusschock @awaelchli @akihironitta @rohitgr7 @ananthsub @ninginthecloud @tchaton

[awaelchli]

At the most we would need only one of these attributes. It does not make sense to have one dedicated to each trainer method.

[jjenniferdai]

At the most we would need only one of these attributes. It does not make sense to have one dedicated to each trainer method.

updated this in pitch option 1 !

[awaelchli]

Since lately we changed the resume arguments to ckpt_path, I would say trainer.ckpt_path is most reasonable.

[carmocca]

I agree with ^

[awaelchli]

If I understand correctly, #11696 addressed this. Closing.