Ability to set default_root_dir as per the pl trainer specifications

[vmajor]

Description

The ability to set the default_root_dir for checkpointing during train() of PyTorch Lighning GluonTS models seems to be missing. Being able to set default_root_dir arbitrarily would allow training of multiple models within the same root path and would simplify resource sharing.

Currently all models place their checkpoints and logs inside lightning_logs making it unnecessarily cumbersome to manage the trained models programmatically.

References

Reference: https://lightning.ai/docs/pytorch/stable/common/checkpointing_basic.html

# saves checkpoints to 'some/path/' at every epoch end
trainer = Trainer(default_root_dir="some/path/")

[lostella]

@vmajor is this something the trainer_kwargs argument to estimators does?

See here and here

[vmajor]

Yes it is and I tried this, but it was not working. Now I tried it again, and it is working... sigh. I will close this issue.

edit: thank you :)