Deprecated sg_logger_params

[shaydeci]

No description provided.

[BloodAxe]

Should we also introduce logger names consistent with class names and deprecate old snake_case names? E.g "DagsHubSGLogger" instead of "dagshub_sg_logger"?

[shaydeci]

I was thinking about that too actually.
Should I? Then since we don't want to break (not actually breaking since this is a new format) we would also deprecate the old way?
Or should we just leave both options ?

[Louis-Dupont]

I think it would be great, add the class names, and deprecate the old names.

[BloodAxe]

How would one "unset" the base_sg_logger once it is already in defaults?

I feel this may be the problem if someone would like to use non-default logger:

defaults:
  training_hyperparams: default_train_params

training_hyperparams:
  sg_logger:
    wandb_sg_logger:
      ...

In this case the resolved config would include:

training_hyperparams:
  sg_logger:
    base_sg_logger:
      ...
    wandb_sg_logger:
      ...

And that would lead to an exception in factory.

Second use case that would not work is using W&B via Makefile (To keep YAML files intact):

WANDB_PARAMS = training_hyperparams.sg_logger=wandb_sg_logger +training_hyperparams.sg_logger_params.api_server=https://wandb.research.deci.ai +training_hyperparams.sg_logger_params.entity=super-gradients training_hyperparams.sg_logger_params.launch_tensorboard=false training_hyperparams.sg_logger_params.monitor_system=true +training_hyperparams.sg_logger_params.project_name=PoseEstimation training_hyperparams.sg_logger_params.save_checkpoints_remote=true training_hyperparams.sg_logger_params.save_logs_remote=true training_hyperparams.sg_logger_params.save_tensorboard_remote=false training_hyperparams.sg_logger_params.tb_files_user_prompt=false

coco2017_yolo_nas_s:
	python src/super_gradients/train_from_recipe.py --config-name=coco2017_yolo_nas_s $(WANDB_PARAMS)

[shaydeci]

Strong point right there...
I will think of something.