v3.2.0

Fix PhysFormer(++) save and load path from epoch, not step.
Fix pred df path in predict.
Add default scripts.

v3.1.3

Update save detail to clarify model hyper-parameters.
Improve code readability.

v3.1.2

Updates combined distributed loss function.
Refactor following modules and configs.

v3.1.1

Add PhysFormer++ model.
Able to select PhysFormer ++ or not at main config and tuner.
Set default model as PhysFormer++.

v3.1.0

Able to use PhysFormer model(default).
Able to select num_workers of DataLoader using workers_ratio and using gpu devices.
Able to select num_workers of DataLoader using all cpu workers.
Fix tuner suggest param types at scheduler.
Refactor a scheduler.
Change some default parameter values.
Match optuna suggest ranges with default param values.

v3.0.0

Scheduler works correctly.
Hyper-parameters(optimizer, scheduler) scientific notation.
Refactor some parameter names.
Python version and packages update.
Delete unnecessary dirs.
Fix typo.

v2.9.0

Change default values of parameters to fit the task.
Get total number of train batches in whole training process correctly.
A scheduler type changes.
Save_top_k follows max_epochs.

v2.8.0

Upload model based on upload_tag and model_detail.
Change default scripts.

v2.7.2

Versions of libraries update.

v2.7.1

Parameter names change clearly.
eta_min depends on LR.