Fix scheduler step bug

[alphinside]

Type of changes

Please check the type of change your PR introduces: - [x] Bugfix - [ ] Feature - [ ] Code style update (formatting, renaming) - [ ] Refactoring (no functional changes, no api changes) - [ ] Build related changes - [x] Documentation content changes - [ ] Other (please describe): ## What is the current behavior? Currently all scheduler step update is batch update, this caused several scheduler have unintended result - close #58 ## What is the new behavior?

• make general mechanism for scheduler update to hadle batch and epoch update
• update several scheduler name
• update docs
• fix bug on training pipeline if save_best_metrics is not described in experiment file

Checklist

• [x] Unit tests pass locally with my changes
• [x] I have added tests that prove my fix is effective or that my feature works
• [x] I have added necessary documentation (if appropriate)
• [x] Any dependent changes have been merged and published in downstream modules
• [x] I have filled the [Unreleased] Changelog

[alphinside]

Is there no support for ReduceLROnPlateau scheduler? I guess it is because it required validation metrics, but it could be more useful in my opinion compared to other type of scheduler.

Yes correct, I also think it will be useful, but it's implementation is not well fitted into our desgin because scheduler is tightly coupled to trainer whereas validation is in the training pipeline, though it is possible to add 'on_validation_update' type, and add it to training pipeline. And also we dont have validation loss calculation mechanism

[triwahyuu]

so, add TODO feature then?

[alphinside]

so, add TODO feature then?

yes this issue #60

[triwahyuu]

okay then, after that it would be possible to support this scheduler then?

[alphinside]

okay then, after that it would be possible to support this scheduler then?

yes possibly, though wont solve the design problem

[triwahyuu]

okay then