Goal

• As a developer, I want to use a customized training pipeline that differs from the pipeline in Labrador project, so that I can tweak more training knobs and make the training process as efficient as possible.

Acceptance Criteria

Given a customized training pipeline,  when enabling the loss curve,  then the training pipeline can plot the training loss curve easily.

Given a customized training pipeline,  when specifying different loss functions, then the gradient for the weights update in the pre-trained model should rely on the loss functions that the developer specified.

Given a customized training pipeline,  when specifying different optimizers, then the weights update in the pre-trained model should conform to the optimizers that the developer specified.

Common loss functions: https://builtin.com/machine-learning/common-loss-functions
Common optimizers: https://ml-cheatsheet.readthedocs.io/en/latest/optimizers.html