Schedulers

TrainLoop-based training supports the use of learning rate schedulers. The built-in common learning rate schedulers can be found in the aitoolbox.torchtrain.schedulers sub-package. Currently AIToolbox comes out of the bag with the following scheduler types:

Schedulers are given to any TrainLoop type the same way as callbacks via the callbacks list provided to fit().

Implementing New Schedulers

Under the hood the schedulers are just AIToolbox Callbacks. Consequently when desired, new learning rate schedulers can easily be implemented by just inheriting them from the commonly used aitoolbox.torchtrain.callbacks.abstract.AbstractCallback base class and implementing the necessary learning rate scheduling logic.