LRScheduler

class braintools.optim.LRScheduler(base_lr=0.001, last_epoch=0)

Base class for learning rate schedulers.

Can be used either standalone (passed to optimizer at initialization) or attached to an optimizer later.

apply(apply_fn)

Apply a function to modify the current learning rate.

attach_optimizer(optimizer)

Attach this scheduler to an optimizer.

get_last_lr()

Return the most recently computed learning rate(s).

Mirrors the PyTorch LRScheduler.get_last_lr API.

Return type:

List[float]

get_lr()

Calculate learning rate.

load_state_dict(state_dict)

Load scheduler state from dictionary.

state_dict()

Return scheduler state as dictionary.

step(epoch=None)

Update learning rate.

step_epoch()

Step the scheduler by one epoch.