helios.scheduler.utils¶

Attributes¶

SCHEDULER_REGISTRY

Global instance of the registry for schedulers.

Functions¶

create_scheduler(→ torch.nn.Module)

Create the scheduler for the given type.

Module Contents¶

helios.scheduler.utils.SCHEDULER_REGISTRY¶

Global instance of the registry for schedulers.

By default, the registry contains the following schedulers:

Schedulers¶
Scheduler	Name
`torch.optim.lr_scheduler.LambdaLR`	LambdaLR
`torch.optim.lr_scheduler.MultiplicativeLR`	MultiplicativeLR
`torch.optim.lr_scheduler.StepLR`	StepLR
`torch.optim.lr_scheduler.MultiStepLR`	MultiStepLR
`torch.optim.lr_scheduler.ConstantLR`	ConstantLR
`torch.optim.lr_scheduler.LinearLR`	LinearLR
`torch.optim.lr_scheduler.ExponentialLR`	ExponentialLR
`torch.optim.lr_scheduler.PolynomialLR`	PolynomialLR
`torch.optim.lr_scheduler.CosineAnnealingLR`	CosineAnnealingLR
`torch.optim.lr_scheduler.SequentialLR`	SequentialLR
`torch.optim.lr_scheduler.ReduceLROnPlateau`	ReduceLROnPlateau
`torch.optim.lr_scheduler.CyclicLR`	CyclicLR
`torch.optim.lr_scheduler.OneCycleLR`	OneCycleLR
`torch.optim.lr_scheduler.CosineAnnealingWarmRestarts`	CosineAnnealingWarmRestarts
`helios.scheduler.schedulers.CosineAnnealingRestartLR`	CosineAnnealingRestartLR
`helios.scheduler.schedulers.MultiStepRestartLR`	MultiStepRestartLR

Example

import helios.optim as hlo
import helios.scheduler as hls

# This automatically registers your optimizer.
@hls.SCHEDULER_REGISTRY.register
class MyScheduler:
    ...

# Alternatively you can manually register a scheduler. like this:
hls.SCHEDULER_REGISTRY.register(MyScheduler)

helios.scheduler.utils.create_scheduler(type_name: str, *args: Any, **kwargs: Any) → torch.nn.Module¶

Create the scheduler for the given type.

Parameters:

type_name – the type of the scheduler to create.
args – positional arguments to pass into the scheduler.
kwargs – keyword arguments to pass into the scheduler.

Returns:

The scheduler.