composer.optim.scheduler_hparams

Hyperparameters for schedulers.

Hparams

These classes are used with yahp for YAML-based configuration.

ConstantSchedulerHparams	Hyperparameters for the ConstantScheduler scheduler.
CosineAnnealingSchedulerHparams	Hyperparameters for the CosineAnnealingScheduler scheduler.
CosineAnnealingWarmRestartsSchedulerHparams	Hyperparameters for the CosineAnnealingWarmRestartsScheduler scheduler.
CosineAnnealingWithWarmupSchedulerHparams	Hyperparameters for the CosineAnnealingWithWarmupScheduler scheduler.
ExponentialSchedulerHparams	Hyperparameters for the ExponentialScheduler scheduler.
LinearSchedulerHparams	Hyperparameters for the LinearScheduler scheduler.
LinearWithWarmupSchedulerHparams	Hyperparameters for the LinearWithWarmupScheduler scheduler.
MultiStepSchedulerHparams	Hyperparameters for the MultiStepScheduler scheduler.
MultiStepWithWarmupSchedulerHparams	Hyperparameters for the MultiStepWithWarmupScheduler scheduler.
PolynomialSchedulerHparams	Hyperparameters for the PolynomialScheduler scheduler.
SchedulerHparams	Base class for scheduler hyperparameter classes.
StepSchedulerHparams	Hyperparameters for the StepScheduler scheduler.

class composer.optim.scheduler_hparams.ConstantSchedulerHparams(alpha=1.0, t_max='1dur')

Bases: composer.optim.scheduler_hparams.SchedulerHparams

Hyperparameters for the ConstantScheduler scheduler.

See ConstantScheduler for documentation.

Parameters

• alpha (float, optional) – See ConstantScheduler.

• t_max (str, optional) – See ConstantScheduler.

class composer.optim.scheduler_hparams.CosineAnnealingSchedulerHparams(t_max='1dur', alpha_f=0.0)

Bases: composer.optim.scheduler_hparams.SchedulerHparams

Hyperparameters for the CosineAnnealingScheduler scheduler.

See CosineAnnealingScheduler for documentation.

Parameters

• t_max (str, optional) – See CosineAnnealingScheduler.

• alpha_f (float, optional) – See CosineAnnealingScheduler.

class composer.optim.scheduler_hparams.CosineAnnealingWarmRestartsSchedulerHparams(t_0='1dur', alpha_f=0.0, t_mult=1.0)

Bases: composer.optim.scheduler_hparams.SchedulerHparams

Hyperparameters for the CosineAnnealingWarmRestartsScheduler scheduler.

See CosineAnnealingWarmRestartsScheduler for documentation.

Parameters

• t_0 (str, optional) – See CosineAnnealingWarmRestartsScheduler.

• alpha_f (float, optional) – See CosineAnnealingWarmRestartsScheduler.

• t_mult (float, optional) – See CosineAnnealingWarmRestartsScheduler.

class composer.optim.scheduler_hparams.CosineAnnealingWithWarmupSchedulerHparams(t_warmup, t_max='1dur', alpha_f=0.0)

Bases: composer.optim.scheduler_hparams.SchedulerHparams

Hyperparameters for the CosineAnnealingWithWarmupScheduler scheduler.

See CosineAnnealingWithWarmupScheduler for documentation.

Parameters

• t_warmup (str) – See CosineAnnealingWithWarmupScheduler.

• t_max (str, optional) – See CosineAnnealingWithWarmupScheduler.

• alpha_f (float, optional) – See CosineAnnealingWithWarmupScheduler.

class composer.optim.scheduler_hparams.ExponentialSchedulerHparams(gamma, decay_period='1ep')

Bases: composer.optim.scheduler_hparams.SchedulerHparams

Hyperparameters for the ExponentialScheduler scheduler.

See ExponentialScheduler for documentation.

Parameters

• gamma (float) – See ExponentialScheduler.

• decay_period (str, optional) – See ExponentialScheduler.

class composer.optim.scheduler_hparams.LinearSchedulerHparams(alpha_i=1.0, alpha_f=0.0, t_max='1dur')

Bases: composer.optim.scheduler_hparams.SchedulerHparams

Hyperparameters for the LinearScheduler scheduler.

See LinearScheduler for documentation.

Parameters

• alpha_i (float, optional) – See LinearScheduler.

• alpha_f (float, optional) – See LinearScheduler.

• t_max (str, optional) – See LinearScheduler.

class composer.optim.scheduler_hparams.LinearWithWarmupSchedulerHparams(t_warmup, alpha_i=1.0, alpha_f=0.0, t_max='1dur')

Bases: composer.optim.scheduler_hparams.SchedulerHparams

Hyperparameters for the LinearWithWarmupScheduler scheduler.

See LinearWithWarmupScheduler for documentation.

Parameters

• t_warmup (str) – See LinearWithWarmupScheduler.

• alpha_i (float, optional) – See LinearWithWarmupScheduler.

• alpha_f (float, optional) – See LinearWithWarmupScheduler.

• t_max (str, optional) – See LinearWithWarmupScheduler.

class composer.optim.scheduler_hparams.MultiStepSchedulerHparams(milestones, gamma=0.1)

Bases: composer.optim.scheduler_hparams.SchedulerHparams

Hyperparameters for the MultiStepScheduler scheduler.

See MultiStepScheduler for documentation.

Parameters

• milestones (List[str]) – See MultiStepScheduler.

• gamma (float, optional) – See MultiStepScheduler.

class composer.optim.scheduler_hparams.MultiStepWithWarmupSchedulerHparams(t_warmup, milestones, gamma=0.1)

Bases: composer.optim.scheduler_hparams.SchedulerHparams

Hyperparameters for the MultiStepWithWarmupScheduler scheduler.

See MultiStepWithWarmupScheduler for documentation.

Parameters

• t_warmup (str,) – See MultiStepWithWarmupScheduler.

• milestones (List[str]) – See MultiStepWithWarmupScheduler.

• gamma (float, optional) – See MultiStepWithWarmupScheduler.

class composer.optim.scheduler_hparams.PolynomialSchedulerHparams(power, t_max='1dur', alpha_f=0.0)

Bases: composer.optim.scheduler_hparams.SchedulerHparams

Hyperparameters for the PolynomialScheduler scheduler.

See PolynomialScheduler for documentation.

Parameters

• power (float) – See PolynomialScheduler.

• t_max (str, optional) – See PolynomialScheduler.

• alpha_f (float, optional) – See PolynomialScheduler.

class composer.optim.scheduler_hparams.SchedulerHparams

Bases: yahp.hparams.Hparams, abc.ABC

Base class for scheduler hyperparameter classes.

Scheduler parameters that are added to TrainerHparams (e.g. via YAML or the CLI) are initialized in the training loop.

initialize_object()

Initializes the scheduler.

class composer.optim.scheduler_hparams.StepSchedulerHparams(step_size, gamma=0.1)

Bases: composer.optim.scheduler_hparams.SchedulerHparams

Hyperparameters for the StepScheduler scheduler.

See StepScheduler for documentation.

Parameters

• step_size (str, optional) – See StepScheduler.

• gamma (float, optional) – See StepScheduler.