kosmos.ml.config.factories.lr_scheduler

Classes

class LearningRateSchedulerConfig

Bases: abc.ABC

Learning rate scheduler configuration.

Methods

get_instance(optimizer: torch.optim.Optimizer) → torch.optim.lr_scheduler.LRScheduler

Get the learning rate scheduler instance.

Returns:

Learning rate scheduler instance.

Return type:

LRScheduler

---

class StepLearningRateSchedulerConfig(step_size: int, gamma: float = 0.1)

Bases: LearningRateSchedulerConfig

Step learning rate scheduler configuration.

step_size

Period of learning rate decay.

Type:

int

gamma

Multiplicative factor of learning rate decay.

Type:

float

Initialize the step learning rate scheduler configuration.

Parameters:

• step_size (int) – Period of learning rate decay.

• gamma (float) – Multiplicative factor of learning rate decay. Defaults to 0.1.

Methods

get_instance(optimizer: torch.optim.Optimizer) → torch.optim.lr_scheduler.StepLR

Get the step learning rate scheduler instance.

Parameters:

optimizer (Optimizer) – Optimizer instance.

Returns:

Step learning rate scheduler instance.

Return type:

StepLR

---

class ExponentialLearningRateSchedulerConfig(gamma: float)

Bases: LearningRateSchedulerConfig

Exponential learning rate scheduler configuration.

gamma

Multiplicative factor of learning rate decay.

Type:

float

Initialize the exponential learning rate scheduler configuration.

Parameters:

gamma (float) – Multiplicative factor of learning rate decay.

Methods

get_instance(optimizer: torch.optim.Optimizer) → torch.optim.lr_scheduler.ExponentialLR

Get the exponential learning rate scheduler instance.

Parameters:

optimizer (Optimizer) – Optimizer instance.

Returns:

Exponential learning rate scheduler instance.

Return type:

ExponentialLR

---

class CosineLearningRateSchedulerConfig(max_epochs: int, min_lr: float = 0.0)

Bases: LearningRateSchedulerConfig

Cosine annealing learning rate scheduler configuration.

max_epochs

Maximum number of epochs (iterations for the scheduler).

Type:

int

min_lr

Minimum learning rate.

Type:

float

Initialize the cosine learning rate scheduler configuration.

Parameters:

• max_epochs (int) – Maximum number of epochs (iterations for the scheduler).

• min_lr (float) – Minimum learning rate. Defaults to 0.0.

Methods

get_instance(optimizer: torch.optim.Optimizer) → torch.optim.lr_scheduler.CosineAnnealingLR

Get the cosine learning rate scheduler instance.

Parameters:

optimizer (Optimizer) – Optimizer instance.

Returns:

Cosine learning rate scheduler instance.

Return type:

CosineAnnealingLR