espnet2.schedulers.abs_scheduler.AbsEpochStepScheduler

source

class espnet2.schedulers.abs_scheduler.AbsEpochStepScheduler

Bases: AbsScheduler

Abstract base class for epoch-based learning rate schedulers.

This class defines the interface for all epoch-based schedulers. It inherits from the AbsScheduler class and requires implementations for methods to manage the learning rate at each epoch.

None

• Parameters:epoch (int , optional) – The current epoch. Defaults to None.

step(epoch

int = None): Updates the learning rate based on the current epoch.

state_dict()

Returns the state of the scheduler as a dictionary.

load_state_dict(state)

Loads the state of the scheduler from a given dictionary.

########### Examples

Example of a custom epoch scheduler implementation

class CustomEpochScheduler(AbsEpochStepScheduler):

def __init__(self, optimizer): : self.optimizer = optimizer self.initial_lr = optimizer.param_groups[0][‘lr’]

def step(self, epoch: int = None): : new_lr = self.initial_lr * (0.1 ** (epoch // 10)) for param_group in self.optimizer.param_groups: <br/>

param_group[‘lr’] = new_lr

def state_dict(self): : return

def load_state_dict(self, state): : self.initial_lr = state[‘initial_lr’]

abstract load_state_dict(state)

Abstract base class for epoch step learning rate schedulers.

This class defines the interface for epoch-based learning rate schedulers. Subclasses must implement the step, state_dict, and load_state_dict methods to manage the learning rate scheduling.

None

• Parameters:state – The state dictionary to load into the scheduler.
• Returns: None
• Yields: None
• Raises:NotImplementedError – If the method is not implemented in a subclass.

########### Examples

class MyScheduler(AbsEpochStepScheduler): : def step(self, epoch: int = None): : # Implementation for stepping the scheduler pass <br/> def state_dict(self): : # Implementation for saving the state pass <br/> def load_state_dict(self, state): : # Implementation for loading the state pass

my_scheduler = MyScheduler() my_scheduler.load_state_dict({‘some_key’: ‘some_value’})

NOTE

This class is meant to be subclassed and should not be instantiated directly.

abstract state_dict()

Abstract base class for epoch-based step learning rate schedulers.

This class provides the interface for implementing learning rate schedulers that adjust the learning rate at the end of each epoch.

None

• Parameters:epoch (int , optional) – The current epoch. Defaults to None.
• Returns: A state dictionary containing the current state of the scheduler.
• Return type: dict
• Yields: None
• Raises:
  • NotImplementedError – If the method is not implemented in a
  • subclass. –

########### Examples

To create a custom epoch step scheduler, inherit from this class and implement the required methods:

``

`

python class MyCustomScheduler(AbsEpochStepScheduler):

def step(self, epoch: int = None): : # Custom logic to update the learning rate pass

def state_dict(self): : # Return the state of the scheduler return {}

def load_state_dict(self, state): : # Load the state into the scheduler pass

``

`

abstract step(epoch: int | None = None)

Abstract base class for epoch-based learning rate schedulers.

This class serves as a blueprint for creating custom epoch-based learning rate schedulers by inheriting from it. Implementations must define the step, state_dict, and load_state_dict methods.

None

• Parameters:epoch (int , optional) – The current epoch number. Defaults to None.
• Returns: None
• Yields: None
• Raises:NotImplementedError – If the step method is not implemented by a subclass.

########### Examples

To create a custom scheduler, you would inherit from this class and implement the required methods. For example:

class CustomEpochScheduler(AbsEpochStepScheduler): : def step(self, epoch: int = None): : # Custom logic to update learning rate pass <br/> def state_dict(self): : # Logic to return the state dictionary pass <br/> def load_state_dict(self, state): : # Logic to load the state dictionary pass