exp

ExponentDecayLr

Bases: OptimScheduler

ExponentDecayLr is a class that inherits from OptimScheduler and implements an exponential decay learning rate scheduler.

It updates the learning rate based on the epoch number using the provided decay factor.

Source code in speechain/optim_sche/exp.py

class ExponentDecayLr(OptimScheduler):
    """ExponentDecayLr is a class that inherits from OptimScheduler and implements an
    exponential decay learning rate scheduler.

    It updates the learning rate based on the epoch number using the provided decay
    factor.
    """

    def sche_init(self, decay_factor: float = 0.999):
        """Initializes the exponential decay learning rate scheduler with the given
        decay factor.

        Args:
            decay_factor (float):
                The decay factor that will be used to exponentially decay the learning rate. Defaults to 0.999.
        """
        self.decay_factor = decay_factor

    def update_lr(self, real_step: int, epoch_num: int) -> float:
        """Updates the learning rate based on the current epoch number and the decay
        factor.

        Args:
            real_step (int):
                The current step in the optimization process.
                Not used in this implementation, but required for compatibility with the base class.
            epoch_num (int):
                The current epoch number.

        Returns:
            float: The updated learning rate.
        """
        return self.get_lr() * pow(self.decay_factor, epoch_num - 1)

    def extra_repr_fn(self) -> str:
        """Returns a string representation of the ExponentDecayLr object, including the
        decay factor.

        Returns:
            str: A string containing the class name and the decay factor value.
        """
        return f"decay_factor={self.decay_factor}"

extra_repr_fn()

Returns a string representation of the ExponentDecayLr object, including the decay factor.

Returns:

Name	Type	Description
str	str	A string containing the class name and the decay factor value.

Source code in speechain/optim_sche/exp.py

def extra_repr_fn(self) -> str:
    """Returns a string representation of the ExponentDecayLr object, including the
    decay factor.

    Returns:
        str: A string containing the class name and the decay factor value.
    """
    return f"decay_factor={self.decay_factor}"

sche_init(decay_factor=0.999)

Initializes the exponential decay learning rate scheduler with the given decay factor.

Parameters:

Name	Type	Description	Default
decay_factor	float	The decay factor that will be used to exponentially decay the learning rate. Defaults to 0.999.	0.999

Source code in speechain/optim_sche/exp.py

def sche_init(self, decay_factor: float = 0.999):
    """Initializes the exponential decay learning rate scheduler with the given
    decay factor.

    Args:
        decay_factor (float):
            The decay factor that will be used to exponentially decay the learning rate. Defaults to 0.999.
    """
    self.decay_factor = decay_factor

update_lr(real_step, epoch_num)

Updates the learning rate based on the current epoch number and the decay factor.

Parameters:

Name	Type	Description	Default
real_step	int	The current step in the optimization process. Not used in this implementation, but required for compatibility with the base class.	required
epoch_num	int	The current epoch number.	required

Returns:

Name	Type	Description
float	float	The updated learning rate.

Source code in speechain/optim_sche/exp.py

def update_lr(self, real_step: int, epoch_num: int) -> float:
    """Updates the learning rate based on the current epoch number and the decay
    factor.

    Args:
        real_step (int):
            The current step in the optimization process.
            Not used in this implementation, but required for compatibility with the base class.
        epoch_num (int):
            The current epoch number.

    Returns:
        float: The updated learning rate.
    """
    return self.get_lr() * pow(self.decay_factor, epoch_num - 1)