Convergence

This module implements convergence criteria

Convergence

Implements convergence criterium. Convergence is when change in patience epochs is smaller than delta.

__call__(self, iteration, l1_norm) special

Parameters:

Name	Type	Description	Default
epoch	int	Current epoch of the optimization	required
l1_norm	Tensor	Value of the L1 norm	required

Source code in deepymod/training/convergence.py

def __call__(self, iteration: int, l1_norm: torch.Tensor) -> bool:
    """

    Args:
        epoch (int): Current epoch of the optimization
        l1_norm (torch.Tensor): Value of the L1 norm
    """
    converged = False # overwrite later

    # Initialize if doesn't exist
    if self.start_l1 is None:
        self.start_l1 = l1_norm
        self.start_iteration = iteration

    # Check if change is smaller than delta and if we've exceeded patience
    elif torch.abs(self.start_l1 - l1_norm).item() < self.delta:
        if (iteration - self.start_iteration) >= self.patience:
            converged = True

    # If not, reset and keep going
    else:
        self.start_l1 = l1_norm
        self.start_iteration = iteration

    return converged

__init__(self, patience=200, delta=0.001) special

Implements convergence criterium. Convergence is when change in patience epochs is smaller than delta. Args: patience (int): how often to check for convergence delta (float): desired accuracy

Source code in deepymod/training/convergence.py

def __init__(self, patience: int = 200, delta: float = 1e-3) -> None:
    '''Implements convergence criterium. Convergence is when change in patience
    epochs is smaller than delta.
    Args: 
        patience (int): how often to check for convergence
        delta (float): desired accuracy
    '''
    self.patience = patience
    self.delta = delta
    self.start_iteration = None
    self.start_l1 = None