gigl.src.common.modeling_task_specs.utils.early_stop

Attributes

logger

Classes

EarlyStopper

Handles early stopping logic, keeping track of the best performing model provided some criterion

Module Contents

class gigl.src.common.modeling_task_specs.utils.early_stop.EarlyStopper(early_stop_patience, should_maximize, model=None)

Handles early stopping logic, keeping track of the best performing model provided some criterion

Parameters:

• early_stop_patience (int) – Maximum allowed number of steps for consecutive decreases in performance

• should_maximize (bool) – Whether we minimize or maximize the provided criterion

• model (Optional[nn.Module]) – Optional model to provide to early stopper class. If provided, will keep track of the state dict of the best model.

step(value)

Steps through the early stopper provided some criterion. Returns whether the provided criterion improved over the previous best criterion and whether we should early stop. :param value: Criterion used for stepping through early stopper :type value: float

Returns:

Whether there was improvement over previous best criterion bool: Whether early stop patience has been reached, indicating early stopping

Return type:

bool

Parameters:

value (float)

property best_criterion: float

Return type:

float

property best_model_state_dict: dict[str, torch.Tensor] | None

Return type:

Optional[dict[str, torch.Tensor]]

gigl.src.common.modeling_task_specs.utils.early_stop.logger