classy.pl_callbacks.best_checkpoint

Classes

ModelCheckpointWithBest

class ModelCheckpointWithBest(pytorch_lightning.callbacks.model_checkpoint.ModelCheckpoint)

##

A callback that explicitly saves the best checkpoint with best.ckpt. Note that the best checkpoint is duplicated, rather than linked, in best.ckpt

__init__

def __init__(
    dirpath: Union[str, pathlib.Path, None] = None,
    filename: Optional[str] = None,
    monitor: Optional[str] = None,
    verbose: bool = False,
    save_last: Optional[bool] = None,
    save_top_k: int = 1,
    save_weights_only: bool = False,
    mode: str = 'min',
    auto_insert_metric_name: bool = True,
    every_n_train_steps: Optional[int] = None,
    train_time_interval: Optional[datetime.timedelta] = None,
    every_n_epochs: Optional[int] = None,
    save_on_train_epoch_end: Optional[bool] = None,
    every_n_val_epochs: Optional[int] = None,
)

##

on_validation_end

def on_validation_end(
    self,
    trainer,
    pl_module,
)

##

Save a checkpoint at the end of the validation stage.