Module epiclass.core.trainer

Trainer class extensions module

Functions

def define_callbacks(early_stop_limit: int | None, show_summary=True)

Returns list of PyTorch trainer callbacks. RichModelSummary, EarlyStopping, ModelCheckpoint

Will only save last epoch model if there is no early stopping.

Classes

class MyTrainer (general_log_dir: str, model, **kwargs)

Personalized trainer

Metrics expect probabilities and not logits.

Expand source code

class MyTrainer(pl.Trainer):
    """Personalized trainer"""

    def __init__(self, general_log_dir: str, model, **kwargs):
        """Metrics expect probabilities and not logits."""
        super().__init__(**kwargs)

        self.best_checkpoint_file = Path(general_log_dir) / "best_checkpoint.list"
        self.my_model = model
        self.batch_size = None

    def fit(self, *args, verbose=True, **kwargs):
        """Base pl.Trainer.fit function, but also prints the batch size."""
        self.batch_size = kwargs["train_dataloaders"].batch_size
        if verbose:
            print(f"Training batch size : {self.batch_size}")
        super().fit(*args, **kwargs)

    def save_model_path(self):
        """Save best checkpoint path to a file."""
        try:
            model_path = self.checkpoint_callback.best_model_path  # type: ignore
            print(f"Saving model to {model_path}")
            with open(self.best_checkpoint_file, "a", encoding="utf-8") as ckpt_file:
                ckpt_file.write(f"{model_path} {datetime.now()}\n")
        except AttributeError:
            print("Cannot save model, no checkpoint callback.")

    def print_hyperparameters(self):
        """Print training hyperparameters."""
        print("--TRAINING HYPERPARAMETERS--")
        print(f"L2 scale : {self.my_model.l2_scale}")
        print(f"Dropout rate : {self.my_model.dropout_rate}")
        print(f"Learning rate : {self.my_model.learning_rate}")
        try:
            stop_callback = self.early_stopping_callback
            print(f"Patience : {stop_callback.patience}")  # type: ignore
            print(f"Monitored value : {stop_callback.monitor}")  # type: ignore
        except AttributeError:
            print("No early stopping.")

Ancestors

• pytorch_lightning.trainer.trainer.Trainer

Methods

def fit(self, *args, verbose=True, **kwargs)

Base pl.Trainer.fit function, but also prints the batch size.

def print_hyperparameters(self)

Print training hyperparameters.

def save_model_path(self)

Save best checkpoint path to a file.