Skip to content

Utils

Source

Lightning utils.

Series dataclass

An easy to plot series extracting from training log.

epoch instance-attribute

X-axis values representing training epochs.

value instance-attribute

Y-axis values representing the quantity recorded over the epoch.

TrainingReport dataclass

Series extracted from training logs.

learning_rate instance-attribute

Learning rate.

metrics instance-attribute

Metrics, can be empty.

train_loss instance-attribute

Training loss.

val_loss instance-attribute

Validation loss.

load_config_from_checkpoint(checkpoint_path)

Load a CAREamics config from a checkpoint.

Some fields, if missing, will be populated by defaults. Namely, version, training_config and experiment_name.

The default for experiment_name will be "loaded_from_<checkpoint_filename>".

Parameters:

  • checkpoint_path (Path) –

    Path to the PyTorch Lightning checkpoint file.

Returns:

Raises:

  • ValueErrors: –

    If certain required information is not found in the checkpoint.

load_module_from_checkpoint(checkpoint_path)

Load a trained CAREamics module from checkpoint.

Automatically detects the algorithm type from the checkpoint and loads the appropriate module with trained weights.

Parameters:

  • checkpoint_path (Path) –

    Path to the PyTorch Lightning checkpoint file.

Returns:

  • CAREamicsModule –

    Lightning module with loaded weights.

Raises:

  • ValueError –

    If the algorithm type cannot be determined from the checkpoint.

read_csv_logger(log_folder, experiment_name, version=None)

Return plottable training curves from Lightning CSV logs.

Parameters:

  • log_folder (Path or str) –

    Path to the folder containing the csv logs.

  • experiment_name (str) –

    Name of the experiment.

  • version (int or None, default: = None ) –

    Version number to load, if None then the latest version is loaded.

Returns:

  • TrainingReport –

    Dataclass containing train and validation loss, learning rate, and any discovered validation metrics.

Raises:

  • ValueError –

    If version is specified but not found.