nexuml_library.scenarios.training.defaults

nexuml_library.scenarios.training.defaults

Default training scenario fragments.

default_training

default_training(
    lr: float = 0.001,
    batch_size: BatchSizeSpec | None = 64,
    max_epochs: int = 10,
    loss_keys: dict[str, float] | None = None,
    metric_keys: list[str] | None = None,
    optimizer_factory: Callable[
        ..., Optimizer
    ] = torch.optim.Adam,
) -> TrainingSpec

Create a default TrainingSpec.

Returns:

Name	Type	Description
TrainingSpec	TrainingSpec	Default training configuration with optimizer, scheduler and loss keys.

default_logging

default_logging(
    experiment_name: str = "NexuML",
    run_name: str | None = None,
    log_system_metrics: bool = False,
    use_tensorboard: bool = True,
    use_mlflow: bool = True,
    use_dvclive: bool = False,
) -> LoggingSpec

Create a default LoggingSpec.

Returns:

Name	Type	Description
LoggingSpec	LoggingSpec	Default logging configuration with optional TensorBoard, MLflow and DVC Live backends.

default_callbacks

default_callbacks() -> list[CallbackSpec]

Create a default list of CallbackSpec.

Returns:

Type	Description
list[CallbackSpec]	list[CallbackSpec]: Default callbacks for training.

default_checkpoint

default_checkpoint(
    path: str | None = None,
) -> CheckpointLoadSpec

Create a default CheckpointLoadSpec.

Returns:

Name	Type	Description
CheckpointLoadSpec	CheckpointLoadSpec	Default checkpoint loading configuration.

default_tuning

default_tuning() -> TuningSpec

Create a default tuning configuration.

Returns:

Name	Type	Description
TuningSpec	TuningSpec	Default tuning configuration.

default_exports

default_exports(
    path: str | None = None,
) -> list[ExportSpec]

Create a default exports configuration.

Returns:

Type	Description
list[ExportSpec]	list[ExportSpec]: Default exports configuration.