mednet.utils.checkpointer¶

Module Attributes

`CheckpointAliasType`	Definition of a Checkpoint alias type to make it flexible to user defined metric to monitor.
`CHECKPOINT_ALIASES`	Standard paths where checkpoints may be (if produced with this framework).

Functions

`get_checkpoint_to_resume_training`(path)	Return the best checkpoint file path to resume training from.
`get_checkpoint_to_run_inference`(path, metric)	Return the best checkpoint file path to run inference with.

mednet.utils.checkpointer.CheckpointAliasType = str | collections.abc.Callable[[str, str], str]¶: Definition of a Checkpoint alias type to make it flexible to user defined metric to monitor.

mednet.utils.checkpointer.CHECKPOINT_ALIASES: dict[str, str | Callable[[str, str], str]] = {'best': <function <lambda>>, 'periodic': 'model-at-{epoch}'}¶: Standard paths where checkpoints may be (if produced with this framework).

mednet.utils.checkpointer.get_checkpoint_to_resume_training(path)[source]¶

Return the best checkpoint file path to resume training from.

Parameters:: path (Path) – The base directory containing either the “periodic” checkpoint to start the training session from.
Returns:: Path to a checkpoint file that exists on disk.
Return type:: Path
Raises:: FileNotFoundError – If none of the checkpoints can be found on the provided directory.

mednet.utils.checkpointer.get_checkpoint_to_run_inference(path, metric, mode='min')[source]¶

Return the best checkpoint file path to run inference with.

Parameters:

path (Path) – The base directory containing either the “best”, “last” or “periodic” checkpoint to start the training session from.
metric (str) – Name of the metric used for monitoring and saving the best checkpoint (default: “loss”).
mode (Literal['min', 'max']) – One of {“min”, “max”}.

Returns:

Path to a checkpoint file that exists on disk.

Return type:

Path

Raises:

FileNotFoundError – If none of the checkpoints can be found on the provided directory.