fl4health.utils.losses module

class EvaluationLosses(checkpoint, additional_losses=None)

Bases: Losses

__init__(checkpoint, additional_losses=None)

A class to store the checkpoint and additional_losses of a model along with a method to return a dictionary representation.

Parameters:

• checkpoint (torch.Tensor) – The loss used to checkpoint model (if checkpointing is enabled).

• additional_losses (dict[str, torch.Tensor] | None) – Optional dictionary of additional losses.

static aggregate(loss_meter)

Aggregates the losses in the given LossMeter into an instance of EvaluationLosses

Parameters:

loss_meter (LossMeter[EvaluationLosses]) – The loss meter object with the collected evaluation losses.

Returns:

An instance of EvaluationLosses with the aggregated losses.

Return type:

EvaluationLosses

as_dict()

Produces a dictionary representation of the object with all of the losses.

Returns:

A dictionary with the checkpoint loss, plus each one of the keys in

additional losses if they exist.

Return type:

dict[str, float]

class LossMeter(loss_meter_type, losses_type)

Bases: Generic[LossesType]

__init__(loss_meter_type, losses_type)

A meter to store a list of losses.

Parameters:

• loss_meter_type (LossMeterType) – The type of this loss meter

• losses_type (type[Losses]) – The type of the loss that will be stored. Should be one of the subclasses of Losses

static aggregate_losses_dict(loss_list, loss_meter_type)

Aggregates a list of losses dictionaries into a single dictionary according to the loss meter aggregation type

Parameters:

• loss_list (list[dict[str, torch.Tensor]]) – A list of loss dictionaries

• loss_meter_type (LossMeterType) – The type of the loss meter to perform the aggregation

Returns:

A single dictionary with the aggregated losses according to the given loss

meter type

Return type:

dict[str, torch.Tensor]

clear()

Resets the meter by re-initializing losses_list to be empty

Return type:

None

compute()

Computes the aggregation of current list of losses if non-empty.

Returns:

New Losses object with the aggregation of losses in losses_list.

Return type:

LossesType

update(losses)

Appends loss to list of losses.

Parameters:

losses (LossesType) – A losses object with checkpoint, backward and additional losses.

Return type:

None

class LossMeterType(value)

Bases: Enum

An enumeration.

ACCUMULATION = 'ACCUMULATION'

AVERAGE = 'AVERAGE'

class Losses(additional_losses=None)

Bases: ABC

__init__(additional_losses=None)

An abstract class to store the losses

Parameters:

additional_losses (dict[str, torch.Tensor] | None) – Optional dictionary of additional losses.

abstract static aggregate(loss_meter)

Aggregates the losses in the given LossMeter into an instance of Losses

Parameters:

loss_meter (LossMeter) – The loss meter object with the collected losses.

Raises:

NotImplementedError – To be implemented by child classes.

Return type:

Losses

as_dict()

Produces a dictionary representation of the object with all of the losses.

Returns:

A dictionary with the additional losses if they exist.

Return type:

dict[str, float]

class TrainingLosses(backward, additional_losses=None)

Bases: Losses

__init__(backward, additional_losses=None)

A class to store the backward and additional_losses of a model along with a method to return a dictionary representation.

Parameters:

• backward (torch.Tensor | dict[str, torch.Tensor]) – The backward loss or losses to optimize. In the normal case, backward is a Tensor corresponding to the loss of a model. In the case of an ensemble_model, backward is dictionary of losses.

• additional_losses (dict[str, torch.Tensor] | None) – Optional dictionary of additional losses.

static aggregate(loss_meter)

Aggregates the losses in the given LossMeter into an instance of TrainingLosses

Parameters:

loss_meter (LossMeter[TrainingLosses]) – The loss meter object with the collected training losses.

Returns:

An instance of TrainingLosses with the aggregated losses.

Return type:

TrainingLosses

as_dict()

Produces a dictionary representation of the object with all of the losses.

Returns:

A dictionary where each key represents one of the backward losses,

plus additional losses if they exist.

Return type:

dict[str, float]