helios.losses.utils

Attributes

LOSS_REGISTRY

Global instance of the registry for loss functions.

Functions

create_loss(→ torch.nn.Module)

Create the loss function for the given type.

Module Contents

helios.losses.utils.LOSS_REGISTRY

Global instance of the registry for loss functions.

By default, the registry contains the following losses:

Optimizers

Loss	Name
torch.nn.L1Loss	L1Loss
torch.nn.MSELoss	MSELoss
torch.nn.CrossEntropyLoss	CrossEntropyLoss
torch.nn.CTCLoss	CTCLoss
torch.nn.NLLLoss	NLLLoss
torch.nn.PoissonNLLLoss	PoissonNLLLoss
torch.nn.GaussianNLLLoss	GaussianNLLLoss
torch.nn.KLDivLoss	KLDivLoss
torch.nn.BCELoss	BCELoss
torch.nn.BCEWithLogitsLoss	BCEWithLogitsLoss
torch.nn.MarginRankingLoss	MarginRankingLoss
torch.nn.HingeEmbeddingLoss	HingeEmbeddingLoss
torch.nn.MultiLabelMarginLoss	MultiLabelMarginLoss
torch.nn.HuberLoss	HuberLoss
torch.nn.SmoothL1Loss	SmoothL1Loss
torch.nn.SoftMarginLoss	SoftMarginLoss
torch.nn.MultiLabelMarginLoss	MultiLabelMarginLoss
torch.nn.CosineEmbeddingLoss	CosineEmbeddingLoss
torch.nn.MultiMarginLoss	MultiMarginLoss
torch.nn.TripletMarginLoss	TripletMarginLoss
torch.nn.TripletMarginWithDistanceLoss	TripletMarginWithDistanceLoss

Example

import helios.losses as hll

# This automatically registers your loss function.
@hll.LOSS_REGISTRY.register
class MyLoss:
    ...

# Alternatively you can manually register a loss function like this:
hll.LOSS_REGISTRY.register(MyLoss)

helios.losses.utils.create_loss(type_name: str, *args: Any, **kwargs: Any) → torch.nn.Module

Create the loss function for the given type.

Parameters:

• type_name – the type of the loss to create.

• args – positional arguments to pass into the loss.

• kwargs – keyword arguments to pass into the loss.

Returns:

The loss function.