loss.cross_entropy_loss

LabelSmoothCELoss

class ding.torch_utils.loss.cross_entropy_loss.LabelSmoothCELoss(ratio: float)

Overview:

Label smooth cross entropy loss.

Interfaces:

forward

forward(logits: torch.Tensor, labels: torch.LongTensor) → torch.Tensor

Overview:

Calculate label smooth cross entropy loss.

Arguments:

• logits (torch.Tensor): Predicted logits.

• labels (torch.LongTensor): Ground truth.

Returns:

• loss (torch.Tensor): Calculated loss.

SoftFocalLoss

class ding.torch_utils.loss.cross_entropy_loss.SoftFocalLoss(gamma: int = 2, weight: Optional[Any] = None, size_average: bool = True, reduce: Optional[bool] = None)

Overview:

Soft focal loss.

Interfaces:

forward

forward(inputs: torch.Tensor, targets: torch.LongTensor) → torch.Tensor

Overview:

Calculate soft focal loss.

Arguments:

• logits (torch.Tensor): Predicted logits.

• labels (torch.LongTensor): Ground truth.

Returns:

• loss (torch.Tensor): Calculated loss.

build_ce_criterion

Overview:

Get a cross enntropy loss instance according to given config.

Arguments:

• cfg (dict)

Returns:

• loss (nn.Module): loss function instance

loss.multi_logits_loss

MultiLogitsLoss

class ding.torch_utils.loss.multi_logits_loss.MultiLogitsLoss(criterion: Optional[str] = None, smooth_ratio: float = 0.1)

Overview:

Base class for supervised learning on linklink, including basic processes.

Interface:

forward

forward(logits: torch.Tensor, labels: torch.LongTensor) → torch.Tensor

Overview:

Calculate multiple logits loss.

Arguments:

• logits (torch.Tensor): Predicted logits, whose shape must be 2-dim, like (B, N).

• labels (torch.LongTensor): Ground truth.

Returns:

• loss (torch.Tensor): Calculated loss.