Losses

LabelSmoothingCrossEntropy

Bases: Module

Label smoothing is designed to make the model a little bit less certain of it's decision by changing a little bit its target: instead of wanting to predict 1 for the correct class and 0 for all the others, we ask it to predict 1 - ε for the correct class and ε for all the others, with ε a (small) positive number.

__init__(eps=0.1, reduction='mean')

Parameters:

Name	Type	Description	Default
eps	float	Weight for the interpolation formula.	0.1
reduction	str	Reduction applied to the loss tensor.	mean

NoneReduce

Force non-reduction on the loss tensor so it can used later in methods such as Mixup or LabelSmoothing.

__init__(loss_func)

Parameters:

Name	Type	Description	Default
loss_func	Callable	Loss function.	required

reduce_loss(loss, reduction=None)

Reduce the loss tensor using reduction method. If reduction is None, returns the passed loss tensor.

Parameters:

Name	Type	Description	Default
loss	Tensor	Loss tensor.	required
reduction	str | None	Reduction applied to the loss tensor.	None

Returns:

Type	Description
Tensor	Reduced loss tensor.