class

BCELoss

extendsModule

BCELoss(weight: Tensor | None = None, reduction: str = 'mean')

source

Binary cross-entropy loss.

Measures the element-wise binary cross-entropy between predictions x (probabilities in $(0, 1)$ ) and binary targets y (values in $\{0, 1\}$ ):

\ell(x, y) = -\bigl[y \log x + (1 - y) \log(1 - x)\bigr]

The input must be valid probabilities, i.e. produced by a Sigmoid activation. Passing raw logits directly leads to undefined behaviour; use BCEWithLogitsLoss instead for better numerical stability when starting from logits.

Parameters

weightTensor= None

Element-wise weight tensor that is broadcast over the input and target tensors. Must be broadcastable to their shape.

reductionstr= 'mean'

'none' | 'mean' (default) | 'sum'.

Attributes

weightTensor or None

Optional element-wise weighting.

reductionstr

The reduction mode.

Notes

Input x : $(*)$ — probabilities in $(0, 1)$ .
Target y : $(*)$ — binary labels in $\{0, 1\}$ .
Output : scalar for 'mean' / 'sum'; $(*)$ for 'none'.

Values of x outside $(0, 1)$ will produce NaN or inf losses due to the logarithm.
For logits (pre-sigmoid values), prefer BCEWithLogitsLoss which uses the identity $\log(1 + e^{-x}) = \max(x,0) + \log(1 + e^{-|x|})$ .

Examples

Basic binary classification:
>>> import lucid
>>> import lucid.nn as nn
>>> criterion = nn.BCELoss()
>>> probs   = lucid.tensor([0.9, 0.1, 0.8, 0.3])
>>> targets = lucid.tensor([1.0, 0.0, 1.0, 0.0])
>>> loss = criterion(probs, targets)
With element-wise sample weighting:
>>> import lucid
>>> import lucid.nn as nn
>>> criterion = nn.BCELoss(weight=lucid.tensor([2.0, 1.0, 2.0, 1.0]))
>>> probs   = lucid.tensor([0.7, 0.2, 0.6, 0.4])
>>> targets = lucid.tensor([1.0, 0.0, 1.0, 0.0])
>>> loss = criterion(probs, targets)

Methods (3)

dunder

init

→None

__init__(weight: Tensor | None = None, reduction: str = 'mean')

source

Initialise the BCELoss module. See the class docstring for parameter semantics.

forward

→Tensor

forward(x: Tensor, target: Tensor)

source

Compute the loss between predictions and targets.

Parameters

xTensor

Input tensor.

targetTensor

Input tensor.

Returns

Tensor

Scalar loss (or unreduced tensor depending on reduction).

extra_repr

→str

extra_repr()

source

Return a string representation of the layer's configuration.