nn.functional.binary_cross_entropy_with_logits¶

lucid.nn.functional.binary_cross_entropy_with_logits(input_: Tensor, target: Tensor, weights: Tensor | None = None, reduction: Literal['mean', 'sum'] | None = 'mean', eps: float = 1e-07) → Tensor¶

The binary_cross_entropy_with_logits function computes the binary cross-entropy loss directly from raw logits (before sigmoid activation), using a numerically stable formulation.

Function Signature¶

def binary_cross_entropy_with_logits(
    input_: Tensor,
    target: Tensor,
    weight: Tensor | None = None,
    reduction: str | None = "mean",
) -> Tensor

Parameters¶

input_ (Tensor): The input tensor containing raw logits of shape (N, …), where N is the batch size.
target (Tensor): The target tensor of shape matching input_, with binary values (0 or 1).
weight (Tensor | None, optional): A manual rescaling weight tensor of shape matching input_. Default: None.
reduction (str | None, optional): Specifies the reduction to apply: ‘mean’, ‘sum’, or ‘none’. Default: ‘mean’.

Returns¶

Tensor: The computed binary cross-entropy loss as a scalar or tensor, depending on the reduction method.

Examples¶

>>> import lucid.nn.functional as F
>>> input_ = Tensor([2.0, -1.0])  # raw logits
>>> target = Tensor([1.0, 0.0])
>>> loss = F.binary_cross_entropy_with_logits(input_, target)
>>> print(loss)
Tensor(...)  # Scalar loss value