logaddexp

→Tensor

logaddexp(input: Tensor, other: Tensor | Scalar)

Element-wise log-sum-exp of two operands.

Computes $\log(e^{x} + e^{y})$ using the log-sum-exp trick to avoid overflow when x or y is large.

inputTensor

Left operand.

otherTensor or scalar

Right operand. Broadcasts against input following the standard broadcasting rules; Python scalars are promoted to a tensor of matching dtype.

Tensor

Element-wise result with shape broadcast(input.shape, other.shape) and dtype determined by the usual type-promotion rules.

Mathematical definition:

\text{out}_i = \log(e^{\text{input}_i} + e^{\text{other}_i}) = \max(x, y) + \log(1 + e^{-|x - y|})

Widely used in probabilistic models for marginalizing in log-space.

>>> import lucid
>>> a = lucid.tensor([1.0, 2.0, 3.0])
>>> b = lucid.tensor([4.0, 5.0, 6.0])
>>> lucid.logaddexp(a, b)
Tensor([...])