where

→Tensor

where(cond: Tensor, x: Tensor, y: Tensor)

source edit

Implementing kernel

C++where_opfree fn

Elementwise selection between x and y based on cond.

For each element position the result is x where cond is True and y otherwise. All three inputs broadcast against each other.

Parameters

condTensor

Boolean tensor.

xTensor

Values used where cond is True.

yTensor

Values used where cond is False.

Returns

Tensor

Tensor of the broadcast shape.

Notes

Mathematically,

\text{out}_i \;=\; \begin{cases} x_i & \text{if } \text{cond}_i \\ y_i & \text{otherwise} . \end{cases}

Differentiable through both x and y — the gradient is gated by cond.

Examples

>>> import lucid
>>> a = lucid.tensor([1.0, 2.0, 3.0])
>>> b = lucid.tensor([-1.0, -2.0, -3.0])
>>> lucid.where(a > 1.5, a, b)
Tensor([-1.,  2.,  3.])