rsub

→Tensor

rsub(a: Tensor, b: Tensor | Scalar, alpha: float = ...)

source

Reverse subtraction: $b - \alpha \cdot a$ .

Useful when b is a Python scalar that cannot left-bind the - operator against a tensor in a particular calling convention.

Parameters

aTensor

Right operand (subtrahend, multiplied by alpha).

bTensor | Scalar

Left operand (minuend).

alphafloat

Scalar multiplier applied to a. Defaults to 1.0.

Returns

Tensor

Element-wise difference of the broadcast shape of a and b.

Notes

Mathematical definition:

\text{out}_i = b_i - \alpha \cdot a_i

Equivalent to subtract with the operands reversed (and alpha re-attached to the new second argument). Gradient w.r.t. a is $-\alpha$ ; w.r.t. b is $1$ .

Examples

>>> import lucid
>>> a = lucid.tensor([1.0, 2.0])
>>> lucid.rsub(a, 10.0)
Tensor([9., 8.])