lucid.sub

lucid.sub(a: Tensor, b: Tensor, /) → Tensor

The sub function performs element-wise subtraction between two Tensor objects. It returns a new Tensor representing the difference, with gradient support for backpropagation.

Function Signature

def sub(a: Tensor, b: Tensor) -> Tensor

Parameters

  • a (Tensor): The first tensor in the subtraction operation.

  • b (Tensor): The second tensor in the subtraction operation.

Returns

  • Tensor:

    A new Tensor representing the element-wise difference between a and b. If either a or b requires gradients, the resulting tensor will also require gradients.

Forward Calculation

The forward calculation for the subtraction operation is:

\[\text{out} = a - b\]

where \(a\) and \(b\) are the data contained in the tensors a and b, respectively.

Backward Gradient Calculation

For each tensor a and b involved in the subtraction, the gradient with respect to the output (out) is computed as follows:

\[\frac{\partial \text{out}}{\partial a} = 1, \quad \frac{\partial \text{out}}{\partial b} = -1\]

Examples

Using sub to subtract one tensor from another:

>>> import lucid
>>> a = Tensor([5.0, 7.0, 9.0], requires_grad=True)
>>> b = Tensor([4.0, 5.0, 6.0], requires_grad=True)
>>> out = lucid.sub(a, b)
>>> print(out)
Tensor([1.0, 2.0, 3.0], grad=None)

After calling backward() on out, gradients for a and b will be accumulated based on the backpropagation rules:

>>> out.backward()
>>> print(a.grad)
[1.0, 1.0, 1.0]
>>> print(b.grad)
[-1.0, -1.0, -1.0]