Elementwise exponential transforms
Transforming logits or other tensor values into positive exponential-scale quantities.
PyTorch · Function
torch.exp
Computes the elementwise exponential of a tensor. It can optionally write the result to a supplied output tensor.
How it works
torch.exp applies the exponential function element by element, producing a tensor whose values satisfy yᵢ = eˣᵢ. You can optionally provide an output tensor through the keyword-only out argument.
Syntax
torch.exp(input, *, out=None) → TensorParameters
| Name | Type | Description |
|---|---|---|
input | Tensor | The input tensor whose elements are exponentiated. |
out | Tensor, optional | Optional output tensor that receives the result. |
Returns
A new Tensor containing the elementwise exponential of input, or the supplied output tensor when out is used.
Key points
- The function creates a result tensor containing the exponential of each input element.
- The required input is a PyTorch Tensor.
- An optional destination tensor can receive the computed values through out.
- A tensor containing 0 and log(2) demonstrates outputs of 1 and 2.
Common mistakes
- If supplied, out must be a tensor intended to receive the result; use it as a keyword argument.
Example: Compute elementwise exponentials
import math
import torch
x = torch.tensor([0.0, math.log(2.0)])
y = torch.exp(x)
print(y) # tensor([1., 2.])Computes the exponential independently for each element of the input tensor.
When to use it
Recovering values from natural logarithms
Constructing values such as powers of e from tensorized logarithms.
Destination-buffer computation
Writing results into an existing tensor buffer when an output destination is available.
Interview questions
QWhat does torch.exp compute?
torch.exp returns a new tensor containing the exponential of every element in the input. For each input value xᵢ, the corresponding output is eˣᵢ.
Follow-up: How can the result be written into a preallocated tensor? Pass that tensor with the keyword-only out argument.
QWhat are torch.exp's parameter and return types?
The input parameter is a Tensor, and the function returns a Tensor. The operation is pointwise, so the output has the corresponding elementwise exponential values.
Follow-up: What optional argument is available? The documentation exposes out, an optional Tensor used as the output destination.
QWhat is the function signature?
Its signature is torch.exp(input, *, out=None) → Tensor. The asterisk means out is keyword-only, while input is the tensor being transformed.
Follow-up: What mathematical relationship does the operation implement? Each output element follows yᵢ = eˣᵢ.
Related
torch.exp_torch.Tensor.exptorch.special.exp2torch.special.expm1