.expand()

The .expand() function in PyTorch creates a new view of a tensor by expanding its singleton dimensions (dimensions with size 1) to a larger size. This operation is memory-efficient, as it doesn’t allocate new memory but instead returns a view where the singleton dimensions are expanded to the desired size.

Syntax

tensor.expand(*sizes)

• tensor: The input tensor to be expanded.
• *sizes: The desired expanded size.

The function returns a new view of the original tensor with the specified size. Note that only dimensions of size 1 can be expanded to a larger size.

Example

Here’s an example of how .expand() can be used to expand a tensor’s dimensions:

import torch

# Define a tensor with a singleton dimension
x = torch.tensor([[1], [2], [3]])

# Expand the tensor to size (3, 4)
expanded_x = x.expand(3, 4)

print(expanded_x)

Copy to clipboard
Copy to clipboard

This example results in the following output:

tensor([[1, 1, 1, 1],
        [2, 2, 2, 2],
        [3, 3, 3, 3]])

Copy to clipboard
Copy to clipboard

In this example, .expand() creates a new view of x, with the second dimension expanded from size 1 to size 4. Each element in the original tensor is repeated along the expanded dimension, but no new memory is allocated for these repeats.

Note: Since the expanded tensor is a view, modifying its contents in-place may lead to unexpected behavior. To safely modify the expanded tensor, use .clone() to create a copy before making changes.