.unsqueeze()

The .unsqueeze() method in PyTorch adds a new dimension of size one at the specified position in a tensor. It is commonly used to reshape tensors for operations like broadcasting or to match input dimensions for neural networks.

Syntax

torch.unsqueeze(input, dim)

• input: The input tensor to which a new dimension will be added.
• dim: The index at which to insert the new dimension. This must be a valid dimension index for the tensor, including both positive and negative indices.

The method returns a new tensor with the additional dimension.

Example

This example demonstrates adding a new dimension to a tensor using .unsqueeze():

import torch

# Create a 1D tensor
tensor = torch.tensor([1, 2, 3])

# Add a new dimension at position 0
unsqueezed_tensor_0 = tensor.unsqueeze(0)

# Add a new dimension at position 1
unsqueezed_tensor_1 = tensor.unsqueeze(1)

print("Original Tensor Shape:", tensor.shape)
print("Shape after unsqueeze(0):", unsqueezed_tensor_0.shape)
print("Shape after unsqueeze(1):", unsqueezed_tensor_1.shape)

This example results in the following output:

Original Tensor Shape: torch.Size([3])
Shape after unsqueeze(0): torch.Size([1, 3])
Shape after unsqueeze(1): torch.Size([3, 1])

In this example:

• .unsqueeze(0) adds a new dimension at the start, converting the tensor from shape [3] to [1, 3].
• .unsqueeze(1) adds a new dimension after the first axis, resulting in the shape [3, 1].