.reshape_as()

Anonymous contributor's avatar
Anonymous contributor
Published May 27, 2025
Contribute to Docs

In PyTorch, The .reshape_as() method reshapes a tensor to match the shape of the input tensor. If possible, it returns a view of the original tensor; otherwise, it returns a new tensor with copied data. This method is equivalent to .reshape().

Syntax

Tensor.reshape_as(other)

Parameters:

  • other: The tensor whose shape will be used to reshape the calling tensor.

Return value:

  • Tensor: A new tensor with the same shape as other, containing the same data as the original tensor (either as a view or with copied data if needed).

Example 1: Reshape 1D Tensor to 2D with .reshape_as()

This example demonstrates the usage of the .reshape_as() method:

import torch


# Source tensor
# 1D tensor with values [0, 1, 2, 3, 4, 5]
tensor_a = torch.arange(6)


# Other tensor
# 2D tensor with shpae (2, 3)
tensor_b = torch.zeros(2, 3)


# Reshape tensor_a to the shape of tensor_b
reshaped_tensor = tensor_a.reshape_as(tensor_b)


print("Original tensor_a:", tensor_a)
print("Shape of tensor_a:", tensor_a.shape)
print("Reference tensor_b:", tensor_b)
print("Shape of tensor_b:", tensor_b.shape)
print("Reshaped tensor:", reshaped_tensor)
print("Shape after reshape_as:", reshaped_tensor.shape)

The code will produce this output:

Original tensor_a: tensor([0, 1, 2, 3, 4, 5])
Shape of tensor_a: torch.Size([6])
Reference tensor_b: tensor([[0., 0., 0.],
                             [0., 0., 0.]])
Shape of tensor_b: torch.Size([2, 3])
Reshaped tensor: tensor([[0, 1, 2],
                         [3, 4, 5]])
Shape after reshape_as: torch.Size([2, 3])

This code reshapes the source tensor from a 1D tensor containing numbers 0 to 5 into a 2D tensor with shape (2, 3), preserving the data.

Example 2: Convert 1D Tensor to 4x2 Matrix Using .reshape_as()

import torch


# Create a 1D tensor with 8 elements
tensor_a = torch.arange(8)


# Create a new reference tensor with a different shape (4x2)
tensor_b = torch.zeros(4, 2)


# Reshape tensor_a to match the shape of tensor_b
reshaped_tensor = tensor_a.reshape_as(tensor_b)


print("Original tensor_a:", tensor_a)
print("Shape of tensor_a:", tensor_a.shape)
print("Reference tensor_b:", tensor_b)
print("Shape of tensor_b:", tensor_b.shape)
print("Reshaped tensor:", reshaped_tensor)
print("Shape after reshape_as:", reshaped_tensor.shape)

The code will produce this output:

Original tensor_a: tensor([0, 1, 2, 3, 4, 5, 6, 7])
Shape of tensor_a: torch.Size([8])
Reference tensor_b: tensor([[0., 0.],
                             [0., 0.],
                             [0., 0.],
                             [0., 0.]])
Shape of tensor_b: torch.Size([4, 2])
Reshaped tensor: tensor([[0, 1],
                         [2, 3],
                         [4, 5],
                         [6, 7]])
Shape after reshape_as: torch.Size([4, 2])

In this example, the code reshapes tensor_a from a 1D vector to match tensor_b, a 4-row, 2-column matrix, preserving the original data.

All contributors

Contribute to Docs

Learn PyTorch on Codecademy