How to Use Reshape in PyTorch: Syntax and Examples

In PyTorch, you use tensor.reshape(new_shape) to change the shape of a tensor without changing its data. The new_shape can be a tuple or list specifying the desired dimensions, and one dimension can be set to -1 to infer its size automatically.

Syntax

The basic syntax to reshape a tensor in PyTorch is:

• tensor.reshape(new_shape): Returns a tensor with the same data but a different shape.
• new_shape: A tuple or list of integers specifying the desired dimensions.
• Use -1 in new_shape to let PyTorch automatically calculate that dimension size.

python

reshaped_tensor = tensor.reshape(new_shape)

Example

This example shows how to reshape a 1D tensor of 6 elements into a 2D tensor with 2 rows and 3 columns.

python

import torch

tensor = torch.arange(6)  # Creates tensor([0, 1, 2, 3, 4, 5])
reshaped = tensor.reshape((2, 3))
print("Original tensor:", tensor)
print("Reshaped tensor (2x3):\n", reshaped)

Output

Original tensor: tensor([0, 1, 2, 3, 4, 5])
Reshaped tensor (2x3):
 tensor([[0, 1, 2],
        [3, 4, 5]])

Common Pitfalls

Common mistakes when using reshape include:

• Trying to reshape to a shape that does not match the total number of elements.
• Forgetting that -1 can only be used once in the new shape.
• Confusing reshape with view; reshape can handle non-contiguous tensors, while view requires contiguous memory.

python

import torch

# Wrong: total elements mismatch
try:
    tensor = torch.arange(6)
    tensor.reshape((4, 2))  # 4*2=8 elements, but tensor has 6
except Exception as e:
    print("Error:", e)

# Right: using -1 to infer dimension
tensor = torch.arange(6)
reshaped = tensor.reshape((2, -1))  # Automatically infers 3 columns
print("Reshaped with -1:", reshaped)

Output

Error: shape '[4, 2]' is invalid for input of size 6
Reshaped with -1: tensor([[0, 1, 2],
        [3, 4, 5]])

Quick Reference

Usage	Description
tensor.reshape(new_shape)	Change tensor shape without changing data
-1 in new_shape	Automatically infer dimension size
new_shape must match total elements	Total elements before and after reshape must be equal
Use reshape over view for non-contiguous tensors	reshape works more generally than view

Key Takeaways

Use tensor.reshape(new_shape) to change tensor shape without altering data.

One dimension in new_shape can be -1 to let PyTorch infer its size automatically.

The total number of elements must remain the same before and after reshape.

reshape works on non-contiguous tensors, unlike view which requires contiguous memory.

Common errors come from mismatched shapes or multiple -1 dimensions.