How to Create Tensor in PyTorch: Syntax and Examples

In PyTorch, you create a tensor using the torch.tensor() function by passing a list or array of values. You can also create tensors filled with zeros, ones, or random numbers using functions like torch.zeros(), torch.ones(), and torch.rand().

Syntax

The basic syntax to create a tensor in PyTorch is torch.tensor(data, dtype=None, device=None). Here, data is a list, tuple, or NumPy array of values. The optional dtype sets the data type (like float or int), and device specifies if the tensor is on CPU or GPU.

Other common tensor creation functions include:

• torch.zeros(size): creates a tensor filled with zeros.
• torch.ones(size): creates a tensor filled with ones.
• torch.rand(size): creates a tensor with random values between 0 and 1.

python

import torch

# Create tensor from list
tensor = torch.tensor([1, 2, 3], dtype=torch.float32)

# Create tensor of zeros
zeros = torch.zeros((2, 3))

# Create tensor of ones
ones = torch.ones((2, 3))

# Create tensor with random values
random = torch.rand((2, 3))

Example

This example shows how to create different types of tensors and print their values and shapes.

python

import torch

# Tensor from list
tensor = torch.tensor([4, 5, 6], dtype=torch.float32)
print('Tensor:', tensor)
print('Shape:', tensor.shape)

# Tensor of zeros
zeros = torch.zeros((2, 2))
print('Zeros tensor:\n', zeros)

# Tensor of ones
ones = torch.ones((3, 1))
print('Ones tensor:\n', ones)

# Tensor with random values
random = torch.rand((2, 3))
print('Random tensor:\n', random)

Output

Tensor: tensor([4., 5., 6.])
Shape: torch.Size([3])
Zeros tensor:
 tensor([[0., 0.],
        [0., 0.]])
Ones tensor:
 tensor([[1.],
        [1.],
        [1.]])
Random tensor:
 tensor([[0.1234, 0.5678, 0.9101],
        [0.2345, 0.6789, 0.3456]])

Common Pitfalls

Common mistakes when creating tensors include:

• Passing a Python list without specifying dtype can lead to unexpected types.
• Trying to create a tensor from nested lists with inconsistent lengths causes errors.
• Forgetting to move tensors to GPU with to(device) if needed.

Example of wrong and right ways:

python

import torch

# Wrong: inconsistent nested list lengths
try:
    bad_tensor = torch.tensor([[1, 2], [3]])
except Exception as e:
    print('Error:', e)

# Right: consistent nested list lengths
good_tensor = torch.tensor([[1, 2], [3, 4]])
print('Good tensor:', good_tensor)

Output

Error: expected sequence of length 2 at dim 1 (got 1)
Good tensor: tensor([[1, 2],
        [3, 4]])

Quick Reference

Function	Description	Example
torch.tensor(data)	Create tensor from data	torch.tensor([1, 2, 3])
torch.zeros(size)	Tensor filled with zeros	torch.zeros((2, 2))
torch.ones(size)	Tensor filled with ones	torch.ones((3, 1))
torch.rand(size)	Tensor with random values	torch.rand((2, 3))

Key Takeaways

Use torch.tensor() to create tensors from Python lists or arrays.

Specify dtype to control the data type of the tensor.

Use torch.zeros(), torch.ones(), and torch.rand() for common tensor initializations.

Ensure nested lists have consistent lengths to avoid errors.

Move tensors to GPU with .to(device) if needed for faster computation.