Create Tensors with Gradients in PyTorch

To create a tensor with gradients, we use an extra parameter "requires_grad = True" while creating a tensor.

• requires_grad is a flag that controls whether a tensor requires a gradient or not.

• Only floating point and complex dtype tensors can require gradients.

• If requires_grad is false, then the tensor is same as the tensor without the requires_grad parameter.

Syntax

torch.tensor(value, requires_grad = True)

Parameters

• value – tensor data, user-defined or randomly generated.

• requires_grad – a flag, if True, the tensor is included in the gradient computation.

Output

It returns a tensor with requires_grad as True.

Steps

• Import the required library. The required library is torch.

• Define a tensor with requires_grad = True

• Display the created tensor with gradients.

Let's have a couple of examples for a better understanding of how it works.

Example 1

In the following example, we created two tensors. One tensor is without requires_grad = True and the other is with requires_grad = True.

# import torch library
import torch

# create a tensor without gradient
tensor1 = torch.tensor([1.,2.,3.])

# create another tensor with gradient
tensor2 = torch.tensor([1.,2.,3.], requires_grad = True)

# print the created tensors
print("Tensor 1:", tensor1)
print("Tensor 2:", tensor2)

Output

Tensor 1: tensor([1., 2., 3.])
Tensor 2: tensor([1., 2., 3.], requires_grad=True)

Example 2

# import required library
import torch

# create a tensor without gradient
tensor1 = torch.randn(2,2)

# create another tensor with gradient
tensor2 = torch.randn(2,2, requires_grad = True)

# print the created tensors
print("Tensor 1:
", tensor1)
print("Tensor 2:
", tensor2)

Output

Tensor 1:
 tensor([[-0.9223, 0.1166],
    [ 1.6904, 0.6709]])
Tensor 2:
 tensor([[ 1.1912, -0.1402],
    [-0.2098, 0.1481]], requires_grad=True)

Example 3

In the following example, we created a tensor with gradients using numpy array.

# import the required libraries
import torch
import numpy as np

# create a tensor of random numbers with gradients
# generate 2x2 numpy array of random numbers
v = np.random.randn(2,2)

# create a tensor with above random numpy array
tensor1 = torch.tensor(v, requires_grad = True)

# print above created tensor
print(tensor1)

Output

tensor([[ 0.7128, 0.8310],
   [ 1.6389, -0.3444]], dtype=torch.float64,
requires_grad=True)