Apply Rectified Linear Unit Function Element-wise in PyTorch

To apply a rectified linear unit (ReLU) function element-wise on an input tensor, we use torch.nn.ReLU(). It replaces all the negative elements in the input tensor with 0 (zero), and all the non-negative elements are left unchanged. It supports only real-valued input tensors. ReLU is used as an activation function in neural networks.

Syntax

relu = torch.nn.ReLU()
output = relu(input)

Steps

You could use the following steps to apply rectified linear unit (ReLU) function element-wise −

• Import the required library. In all the following examples, the required Python library is torch. Make sure you have already installed it.

import torch
import torch.nn as nn

• Define input tensor and print it.

input = torch.randn(2,3)
print("Input Tensor:
",input)

• Define a ReLU function relu using torch.nn.ReLU().

relu = torch.nn.ReLU()

• Apply the above-defined ReLU function relu on the input tensor. And optionally assign the output to a new variable

output = relu(input)

• Print the tensor containing ReLU function values.

print("ReLU Tensor:
",output)

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

Example 1

# Import the required library
import torch
import torch.nn as nn
relu = torch.nn.ReLU()
input = torch.tensor([[-1., 8., 1., 13., 9.],
   [ 0., 1., 0., 5., -5.],
   [ 3., -5., 8., -1., 5.],
   [ 0., 3., -1., 13., 12.]])
print("Input Tensor:
",input)
print("Size of Input Tensor:
",input.size())

# Compute the rectified linear unit (ReLU) function element-wise
output = relu(input)
print("ReLU Tensor:
",output)
print("Size of ReLU Tensor:
",output.size())

Output

Input Tensor:
   tensor([[-1., 8., 1., 13., 9.],
      [ 0., 1., 0., 5., -5.],
      [ 3., -5., 8., -1., 5.],
      [ 0., 3., -1., 13., 12.]])
Size of Input Tensor:
   torch.Size([4, 5])
ReLU Tensor:
   tensor([[ 0., 8., 1., 13., 9.],
      [ 0., 1., 0., 5., 0.],
      [ 3., 0., 8., 0., 5.],
      [ 0., 3., 0., 13., 12.]])
Size of ReLU Tensor:
   torch.Size([4, 5])

In the above example, notice that the negative elements in the input tensor are replaced with zero in the output tensor.

Example 2

# Import the required library
import torch
import torch.nn as nn
relu = torch.nn.ReLU(inplace=True)
input = torch.randn(4,5)
print("Input Tensor:
",input)
print("Size of Input Tensor:
",input.size())

# Compute the rectified linear unit (ReLU) function element-wise
output = relu(input)
print("ReLU Tensor:
",output)
print("Size of ReLU Tensor:
",output.size())

Output

Input Tensor:
   tensor([[ 0.4217, 0.4151, 1.3292, -1.3835, -0.0086],
      [-0.7693, -1.7736, -0.3401, -0.7179, -0.0196],
      [ 1.0918, -0.9426, 2.1496, -0.4809, -1.2254],
      [-0.3198, -0.2231, 1.2043, 1.1222, 0.7905]])
Size of Input Tensor:
   torch.Size([4, 5])
ReLU Tensor:
   tensor([[0.4217, 0.4151, 1.3292, 0.0000, 0.0000],
      [0.0000, 0.0000, 0.0000, 0.0000, 0.0000],
      [1.0918, 0.0000, 2.1496, 0.0000, 0.0000],
      [0.0000, 0.0000, 1.2043, 1.1222, 0.7905]])
Size of ReLU Tensor:
   torch.Size([4, 5])