PyTorch .normal()

In PyTorch, the .normal() function is used to generate a tensor of random numbers from a normal (Gaussian) distribution, given the mean and standard deviation. It is useful in scenarios where there is a need to sample from a normal distribution for tasks such as initializing neural network weights, generating synthetic data, or adding noise.

Syntax

torch.normal(mean, std, *, gen=None, out=None)

• mean: A tensor containing the means of the normal distribution.
• std: A tensor containing the standard deviations of the normal distribution.
• gen (Optional): A pseudorandom number generator for sampling.
• out (Optional): A tensor where the output will be stored in-place.

Example

The following example demonstrates the usage of the .normal() function:

import torch

# Create tensors containing means and standard deviations
mean = torch.tensor([0.1, 0.4, 0.7])
std = torch.tensor([0.2, 0.5, 0.8])

# Generate samples from the normal distribution
res = torch.normal(mean, std)

# Print the resultant tensor
print(res)

Copy to clipboard
Copy to clipboard

The above code produces the following output:

tensor([0.3012, 0.3419, 1.6109])

Copy to clipboard
Copy to clipboard

Note: Since the .normal() function generates a tensor of random numbers from the normal distribution, the output may vary each time the code is run.