.randperm()

In PyTorch, the .randperm() function generates a random permutation of integers from 0 to n-1. This function is useful when there is a need to shuffle indices, such as for batch sampling or dataset shuffling.

Syntax

torch.randperm(n, *, generator=None, out=None, dtype=torch.int64, ...)

• n (int): The upper bound (exclusive) for the range of integers.
• generator (Optional): Generates a pseudorandom number for sampling.
• out (Optional): The output tensor to store the result.
• dtype (Optional): The desired data type of the returned tensor.

Note: The ellipsis (…) in the syntax indicates that there can be additional optional parameters beyond those listed here.

Example

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

import torch

# Generate a random permutation of integers from 0 to 9
perm = torch.randperm(10)

# Print the resultant tensor
print(perm)

to clipboard
Copy to clipboard

The above code produces the following output:

tensor([0, 1, 5, 6, 4, 2, 8, 7, 3, 9])

to clipboard
Copy to clipboard

Note: Since the .randperm() function generates a random permutation of integers, the output may vary each time the code is run.