torchrelay.itertools

Same as the itertools python package, but in pytorch.

Overview

product	Cartesian product of a set.
permutations	All possible r-length permutations of a set.
invert_permutation	Return the inverse of a permutation.

product

product(*inputs, r=1, **backend)

Cartesian product of a set.

Parameters:

Name	Type	Description	Default
inputs	iterable of tensor_like	Input sets (tensors are flattened is not vectors) with n[i] elements each.	()
r	int	Repeats. keyword argument only	1

Returns:

Name	Type	Description
output	(prod(n)**r, r) tensor	Cartesian product

permutations

permutations(input, r=None)

All possible r-length permutations of a set.

Note

This function loops over the number of input elements, It can therefore be slow if this number is large.

Parameters:

Name	Type	Description	Default
input	tensor_like	Input vector (tensor is flattened is not a vector) with n elements.	required
r	int	Length of the permutation vectors	n

Returns:

Name	Type	Description
output	(k, r) tensor	All possible r-length permutations of input. k = n! * (n-r)!

invert_permutation

invert_permutation(perm)

Return the inverse of a permutation

Parameters:

Name	Type	Description	Default
perm	(..., N) tensor_like	Permutations. A permutation is a shuffled set of indices.	required

Returns:

Name	Type	Description
iperm	(..., N) tensor	Inverse permutation.

Examples:

perm = [0, 2, 3, 1]
a = torch.rand((len(perm),))
permuted_a = a[perm]
recovered_a = permuted_a[invert_permutation(perm)]
assert((a == recovered_a).all())