# torch.narrow_copy¶

torch.narrow_copy(input, dim, start, length, *, out=None)

Same as Tensor.narrow() except this returns a copy rather than shared storage. This is primarily for sparse tensors, which do not have a shared-storage narrow method.

Parameters:
• input (Tensor) – the tensor to narrow

• dim (int) – the dimension along which to narrow

• start (int) – index of the element to start the narrowed dimension from. Can be negative, which means indexing from the end of dim

• length (int) – length of the narrowed dimension, must be weakly positive

Keyword Arguments:

out (Tensor, optional) – the output tensor.

Example:

>>> x = torch.tensor([[1, 2, 3], [4, 5, 6], [7, 8, 9]])
>>> torch.narrow_copy(x, 0, 0, 2)
tensor([[ 1,  2,  3],
[ 4,  5,  6]])
>>> torch.narrow_copy(x, 1, 1, 2)
tensor([[ 2,  3],
[ 5,  6],
[ 8,  9]])
>>> s = torch.arange(16).reshape(2, 2, 2, 2).to_sparse(2)
>>> torch.narrow_copy(s, 0, 0, 1)
tensor(indices=tensor([[0, 0],
[0, 1]]),
values=tensor([[[0, 1],
[2, 3]],

[[4, 5],
[6, 7]]]),
size=(1, 2, 2, 2), nnz=2, layout=torch.sparse_coo)


torch.narrow() for a non copy variant