How can I add zeroes to the end of a PyTorch tensor based on a modulo operation?

Question

I have a tensor with size: torch.Size([1, 305760])

Since 305760 is not divisible by 400, I want to change the size to: [1, 306000] filling the remaining spaces with 0. How can I do this with PyTorch?

Are you wanting to add zeroes or round up to the nearest thousands place? — David Scott, Oct 14 '20 at 13:28
Both - add enough zeroes so that the size is a multiple of `400` — Shamoon, Oct 14 '20 at 13:37

score 2 · Accepted Answer · answered Oct 14 '20 at 13:45

2

a = torch.randn(1,305760)
N = a.shape[1]
M = 400
b = torch.zeros([a.shape[0], (N // M + 1 ) * 400 - N])
a = torch.cat((a, b), 1)

answered Oct 14 '20 at 13:45

Kris

518
3
13

David Scott · Answer 2 · 2020-10-14T14:02:39.463

1

You'll need to implement a rounding function that lets you specify the base you want to round to. This should work:

def myround(x, base):
    return base * -(-x // base)

torch.Size([1, myround(305760,400)])

Basing this off the following question but modified to force rounding up.

edited Oct 14 '20 at 14:02

answered Oct 14 '20 at 13:54

David Scott

796
2
5
22

How can I add zeroes to the end of a PyTorch tensor based on a modulo operation?

2 Answers2