CODEWITHSUNDEEP
torch
Philippe Remy
Philippe Remy

Reputation: 1821

Decreasing one dimension of a tensor in Lua Torch

I have a tensor X of the shape

(2,[...])

where the three dots can be any vector of dimensions (i.e: 4,4)

I would like to transform this tensor to a tensor whose shape is

(1,[...])

In other words, I want to reduce the dimensionality of the first dimension. Of course, I will loose the information indexed by 2 but it does not matter in this case. The problem is not trivial as I don't know the number of dimensions of the tensors. But at least it's greater or equal to 2. A th code is written below:

th> x = torch.Tensor(2, 4, 4):fill(1)
th> y = x[1]
th> z = torch.Tensor()
th> z[1] = y -- bad argument #1 to '?' (empty tensor at /home/ubuntu/torch/pkg/torch/generic/Tensor.c:684)

Do you have any ideas how to do it?

Many thanks in advance

Upvotes: 1

Views: 747

Answers (2)

Alexander Lutsenko
Alexander Lutsenko

Reputation: 2160

This code does what you need:

y = x[{{1, 1}, {}, {}}]

The unknown number of dimensions? Not a problem, you don't have to specify them all:

y = x[{{1, 1}}]

(1,.,.) = 
  1  1  1  1
  1  1  1  1
  1  1  1  1
  1  1  1  1
[torch.DoubleTensor of size 1x4x4]

Upvotes: 1

deltheil
deltheil

Reputation: 16121

I want to reduce the dimensionality of the first dimension

If by reducing you mean narrow down then you can do e.g.:

x = torch.Tensor(2, 4, 4)
x[1]:fill(0)
x[2]:fill(1)

-- this creates tensors of size 1x4x4
-- the first dimension is narrowed, one index at a time
a = x:narrow(1, 1, 1)
b = x:narrow(1, 2, 1)

This gives:

> print(a)
(1,.,.) = 
  0  0  0  0
  0  0  0  0
  0  0  0  0
  0  0  0  0
[torch.DoubleTensor of size 1x4x4]

> print(b)
(1,.,.) = 
  1  1  1  1
  1  1  1  1
  1  1  1  1
  1  1  1  1
[torch.DoubleTensor of size 1x4x4]

Upvotes: 2

Related Questions