lab01-2 Tensor Manipulation II

Veiw (Reshape)

import numpy as np
import torch

t = np.array([[[0, 1, 2],
              [3, 4, 5]],
             [[ 6, 7, 8],
             [9, 10, 11]]])
ft = torch.FloatTensor(t)
print(ft)
print(ft.shape)

tensor([[[ 0.,  1.,  2.],
         [ 3.,  4.,  5.]],

        [[ 6.,  7.,  8.],
         [ 9., 10., 11.]]])
torch.Size([2, 2, 3])

|ft| = (2, 2, 3) -> (2 x 2, 3) -> (4, 3) : reshaped

print(ft.view([-1, 3]))
print(ft.view([-1, 3]).shape)

tensor([[ 0.,  1.,  2.],
        [ 3.,  4.,  5.],
        [ 6.,  7.,  8.],
        [ 9., 10., 11.]])
torch.Size([4, 3])

print(ft.view([-1, 1, 3]))
print(ft.view([-1, 1, 3]).shape)

tensor([[[ 0.,  1.,  2.]],

        [[ 3.,  4.,  5.]],

        [[ 6.,  7.,  8.]],

        [[ 9., 10., 11.]]])
torch.Size([4, 1, 3])

print(ft.view([-1, 4, 3]))
print(ft.view([-1, 4, 3]).shape)

tensor([[[ 0.,  1.,  2.],
         [ 3.,  4.,  5.],
         [ 6.,  7.,  8.],
         [ 9., 10., 11.]]])
torch.Size([1, 4, 3])

Squeeze

Squeeze deletes dimensions with element count of 1.

ft = torch.FloatTensor([[0], [1], [2]])
print(ft)
print(ft.shape)

tensor([[0.],
        [1.],
        [2.]])
torch.Size([3, 1])

print(ft.squeeze())
print(ft.squeeze().shape) # [3, 1] -> [3]
print()
print(ft.squeeze(dim=0))
print(ft.squeeze(dim=0).shape) # dimension 0의 element count가 1이 아니므로 변화가 없다.

tensor([0., 1., 2.])
torch.Size([3])

tensor([[0.],
        [1.],
        [2.]])
torch.Size([3, 1])

Unsqueeze

put the number 1 in to the dimension I want (do the opposite of squeeze)

ft = torch.Tensor([0, 1, 2])
print(ft.shape)

torch.Size([3])

print(ft.unsqueeze(0))
print(ft.unsqueeze(0).shape)

tensor([[0., 1., 2.]])
torch.Size([1, 3])

print(ft.view(1, -1))
print(ft.view(1, -1).shape)

tensor([[0., 1., 2.]])
torch.Size([1, 3])

print(ft.unsqueeze(1))
print(ft.unsqueeze(1).shape)

tensor([[0.],
        [1.],
        [2.]])
torch.Size([3, 1])

Type Casting

lt = torch.LongTensor([1, 2, 3, 4])
print(lt)

tensor([1, 2, 3, 4])

print(lt.float())

tensor([1., 2., 3., 4.])

bt = torch.ByteTensor([True, False, False, True]) # T -> 1 / F -> 0
print(bt)
print()
bt = (lt == 3) # [1, 2, 3, 4] -> [F, F, T, F]
print(bt)

tensor([1, 0, 0, 1], dtype=torch.uint8)

tensor([False, False,  True, False])

print(bt.long())
print(bt.float())

tensor([0, 0, 1, 0])
tensor([0., 0., 1., 0.])

Concatenate

x = torch.FloatTensor([[1, 2], [3, 4]]) # 2 x 2
y = torch.FloatTensor([[5, 6], [7, 8]]) # 2 x 2

print(torch.cat([x, y], dim=0)) # 4 x 2
print(torch.cat([x, y], dim=0).shape)
print(torch.cat([x, y], dim=1)) # 2 x 4
print(torch.cat([x, y], dim=1).shape)

tensor([[1., 2.],
        [3., 4.],
        [5., 6.],
        [7., 8.]])
torch.Size([4, 2])
tensor([[1., 2., 5., 6.],
        [3., 4., 7., 8.]])
torch.Size([2, 4])

Stacing

x = torch.FloatTensor([1, 4]) # 2
y = torch.FloatTensor([2, 5])
z = torch.FloatTensor([3, 6])

print(torch.stack([x, y, z])) # 3 x 2
print(torch.stack([x, y, z], dim=1)) # 2 x 3

tensor([[1., 4.],
        [2., 5.],
        [3., 6.]])
tensor([[1., 2., 3.],
        [4., 5., 6.]])

print(torch.cat([x.unsqueeze(0), y.unsqueeze(0), z.unsqueeze(0)], dim=0))

tensor([[1., 4.],
        [2., 5.],
        [3., 6.]])

Ones and Zeros

x = torch.FloatTensor([[0, 1, 2], [2, 1, 0]])
print(x)

tensor([[0., 1., 2.],
        [2., 1., 0.]])

print(torch.ones_like(x))
print(torch.zeros_like(x))

tensor([[1., 1., 1.],
        [1., 1., 1.]])
tensor([[0., 0., 0.],
        [0., 0., 0.]])

In-place Operation

x = torch.FloatTensor([[1, 2], [3, 4]])

print(x.mul(2.))
print(x)
print(x.mul_(2.)) # memory를 새로 선언하지 않고 기존의 memory주소에 넣기
print(x) # 다시 동일한 변수를 print 했을 시 변경된 값이 출력됨.

tensor([[2., 4.],
        [6., 8.]])
tensor([[1., 2.],
        [3., 4.]])
tensor([[2., 4.],
        [6., 8.]])
tensor([[2., 4.],
        [6., 8.]])