"Hands on learning in depth" Chapter 2 - preparatory knowledge_ 2.1 data operation_ Learning thinking and exercise answers

---

2.1 Data manipulation

One 、 Summary of key contents

1. Deconstruction of variables

This concept was vague before me , Teacher Li Mu mentioned here , Make a note of ：
Take an example

Y = torch.tensor([[2.0, 1, 4, 3], [1, 2, 3, 4], [4, 3, 2, 1]])
X = torch.arange(12, dtype=torch.float32).reshape((3,4))
before = id(Y)
Y = Y + X
id(Y) == before

Output results ：

False

This example shows that when running some operations , It may cause memory allocation for the new structure .

take Y The address of is recorded in before variable , And then we did Y + X Assigned to a new variable Y The operation of , Although on the surface, it is still Y, But this Y It's not that Y 了 ！Y The address of has changed .

2. Perform in place operation

Continue the above example , We can operate in situ in this way ：

Z = torch.zeros_like(Y)
print('id(Z):', id(Z))
Z[:] = X + Y
print('id(Z):', id(Z))

Output results ：

id(Z): 140040758378960
id(Z): 140040758378960

Construct a and Y Of the same dimension Z,Z All elements of are 0, after Z[:] = X + Y operation , It's equivalent to Z An adaptation of the element of , The address doesn't change , Realize in-situ operation .

Two 、 Exercise answer

1. Run the code in this section . The conditional statements in this section X == Y Change to X < Y or X > Y, Then see what kind of tensor you can get .

X = torch.arange(12, dtype=torch.float32).reshape((3,4))
Y = torch.tensor([[2.0, 1, 4, 3], [1, 2, 3, 4], [4, 3, 2, 1]])
X, Y

Output results ：

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

X > Y

Output results ：

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

2. Use other shapes （ For example, three-dimensional tensors ） Replace two tensors operated by elements in the broadcast mechanism . Whether the result is the same as expected ？

A = torch.tensor([[[1, 2, 3], [4, 5, 6]]])
B = torch.tensor([[[10, 20, 30]], [[40, 50, 60]]])
C = A + B
A.shape,B.shape, C, C.shape

Output results ：

(torch.Size([1, 2, 3]), torch.Size([2, 1, 3]), tensor([[[11, 22, 33],
          [14, 25, 36]],
 
         [[41, 52, 63],
          [44, 55, 66]]]), torch.Size([2, 2, 3]))