Pytorch Foundation (1)

Basic data types and tensor

 

Log in and copy

      
      

       
       1 import torch
       
       
 2 import numpy as np
       
       
 3 
       
       
 4 #array  and  tensor Transformation 
       
       
 5 array = np.array([1.1,2,3])
       
       
 6 tensorArray = torch.from_numpy(array) #array The object becomes tensor object 
       
       
 7 array1  = tensorArray.numpy()#tensor The object becomes array object 
       
       
 8 print(array,'\t', tensorArray, '\t', array1 )
       
       
 9 
       
       
10 #torch Have and numpy The same ability to process data 
       
       
11 print(torch.sin(tensorArray))
       
       
12 print(np.ones([2,5]))# Two lines and five columns 
       
       
13 print(np.ones(2))# Two numbers on a line 
       
       
14 a = torch.randn(2, 3)# A normal distribution of two rows and three columns 
       
       
15 print(a)
       
       
16 print(a.size(0),a.size(1),a.shape[1])#2,3,3   0 On behalf of the line ,1 Represents the corresponding number of columns 
       
       
17 print(a.shape)#torch.Size([2,3])
       
       
18 print(a.type())#torch.FloatTensor
       
       
19 isinstance(a, torch.DoubleTensor)#false
       
       
20 isinstance(a, torch.FloatTensor)#true
       
       
21 a1 = a.cuda()
       
       
22 print(isinstance(a1,torch.FloatTensor))#false
       
       
23 print(isinstance(a1,torch.cuda.FloatTensor))#true,#torch The data inside is different from torch.cuda The data in it 
       
       
24 
       
       
25 #torch Of tensor object 
       
       
26 tensor1 = torch.tensor(1)
       
       
27 print(tensor1)#tensor(1)
       
       
28 tensor2 = torch.tensor(1.2)
       
       
29 print(tensor2)#tensor(1.2000)
       
       
30 print(tensor2.shape)#torch.Size([])
       
       
31 print(len(tensor2.shape))#0, When tensor When it's just a number , His dimension is 0, So his size yes [],shape by 0
       
       
32 tensor3 = torch.tensor([1.1])# One dimensional list , So the output dimension is 1
       
       
33 print(tensor3,tensor3.shape)#tensor([1.1000]) torch.Size([1])
       
       
34 tensor4 = torch.FloatTensor(1)# Note that this time 1 Represents a random return FloatTensor object 
       
       
35 print(tensor4)#tensor([1.1000])
       
       
36 tensor5 = torch.FloatTensor(3)# Think about it tensor and FloatTensor The difference between 
       
       
37 print(tensor5)#tensor([0.0000e+00, 0.0000e+00, 6.8645e+36])
      
      
      
      

       
       
1.
       
       
2.
       
       
3.
       
       
4.
       
       
5.
       
       
6.
       
       
7.
       
       
8.
       
       
9.
       
       
10.
       
       
11.
       
       
12.
       
       
13.
       
       
14.
       
       
15.
       
       
16.
       
       
17.
       
       
18.
       
       
19.
       
       
20.
       
       
21.
       
       
22.
       
       
23.
       
       
24.
       
       
25.
       
       
26.
       
       
27.
       
       
28.
       
       
29.
       
       
30.
       
       
31.
       
       
32.
       
       
33.
       
       
34.
       
       
35.
       
       
36.
       
       
37.
      
      

 

section

 

Log in and copy

      
      

       
       1 import torch
       
       
 2 import numpy as np
       
       
 3 
       
       
 4 #tensor And random numbers 
       
       
 5 a = torch.rand(2, 3, 28, 28)
       
       
 6 print(a, a.shape)# Randomly generate one 2*3*28*28 Four dimensional matrix of （ It can be seen as declaring a four-dimensional matrix ）,torch.Size([2, 3, 28, 28])
       
       
 7                   # 4D is suitable for CNN , Three dimensional fit RNN, Two dimensional fit batch
       
       
 8 print(a.numel())#4707, Count the number of elements 
       
       
 9 print(a.dim())#4
       
       
10 print(torch.tensor(1).dim())#0,
       
       
11 print(torch.empty(1))# One dimensional number 0,tensor([0.])
       
       
12 print(torch.Tensor(2,3).type())# The default is torch.FloatTensor
       
       
13 print(torch.IntTensor(2,3))#2*3
       
       
14 print(torch.tensor([1,1]).type())#torch.LongTensor
       
       
15 print(torch.tensor([1.2,1]).type())#torch.FloatTensor
       
       
16 print(torch.rand(3,3))# The value range is 0 To 1 Between 
       
       
17 print(torch.rand_like(torch.rand(3,3)))#rand_like Rows and columns that directly inherit parameters , Generate 3*3 Of 0 To 1 Between random matrices 
       
       
18 print(torch.randint(1,10,(3,3)))# The value is 1 To 10 Between （ Left closed right away ） The size is 3*3 Matrix 
       
       
19 print(torch.randn(3,3))#3*3 matrix , To obey the mean is 0, The variance of 1 Is a normal distribution 
       
       
20 print(torch.normal(mean=torch.full([10],0),std = torch.arange(1,0,-0.1)))# The mean for 0 Decreasing variance 10*1 One dimensional matrix 
       
       
21 print(torch.full([2,3],7))#2*3 All for 7 Two dimensional matrix of 
       
       
22 print(torch.full([],7))# Numbers 7 dimension 0
       
       
23 print([1],7)# A one-dimensional 1*1 The matrix element is 7
       
       
24 print(torch.logspace(0,1,steps=10))#log(10^0) To log(10^1) Take... In the middle 10 Number 
       
       
25 
       
       
26 # section 
       
       
27 a = torch.rand(4,3,28,28)
       
       
28 print(a[0].shape)#torch.Size([3,28,28])
       
       
29 print(a[0,0].shape)#torch.Size([28, 28])
       
       
30 print(a[0,0,2,4])#tensor(0.6186)
       
       
31 print(a[:2].shape)#torch.Size([2, 3, 28, 28])
       
       
32 print(a[:2,1:,:,:].shape)#torch.Size([2, 2, 28, 28])
       
       
33 print(a[:2,-1:,:,:].shape)#torch.Size([2, 1, 28, 28])
       
       
34 
       
       
35 print(a[:,:,0:28:2,0:28:2].shape)#torch.Size([4, 3, 14, 14])
       
       
36 print(a[:,:,::2,::2].shape)#torch.Size([4, 3, 14, 14])
       
       
37 print(a.index_select(1,torch.arange(1)).shape)#torch.Size([4, 1, 28, 28])
       
       
38 
       
       
39 x = torch.randn(3,4)
       
       
40 mask = x.ge(0.5)# Than 0.5 The large ones are marked as true
       
       
41 print(mask)
       
       
42 torch.masked_select(x,mask)# Put true The choice of 
       
       
43 torch.masked_select(x,mask).shape
       
       
44 
       
       
45 src =torch.tensor([[4,3,5],[6,7,8]])
       
       
46 taa = torch.take(src, torch.tensor([0,3,5]))# Select data according to position after flattening 
       
       
47 print(taa)
      
      
      
      

       
       
1.
       
       
2.
       
       
3.
       
       
4.
       
       
5.
       
       
6.
       
       
7.
       
       
8.
       
       
9.
       
       
10.
       
       
11.
       
       
12.
       
       
13.
       
       
14.
       
       
15.
       
       
16.
       
       
17.
       
       
18.
       
       
19.
       
       
20.
       
       
21.
       
       
22.
       
       
23.
       
       
24.
       
       
25.
       
       
26.
       
       
27.
       
       
28.
       
       
29.
       
       
30.
       
       
31.
       
       
32.
       
       
33.
       
       
34.
       
       
35.
       
       
36.
       
       
37.
       
       
38.
       
       
39.
       
       
40.
       
       
41.
       
       
42.
       
       
43.
       
       
44.
       
       
45.
       
       
46.
       
       
47.
      
      

 

 

Dimensional transformation

 

Log in and copy

      
      

       
       1 import torch
       
       
 2 import numpy as np
       
       
 3 # Dimensional change 
       
       
 4 #View reshape
       
       
 5 a = torch.rand(4,1,28,28)# Four pictures , The number of channels is 1, Length and width are 28*28
       
       
 6 print(a.shape)#torch.Size([4, 1, 28, 28])
       
       
 7 print(a.view(4,1*28*28).shape)#torch.Size([4, 784]), Expand the last three dimensions into a line 
       
       
 8 print(a.view(4*28,28).shape)#torch.Size([112, 28]) become 112 That's ok 28 Two dimensional data for Columns 
       
       
 9 print(a.view(4*1,28,28).shape)#torch.Size([4, 28, 28]) To understand the physical meaning of the corresponding picture 
       
       
10 b = a.view(4,784)
       
       
11 print(b.view(4,28,28,1).shape)#torch.Size([4, 28, 28, 1]),b The data that becomes is not a( Be sure to pay attention to )
       
       
12 #print(a.view(4,783))# If the size is inconsistent, an error will be reported 
       
       
13 
       
       
14 #unsqueeze, Increase the dimension , But it will not affect the change of data 
       
       
15 # The range of data is [-a.dim()-1,a.dim()+1)
       
       
16 print()# The following example is [-5,5)
       
       
17 print(a.unsqueeze(0).shape)#torch.Size([1, 4, 1, 28, 28])
       
       
18 print(a.unsqueeze(-1).shape)#torch.Size([4, 1, 28, 28, 1])
       
       
19 print(a.unsqueeze(4).shape)#torch.Size([4, 1, 28, 28, 1])
       
       
20 print(a.unsqueeze(-4).shape)#torch.Size([4, 1, 1, 28, 28])
       
       
21 print(a.unsqueeze(-5).shape)#torch.Size([1, 4, 1, 28, 28])
       
       
22 #print(a.unsqueeze(5).shape)
       
       
23 a = torch.tensor([1.2,2.3])#a Of shape yes [2]
       
       
24 print(a.unsqueeze(-1))#tensor([[1.2000],
       
       
25                         #[2.3000]]) become 2 Row by column 
       
       
26 print(a.unsqueeze(0))#tensor([[1.2000, 2.3000]])#shape become [1,2], That is, one row and two columns 
       
       
27 b = torch.rand(32)
       
       
28 f = torch.rand(4,3,14,14)
       
       
29 b = b.unsqueeze(1).unsqueeze(2).unsqueeze(0)#torch.Size([1, 32, 1, 1])
       
       
30 print(b.shape)
       
       
31 
       
       
32 # Dimension reduction 
       
       
33 print()
       
       
34 print(b.shape)#torch.Size([1, 32, 1, 1])
       
       
35 print(b.squeeze().shape)#torch.Size([32]), All for 1 Of being squeezed 
       
       
36 print(b.squeeze(-1).shape)#torch.Size([1, 32, 1])
       
       
37 print(b.squeeze(0).shape)#torch.Size([32, 1, 1])
       
       
38 print(b.squeeze(1).shape)#torch.Size([1, 32, 1, 1]), Because it doesn't mean 1  So it wasn't squeezed 
       
       
39 print(b.squeeze(-4).shape)#torch.Size([32, 1, 1])
       
       
40 
       
       
41 #expand Extended data , Copy data , But it will not actively replicate data , It will only be copied when necessary , Recommended 
       
       
42 print()
       
       
43 print(b.shape)#torch.Size([1, 32, 1, 1])
       
       
44 print(b.expand(4,32,14,14).shape)#torch.Size([4, 32, 14, 14]), Only for dimensions that are 1  To expand 
       
       
45 print(b.expand(-1,32,-1,-1).shape)#torch.Size([1, 32, 1, 1]), Other dimensions are -1, In this way, you can expand dimensions of the same size when the original dimension is not one 
       
       
46 print(b.expand(-1,32,-1,-4).shape)#torch.Size([1, 32, 1, -4]) -4 It's meaningless 
       
       
47 
       
       
48 #repeat Indicates the number of copies on the original dimension , Not how much , This method applies for new space , Use more space 
       
       
49 print()
       
       
50 print(b.shape)#torch.Size([1, 32, 1, 1])
       
       
51 print(b.repeat(4,32,1,1).shape)#torch.Size([4, 1024, 1, 1]), The second dimension indicates that the copy is willing to come 32 times 
       
       
52 print(b.repeat(4,1,1,1).shape)#torch.Size([4, 32, 1, 1])
       
       
53 print(b.repeat(4,1,32,32).shape)#torch.Size([4, 32, 32, 32])
       
       
54 
       
       
55 #transpose Realize the exchange between specified dimensions 
       
       
56 a = torch.rand(4,3,32,32)
       
       
57 print(a.shape)#torch.Size([4, 3, 32, 32])
       
       
58 a1 = a.transpose(1,3).contiguous().view(4,3*32*32).view(4,32,32,3).transpose(1,3)
       
       
59 print(a1.shape)#torch.Size([4, 3, 32, 32])
       
       
60 print(torch.all(torch.eq(a,a1)))#tensor(True)
       
       
61 
       
       
62 #premute Realize the exchange from the specified dimension location to the specified location 
       
       
63 print(a.permute(0,2,3,1).shape)#torch.Size([4, 32, 32, 3])
      
      
      
      

       
       
1.
       
       
2.
       
       
3.
       
       
4.
       
       
5.
       
       
6.
       
       
7.
       
       
8.
       
       
9.
       
       
10.
       
       
11.
       
       
12.
       
       
13.
       
       
14.
       
       
15.
       
       
16.
       
       
17.
       
       
18.
       
       
19.
       
       
20.
       
       
21.
       
       
22.
       
       
23.
       
       
24.
       
       
25.
       
       
26.
       
       
27.
       
       
28.
       
       
29.
       
       
30.
       
       
31.
       
       
32.
       
       
33.
       
       
34.
       
       
35.
       
       
36.
       
       
37.
       
       
38.
       
       
39.
       
       
40.
       
       
41.
       
       
42.
       
       
43.
       
       
44.
       
       
45.
       
       
46.
       
       
47.
       
       
48.
       
       
49.
       
       
50.
       
       
51.
       
       
52.
       
       
53.
       
       
54.
       
       
55.
       
       
56.
       
       
57.
       
       
58.
       
       
59.
       
       
60.
       
       
61.
       
       
62.
       
       
63.
      
      

 

author ： Your Rego

The copyright of this article belongs to the author , Welcome to reprint , But without the author's consent, the original link must be given on the article page , Otherwise, the right to pursue legal responsibility is reserved .