NLP introduction + practice: Chapter 2: introduction to pytorch

Last one ： 《nlp introduction + actual combat ： Chapter one ： Deep learning and neural networks 》
Code link of this chapter ： https://gitee.com/ninesuntec/nlp-entry-practice/blob/master/code/2.pytorch.py

1.pytorch Introduction to

Pytorch Is a facebook Published in-depth learning framework , Because of its ease of use , Friendliness . Favored by the majority of users .

2.pytorch Version of

Please refer to pytorch Official website ：https://pytorch.org/get-started/locally/
There are detailed installation steps above , Of course , You can also base on anaconda Environment to install pytorch Environment

Of course , If your computer doesn't have gpu Words , The simple calculation in the early stage depends on cpu No problem at all , The installation command is as follows ：

3.pytorch To get started with

3.1 tensor Tensor

Tensor is a general term , There are many types :

• 0 Order tensor : Scalar 、 constant ,0-D Tensor
• 1 Order tensor : vector ,1-D Tensor
• 2 Order tensor : matrix ,2-D Tensor
• 3 Order tensor
• 5…
• N Order tensor

With numpy, Sequence , Arrays, etc , Why do we need tensors , In us pytorch In the frame , It only knows tensors ,, So we need to convert the data into tensor type ,pytorch To analyze .

3.2 stay Pytorch Create tensors in

Before learning this section , Please make sure you have installed pytorch as well as numpy
I'm through anaconda Installed , No, you can baidu by yourself , Lots of tutorials .

• 《pycharm Middle configuration anaconda Medium tensorflow Environmental Science 》
• 《anaconda Commonly used instructions 》

1. Use python Create tensors from lists or sequences in

import torch
import numpy as np
t1=torch.tensor([[1, -1], [1, -1]])
print(t1)

This article only shows torch and numpy Import of , The following defaults are added , Please fill in by yourself . At the same time, this article does not give a screenshot of the running results , It aims to ensure that everyone manually types the code , Otherwise, it's a bolt , Not much .

2. Use Numpy Create tensors from arrays in

t2 = torch.tensor(np.array([[1, 2, 3], [4, 5, 6]]))
print(t2)

3. Use torch Medium api establish tensor

t3_0 = torch.empty(3, 4)  #  establish 3 That's ok 4 Empty column tensor, Will use useless data to fill 
t3_1 = torch.ones([3, 4])  #  establish 3 That's ok 4 All are listed as 1 Of tensor
t3_2 = torch.zeros([3, 4])  #  establish 3 That's ok 4 All are listed as 0 Of tensor
t3_3 = torch.rand([3, 4])  #  establish 3 That's ok 4 Column of random values tensor, The random interval is （0,1）
t3_4 = torch.randint(low=0, high=10, size=[3, 4])  #  establish 3 That's ok 4 Columns of random integers tensor, The random interval is （low,high）
t3_5 = torch.randn([3, 4])  #  establish 3 That's ok 4 Columns of random numbers tensor, The mean value of the random value is 0, The variance of 1

3.3 Pytorch Medium tensor Common methods

1. obtain tensor Data in ( When tensor Only one element is available in ):tensor.item()

t4 = torch.tensor(np.arange(1))
print(t4)
print(t4.item())

2. Turn into numpy Array

t5_1 = torch.randint(low=0, high=10, size=[3, 4])
t5_2=t5_1.numpy()
print(t5_1)
print(t5_2)

3. Get shape ：tensor.size()

t6 = torch.randint(low=0, high=10, size=[3, 4])
print(t6)
print(t6.size())

4. Shape change ：tensor.view((3,4)). similar numpy Medium reshape, It's a shallow copy , Just a change in shape

t7 = torch.randint(low=0, high=10, size=[3, 4])
print(t7)
print(t7.view(2, 6))

Be careful , No matter how you convert it, you must ensure that the final product remains unchanged , Such as 3* 4=2*6
5. Get dimension :tensor.dim()

t8 = torch.randint(low=0, high=10, size=[3, 4])
print(t8)
print(t8.dim())

6. Get the maximum :tensor.max()

t9 = torch.randint(low=0, high=10, size=[3, 4])
print(t9)
print(t9.max())

7. Transposition ：tensor.t()

t10 = torch.randint(low=0, high=10, size=[3, 4])
print(t10)
print(t10.t())

8. obtain tensor The first m That's ok n The value of the column ：tensor[m,n]

t11 = torch.randint(low=0, high=10, size=[3, 4])
print(t11)
print(t11[1, 2])  #  Get the second line 3 The value of the column 

9. assignment ：tensor[m,n]=x, Will be the first m That's ok n The value of the column is assigned x

t12 = torch.randint(low=0, high=10, size=[3, 4])
print(t12)
t12[1,2]=1000
print(t12)

10. Two dimensions of the commutative matrix :torch.transpose(x, 0, 1)

t13 = torch.randn(2, 3);  # 2 That's ok 3 Column 
print(t13)
t13=torch.transpose(t13, 0, 1)
print(t13)  #  take t13 Turn into 3 That's ok 2 Column 

If we encounter this type of data , Inside [1,2,3] What does it mean to represent respectively ？
It means that the tensor is a three-dimensional tensor , among

• 1： There is only one piece of data
• 2,3： Represents that every piece of data is a 2 That's ok 3 Columns of data

3.4 tensor Data type of

1. obtain tensor Data type of ：x.dtype

t15 = torch.randn([3, 4])
print(t15.dtype)

2. When creating tensors , Can also pass dtype To specify the data type

t16 = torch.randn([3, 4],dtype=torch.double)
print(t16.dtype)

Of course, we can also pass FloatTensor、LongTensor Wait to create the data type we need

t17 = torch.LongTensor([1, 2])
print(t17)

3.5 tensor Other operations of

1.tensor And tensor The addition of

t18 = t17.new_ones(5, 3, dtype=torch.float)
t18_1 = torch.randn(5, 3)
print(t18)
print(t18_1)
print(t18 + t18_1)
print(torch.add(t18, t18_1))
print(t18.add(t18_1))
t18.add_(t18_1)  #  The underlined method will be correct for t18 Make local modifications 
print(t18)

2.tensor And digital operation

t19 = torch.randn(3, 3)
print("t19：", t19)
print("t19+10=", t19 + 10)

3.CUDA Medium tensor
CUDA (Compute Unified Device Architecture), yes NVIDIA The new computing platform .

CUDATM It's a kind of NVIDIA General parallel computing architecture , The architecture enables GPU Able to solve complex computing problems .

torch.cuda This module adds a pair of CUDA tensor Support for , In the cpu and gpu Use the same method on tensor

adopt .to Method can put a tensor Transfer to another device ( For instance from CPU go to GPU)

if torch.cuda.is_available():
    device = torch.device("cuda")  # cuda device object 
    y = torch.ones_like(t19, device=device)  #  Create a cuda Of tensor
    x = t19.to(device)  #  How to use t19 Turn into cuda Of tensor
    z = x + y
    print(z)
    print(z.to("cpu", torch.double))  # .to Method can also set the type 
else:
    print(" Your device does not support gpu operation ")