Use torch NN builds the simplest neural network framework

As usual , Analyze official documents ：
https://pytorch.org/docs/stable/nn.html#containers

Here is the main analysis Container class .

You can see Module Class is the parent class of all neural network modules , therefore , This paper mainly studies Module Usage of .

When we need to create a neural network model , We need to inherit torch.nn.Module class . And call the constructor of the parent class through the constructor .

The constructor of the parent class sets the trained flag The value is True.

among CONV2D It is a two-dimensional convolution algorithm , Mainly used in image operation . about CONV2D Algorithm , Share a blog here , It's very detailed ：https://blog.csdn.net/qq_34107425/article/details/104117670

and forwards Used in CONV1D It is a one-dimensional convolution operation . about rule function , Its official description is ：

———— The picture quotation knows itself .
Generally speaking , One dimensional convolution nn.Conv1d For text data , Only the width is convoluted , It is not necessary to adjust the height .
Conv1d For data only height perhaps weight To deal with , I personally understand that ,Conv1D It deals with linear data , That is to say tensor Low latitude data in tensor field .

And for RELU function , It means nonlinear transformation .
relu The function is to increase the nonlinear relationship between the layers of the neural network , otherwise , If there is no activation function , There is a simple linear relationship between layers , Each layer is equivalent to matrix multiplication , In this way, how can we complete the complex tasks we need neural network to complete .

This is a nonlinear transformation .
When the operation and transformation of functions are not nonlinear , Then there is the possibility of intellectualization , Otherwise, the execution process of the program is in accordance with the system regulations , There is no nonlinear change .

Neural network is equivalent to a processing module , An output is obtained by convolution and nonlinear transformation of the input data , As a result of intelligent processing .

Given in the official presentation document forwards The function shows that the function of this neural network is to carry out once 1 The convolution of dimensions is followed by nonlinear transformation , And repeat the operation once .

Here is a simple neural network skeleton

from torch import nn
import torch

class network(nn.Module):
    def __init__(self):
        super().__init__()
    def forward(self,input):
        output = input + 1
        return output

net = network()
x = torch.tensor(1.0)

output = net(x)
print(output)

among init Method calls the constructor of the parent class ,forwards The method is to do a right input The data is processed once +1 The operation of , The operation here is linear transformation , But he is still the skeleton of a neural network .