A simple neural network model based on MLP full connection layer

Now let's build the simplest neural network , Inspiration and templates come from bilibili One of up Lord Lavita Brother , Relevant video links are as follows ：

Pytorch Advanced tutorial (1)--- Hand to hand MNIST Classification of pictures _ Bili, Bili _bilibili

The following is the simplest neural network classification model that can run through , I will gradually write how to put this simple neural network MLP Model , Change to CNN Neural network model based on convolutional neural network , So as to greatly improve the accuracy , Build a variety of mainstream neural network models, including VGG,Lenet,Alexnet,resnet,transformer wait

The first step is to build the guide package ： Here are some common packages needed to build a neural network , You need to use it api

####################################################
#               MNIST  classification                          #
####################################################

#  Guide pack 
import torch
import torch.nn as nn
import torchvision.datasets as datasets
import torchvision.transforms as transforms
from torch.utils.data import DataLoader
import torch.nn.functional as F

The second step ： Define super parameters ,MLP Type of neural network superparameters often need to be preset .

#  Define super parameters 
num_epochs=3
batch_size=256
learning_rate=0.001

The third step ： load MNIST Data sets .

train_dataset=datasets.MNIST(root='dataset/',train=True,transform=transforms.ToTensor(),download=False)
test_dataset=datasets.MNIST(root='dataset/',train=False,transform=transforms.ToTensor(),download=False)
train_loader=DataLoader(dataset=train_dataset,batch_size=batch_size,shuffle=True)
test_loader=DataLoader(dataset=test_dataset,batch_size=len(test_dataset),shuffle=False)

Step four ： Set up your workout configuration

#  Set up your workout configuration 
cuda = True if torch.cuda.is_available() else False

Step five ： To build the network , With simple MLP The whole connection layer is used for data classification tasks . As shown below, the neural network is composed of three layers , Input layer , Hidden layer , Output layer .

#  Build neural network architecture 
class NeuralNet(nn.Module):
    def __init__(self,num_input,num_hidden,num_class):
        super(NeuralNet,self).__init__()
        self.fc1=nn.Linear(num_input,num_hidden)
        self.fc2=nn.Linear(num_hidden,num_class)
        self.act=nn.ReLU()

    def forward(self,x):
        x = self.fc1(x)
        x = self.act(x)
        x = self.fc2(x)
        return x

Step six ： Initialize schema , Define the loss function and optimizer

#  Initialize schema 
model = NeuralNet(num_input,num_hidden,num_class).cuda()


#  Define the loss function and optimizer 
LossF=nn.CrossEntropyLoss()
optimizer = torch.optim.Adam(model.parameters(),lr=learning_rate)

Step seven , Training ,MINST The dataset is 28*28 Of

# Training 
for epoch in range(num_epochs):
    for batch_index, (images, labels) in enumerate(Train_loader):
        images = images.reshape(-1, 28*28).cuda()
        labels = labels.cuda()
        outputs = model(images)

        #  Calculate the loss 
        loss = LossF(outputs, labels)

        #  Back propagation of gradient 
        optimizer.zero_grad()
        loss.backward()
        optimizer.step()

        if batch_index % 100 == 0:
            print('[{}/{}],[{}/{}],loss={:.4f}'.format(epoch,num_epochs,batch_index,len(Train_loader),loss))

Step eight ： test

# test 
with torch.no_grad():
    correct_num=0
    total_num=0
    for images,labels in Test_loader:
        images=images.reshape(-1,28*28).cuda()
        labels=labels.cuda()
        outputs=model(images)
        _,predictions = torch.max(outputs,1)
        correct_num+=(predictions==labels).sum()
        total_num+=(predictions.size(0))
        print(" The accuracy of the test set is ：{}%".format(correct_num/total_num*100))

Finally, the result diagram is attached

C:\Users\25566\miniconda3\python.exe C:/Users/25566/Desktop/Pytorch-Basics-main/minist classification .py
[0/3],[0/235],loss=2.3095
[0/3],[100/235],loss=0.2861
[0/3],[200/235],loss=0.2120
[1/3],[0/235],loss=0.1810
[1/3],[100/235],loss=0.2953
[1/3],[200/235],loss=0.1477
[2/3],[0/235],loss=0.1991
[2/3],[100/235],loss=0.1295
[2/3],[200/235],loss=0.1428
 The accuracy of the test set is ：95.72000122070312%

Process finished with exit code 0

The above results can be run by copying line by line , Among them, the super parameters can be adjusted manually .