pytorch 一个完整的神经网络训练过程

## TODO: Define your model with dropout added
class Classifier(nn.Module):
    def __init__(self):
        super().__init__()
        self.fc1 = nn.Linear(784, 256)
        self.fc2 = nn.Linear(256, 128)
        self.fc3 = nn.Linear(128, 64)
        self.fc4 = nn.Linear(64, 10)
        
        # Dropout module with 0.2 drop probability
        self.dropout = nn.Dropout(p=0.2)
        
    def forward(self, x):
        # make sure input tensor is flattened
        x = x.view(x.shape[0], -1)
        
        # Now with dropout
        x = self.dropout(F.relu(self.fc1(x)))
        x = self.dropout(F.relu(self.fc2(x)))
        x = self.dropout(F.relu(self.fc3(x)))
        
        # output so no dropout here
        x = F.log_softmax(self.fc4(x), dim=1)
        
        return x

## TODO: Train your model with dropout, and monitor the training progress with the validation loss and accuracy
%matplotlib inline
%config InlineBackend.figure_format = 'retina'
import matplotlib.pyplot as plt
import helper

print("#############################开始训练")
model=Classifier()
optimizer = optim.Adam(model.parameters(),lr = 0.001)
criterion = nn.NLLLoss()
epoches = 10
train_losses=[]
test_losses=[]
for e in range(epoches):
    train_loss=0

    for images,labels in trainloader:
        optimizer.zero_grad()       
        imgs_new = images.view(images.shape[0],-1)
        ps=model.forward(imgs_new)
        loss = criterion(ps,labels)
        train_loss+=loss
        loss.backward()
        optimizer.step()       
        
    else:   
        test_loss=0
        accuracy=0
        model.eval()
        with torch.no_grad():
            for images,labels in testloader: 
                imgs_new= images.view(images.shape[0],-1)
                ps=model.forward(imgs_new)
                test_loss += criterion(ps,labels)

                ps2 = torch.exp(ps2)#为了计算准确率用了exp函数
                t_p,t_c = ps.topk(1,dim=1)
                equals = t_c ==labels
                accuracy = torch.mean(equals.type(torch.FloatTensor))
                
            model.train()
            tr=train_loss/len(trainloader)
            train_losses.append(tr)
            te=test_loss/len(testloader)
            test_losses.append(te)
            print("accuracy={}%".format(accuracy*100),"train_loss={}".format(tr),"test_loss={}".format(te))
print("######################################观察训练与校验损失函数走向便于确定过拟合位置")
plt.plot(train_losses, label='Training loss')
plt.plot(test_losses, label='Validation loss')
plt.legend(frameon=False) 

print("##########################实际预测")
images,labels = next(iter(testloader)) 
imgs_new2= images.view(images.shape[0],-1)
im = imgs_new2[0].view(-1,784)
with torch.no_grad():
    ps=torch.exp(model.forward(im))

helper.view_classify(im.view(1,28,28),ps,version="Fashion")

 

#验证结果

#############################开始训练
accuracy=17.1875% train_loss=0.5980074405670166 test_loss=0.45242178440093994
accuracy=15.234375% train_loss=0.43792349100112915 test_loss=0.39830338954925537
accuracy=13.28125% train_loss=0.3959130644798279 test_loss=0.38223353028297424
accuracy=12.5% train_loss=0.37101325392723083 test_loss=0.37162792682647705
accuracy=15.234375% train_loss=0.354133665561676 test_loss=0.3624916672706604
accuracy=14.453125% train_loss=0.34428462386131287 test_loss=0.357414573431015
accuracy=15.625% train_loss=0.3320519030094147 test_loss=0.3513256013393402
accuracy=17.578125% train_loss=0.32158225774765015 test_loss=0.354927122592926
accuracy=14.84375% train_loss=0.31410548090934753 test_loss=0.3455764949321747
accuracy=15.234375% train_loss=0.3062107563018799 test_loss=0.35034212470054626
######################################观察训练与校验损失函数走向便于确定过拟合位置
##########################实际预测

###备注

最后一个参数version存在的意义是给纵轴注明品类，去掉了最后一个参数就不会有了

helper.view_classify(img.view(1, 28, 28), ps, version='Fashion')