pytorch深度学习：一般分类器

使用的criterion不是MSE而是交叉熵。

numpy.shape，tensor.size()，正确遍历变量。

另外 CrossEntropyLoss的参数真是有够饶人的。

 1 import torch
 2 from torch import nn,optim
 3 import matplotlib.pyplot as plt
 4 
 5 class Classifier(nn.Module):
 6     def __init__(self,input_feature,output_size):
 7         super(Classifier, self).__init__()
 8         self.linear=nn.Linear(input_feature,output_size)
 9         # print(input_feature)
10         # print(output_size)
11 
12     def forward(self,x):
13         # print(x.size())
14         x=self.linear(x)
15         # print(x.size())
16         x=torch.sigmoid(x)
17         # print(x.size())
18         return x
19 
20 
21     def train(self, inp, target, criterion, optimizer, epoches):
22         for epoch in range(epoches):
23             output = self.forward(inp)
24             # print(output.size())
25             # print(target.size())
26             loss = criterion(output, target)
27             optimizer.zero_grad()
28             loss.backward()
29             optimizer.step()
30         return self, loss
31 
32 cluster=torch.ones(100,2)
33 data0=torch.normal(cluster,1)
34 data1=torch.normal(-cluster,1)
35 target0=torch.zeros(100,1)
36 target1=torch.ones(100,1)
37 inputs=torch.cat((data0,data1),dim=0)
38 target=torch.cat((target0,target1),dim=0)
39 print(target.size())
40 target=torch.squeeze(target)
41 print(inputs.size())
42 print(target.size())
43 
44 plt.scatter(inputs.numpy()[:,0],inputs.numpy()[:,1],c=target.numpy()[:,0],s=10,cmap='RdYlGn')
45 plt.show()
46 
47 model=Classifier(2,2)
48 criterion=nn.CrossEntropyLoss()
49 optimizer = optim.SGD(model.parameters(), lr=1e-2)
50 
51 # x=torch.cat((data0,data1),).type(torch.FloatTensor)
52 # y=torch.cat((torch.zeros(100),torch.ones(100)),).type(torch.LongTensor)
53 
54 new_model,loss=model.train(inputs,target.type(torch.LongTensor),criterion,optimizer,100)
55 print(loss)

 

posted @ 2020-09-21 10:24  Lovaer  阅读(443)  评论(0)  编辑  收藏  举报