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。

import torch
from torch import nn,optim
import matplotlib.pyplot as plt

class Classifier(nn.Module):
    def __init__(self,input_feature,output_size):
        super(Classifier, self).__init__()
        self.linear=nn.Linear(input_feature,output_size)
        # print(input_feature)
        # print(output_size)

    def forward(self,x):
        # print(x.size())
        x=self.linear(x)
        # print(x.size())
        x=torch.sigmoid(x)
        # print(x.size())
        return x


    def train(self, inp, target, criterion, optimizer, epoches):
        for epoch in range(epoches):
            output = self.forward(inp)
            # print(output.size())
            # print(target.size())
            loss = criterion(output, target)
            optimizer.zero_grad()
            loss.backward()
            optimizer.step()
        return self, loss

cluster=torch.ones(100,2)
data0=torch.normal(cluster,1)
data1=torch.normal(-cluster,1)
target0=torch.zeros(100,1)
target1=torch.ones(100,1)
inputs=torch.cat((data0,data1),dim=0)
target=torch.cat((target0,target1),dim=0)
print(target.size())
target=torch.squeeze(target)
print(inputs.size())
print(target.size())

plt.scatter(inputs.numpy()[:,0],inputs.numpy()[:,1],c=target.numpy()[:,0],s=10,cmap='RdYlGn')
plt.show()

model=Classifier(2,2)
criterion=nn.CrossEntropyLoss()
optimizer = optim.SGD(model.parameters(), lr=1e-2)

# x=torch.cat((data0,data1),).type(torch.FloatTensor)
# y=torch.cat((torch.zeros(100),torch.ones(100)),).type(torch.LongTensor)

new_model,loss=model.train(inputs,target.type(torch.LongTensor),criterion,optimizer,100)
print(loss)