《PyTorch深度学习实践》02.PyTorch实现Logistic Regression(分类问题)

《PyTorch深度学习实践》.PyTorch实现Logistic Regression(分类问题)

一、Regression v.s. Classification

How to map: R -> [0,1] ?

Sigmoid functions

二、Logistic Function

1.Logistic Regression Model

2.Loss function for Binary Classification

BCE损失：

3. Implementation of Logistic Regression

Init:

BCELoss:

4个步骤：

三、Code

import torch
import torch.nn.functional as F

x_data = torch.Tensor([[1.0],[2.0],[3.0]])
y_data = torch.Tensor([[2.0],[4.0],[6.0]])

class LogisticRegressionModel(torch.nn.Model):
    def __init__(self):
        super(LogisticRegressionModel,self.__init__)
        self.liner = torch.nn.liner(1,1)

    def forward(self,x):
        y_pred = F.sigmoid(self.liner(x))
        return y_pred

model = LogisticRegressionModel()

criterion = torch.nn.BCELoss(size_average=False)
optimizer = torch.optim.SGD(model.parameters(),lr=0.01)

for epoch in range(100):
    y_pred = model(x_data)
    loss = criterion(y_pred,y_data)
    print(epoch,loss)

    optimizer.zero_grad()
    loss.backward()
    optimizer.step()


print('w=',model.liner.weight.item())
print('b=',model.liner.bias.item())

x_test = torch.Tensor([4.0])
y_test = model(x_test)
print('y_pred=',y_test.data)