1.PyTorch基础实现代码
1 import torch 2 from torch.autograd import Variable 3 4 torch.manual_seed(2) 5 x_data = Variable(torch.Tensor([[1.0], [2.0], [3.0], [4.0]])) 6 y_data = Variable(torch.Tensor([[0.0], [0.0], [1.0], [1.0]])) 7 8 #初始化 9 w = Variable(torch.Tensor([-1]), requires_grad=True) 10 b = Variable(torch.Tensor([0]), requires_grad=True) 11 epochs = 100 12 costs = [] 13 lr = 0.1 14 print("before training, predict of x = 1.5 is:") 15 print("y_pred = ", float(w.data*1.5 + b.data > 0)) 16 17 #模型训练 18 for epoch in range(epochs): 19 #计算梯度 20 A = 1/(1+torch.exp(-(w*x_data+b))) #逻辑回归函数 21 J = -torch.mean(y_data*torch.log(A) + (1-y_data)*torch.log(1-A)) #逻辑回归损失函数 22 #J = -torch.mean(y_data*torch.log(A) + (1-y_data)*torch.log(1-A)) +alpha*w**2 23 #基础类进行正则化，加上L2范数 24 costs.append(J.data) 25 J.backward() #自动反向传播 26 27 #参数更新 28 w.data = w.data - lr*w.grad.data 29 w.grad.data.zero_() 30 b.data = b.data - lr*b.grad.data 31 b.grad.data.zero_() 32 33 print("after training, predict of x = 1.5 is:") 34 print("y_pred =", float(w.data*1.5+b.data > 0)) 35 print(w.data, b.data)
2.用PyTorch类实现Logistic regression,torch.nn.module写网络结构
1 import torch 2 from torch.autograd import Variable 3 4 x_data = Variable(torch.Tensor([[0.6], [1.0], [3.5], [4.0]])) 5 y_data = Variable(torch.Tensor([[0.], [0.], [1.], [1.]])) 6 7 class Model(torch.nn.Module): 8 def __init__(self): 9 super(Model, self).__init__() 10 self.linear = torch.nn.Linear(1, 1) 11 self.sigmoid = torch.nn.Sigmoid() ###### **sigmoid** 12 13 def forward(self, x): 14 y_pred = self.sigmoid(self.linear(x)) 15 return y_pred 16 17 18 model = Model() 19 20 21 criterion = torch.nn.BCELoss(size_average=True) #损失函数 22 optimizer = torch.optim.SGD(model.parameters(), lr=0.01) # 随机梯度下降 23 24 25 for epoch in range(500): 26 # Forward pass 27 y_pred = model(x_data) 28 29 30 loss = criterion(y_pred, y_data) 31 if epoch % 20 == 0: 32 print(epoch, loss.item()) 33 34 #梯度归零 35 optimizer.zero_grad() 36 # 反向传播 37 loss.backward() 38 # update weights 39 optimizer.step() 40 41 hour_var = Variable(torch.Tensor([[0.5]])) 42 print("predict (after training)", 0.5, model.forward(hour_var).data[0][0]) 43 hour_var = Variable(torch.Tensor([[7.0]])) 44 print("predict (after training)", 7.0, model.forward(hour_var).data[0][0])
参考：https://blog.csdn.net/ZZQsAI/article/details/90216593