pytorch深度学习：非线性模型

1. 这篇博客使用深度学习框架搭建了一个预测三次函数的模型

2. 正则化很重要，一定要normalize，否则神经网络就是垃圾

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

class unLinear(nn.Module):
    def __init__(self,input_feature,num_hidden,output_size):
        super(unLinear,self).__init__()
        self.hidden=nn.Linear(input_feature,num_hidden)#一个层就是一个函数
        self.out=nn.Linear(num_hidden,output_size)#可以把层理解成函数的右值引用

    def forward(self,x):
        # x=F.relu(self.hidden(x))
        # x = torch.sigmoid(self.hidden(x))
        x=torch.tanh(self.hidden(x))
        x=self.out(x)
        return x

    def train(self,inputs,target,criterion,optimizer,epoches):
        print(inputs.size())
        print(target.size())
        loss=0
        for epoch in range(epoches):
            output = model.forward(inputs)
            # if epoch%1000==0:
            #     plt.scatter(inputs.detach().numpy(), output.detach().numpy(), c='#00CED1', s=10, alpha=0.8, label="test")
            #     plt.show()
            loss = criterion(output, target)
            optimizer.zero_grad()
            loss.backward()
            optimizer.step()
        return self, loss

model = unLinear(input_feature=1,num_hidden=20,output_size=1)
x=torch.torch.arange(-2,2,0.1)
y=x.pow(3)+0.1*torch.rand(x.size())
# print(x)
# print(y)
plt.scatter(x.detach().numpy(), y.detach().numpy(), c='#00CED1', s=10, alpha=0.8, label="test")
plt.show()

inputs=torch.unsqueeze(x,dim=1)
target=torch.unsqueeze(y,dim=1)
criterion=nn.MSELoss()
optimizer = optim.SGD(model.parameters(), lr=1e-2)

new_model=model.train(inputs=inputs,target=target,criterion=criterion,optimizer=optimizer,epoches=10000)

# plt.scatter(x.numpy(),y.numpy(),c='#00CED1',s=10,alpha=0.8,label="test")
# plt.show()

x_predict=torch.unsqueeze(torch.arange(-2,2,0.05),dim=1)
y_predict=model.forward(x_predict)
# y_predict=model.forward(inputs)
# print(inputs.size())
# print(x_predict.size())
# print(y_predict.detach().numpy())
x_predict=torch.squeeze(x_predict)
y_predict=torch.squeeze(y_predict)
x_predict=x_predict.detach().numpy()
y_predict=y_predict.detach().numpy()
# print(y_predict)
plt.scatter(x_predict,y_predict,s=10,alpha=0.8,label="test")
plt.show()