pytorch搭建简单网络

pytorch搭建一个简单神经网络

import torch
import torch.nn as nn

# 定义数据
# x:输入数据
# y:标签
x = torch.Tensor([[0.2, 0.4], [0.2, 0.3], [0.3, 0.4]])
y = torch.Tensor([[0.6], [0.5], [0.7]])


class MyNet(nn.Module):
    def __init__(self):
        # 调用基类构造函数
        super(MyNet, self).__init__()
        # 容器，使用时顺序调用各个层
        self.fc = nn.Sequential(
            # 定义三层
            # 输入层
            nn.Linear(2, 4),
            # 激活函数
            nn.Sigmoid(),
            # 隐藏层
            nn.Linear(4, 4),
            nn.Sigmoid(),
            # 输出层
            nn.Linear(4, 1),
        )
        # 优化器
        # params:优化对象
        # lr:学习率
        self.opt = torch.optim.Adam(params=self.parameters(), lr=0.001)
        # 损失函数,均方差
        self.mls = torch.nn.MSELoss()

    def forward(self, inputs):
        # 前向传播
        return self.fc(inputs)

    def train(self, x, y):
        # 训练
        # 得到输出结果
        out = self.forward(x)
        # 计算误差
        loss = self.mls(out, y)
        # print('loss', loss)
        # 梯度置零
        self.opt.zero_grad()
        # 误差反向传播
        loss.backward()
        # 更新权重
        self.opt.step()

    def test(self, x):
        # 测试，就是前向传播的过程
        return self.forward(x)


net = MyNet()
for i in range(10000):
    net.train(x, y)
x = torch.Tensor([[0.4, 0.1]])
out = net.test(x)
print(out)  # 输出结果 tensor([[0.5205]], grad_fn=<AddmmBackward>)

训练集较少，可能结果不是很好，主要是结构，毕竟刚开始接触这个pytorch