神经网络与深度学习（邱锡鹏）编程练习4 FNN 正向传播 pytorch

NN模型：

 

ref：【人工智能导论：模型与算法】MOOC 8.3 误差后向传播(BP) 例题 【第三版】 - HBU_DAVID - 博客园 (cnblogs.com)

 

实验目标：

理解正向传播过程，熟悉pytorch编程。

初始值：

w1, w2, w3, w4, w5, w6, w7, w8 = 0.2, -0.4, 0.5, 0.6, 0.1, -0.5, -0.3, 0.8

x1, x2 = 0.5, 0.3

y1, y2 = 0.23, -0.07

输出：

 

源代码1

import torch.nn as nn
import torch


def forward_propagate(x, w1, w2):  # 正向传播
    h = nn.Linear(2, 2)  # 输入层，线性（liner）关系
    h.weight.data = w1
    h.bias.data = torch.Tensor([0.0])
    h_out = torch.sigmoid(h(x))

    o = nn.Linear(2, 2)  # 输入层，线性（liner）关系
    o.weight.data = w2
    o.bias.data = torch.Tensor([0.0])
    o_out = torch.sigmoid(o(h_out))

    return o_out


if __name__ == "__main__":
    w1 = torch.Tensor([[0.2, 0.5], [-0.4, 0.6]])
    w2 = torch.Tensor([[0.1, -0.3], [-0.5, 0.8]])

    x = torch.tensor([0.5, 0.3])  # 输入值
    y = torch.tensor([0.23, 0.07])  #
    print("输入值：x1, x2；", x, "\n输出值：y1, y2:", y)

    out_o = forward_propagate(x, w1, w2)
    print("输入值：x1, x2；", x, "\n输出值：y1, y2:", out_o)

源代码2：

整理为 class Net(nn.Module)形式，为将来学习“训练模型”代码做准备

import torch
import torch.nn as nn


class Net(nn.Module):
    # 初始化网络结构
    def __init__(self, input_size, hidden_size, num_classes):
        super(Net, self).__init__()
        print(input_size, hidden_size, num_classes)
        self.fc1 = nn.Linear(input_size, hidden_size)  # 输入层，线性（liner）关系
        self.sigmoid = torch.nn.Sigmoid()  # 隐藏层，使用ReLU函数
        self.fc2 = nn.Linear(hidden_size, num_classes)  # 输出层，线性（liner）关系

    # forword 参数传递函数，网络中数据的流动
    def forward(self, x, w1, w2):
        # print("x:",x)
        self.fc1.weight.data = w1
        self.fc1.bias.data = torch.Tensor([0.0])
        out = self.fc1(x)
        print("h_in:", out)
        # print(self.fc1.weight.data)
        out = self.sigmoid(out)
        print("h_out:", out)
        self.fc2.weight.data = w2
        self.fc2.bias.data = torch.Tensor([0.0])
        out = self.fc2(out)
        print("o_in:", out)
        # print(self.fc2.weight.data)
        out = self.sigmoid(out)
        # print("o_out:", out)
        return out


net = Net(2, 2, 2)

if __name__ == "__main__":
    x = torch.tensor([0.5, 0.3])
    y = torch.tensor([0.23, -0.07])  # y0, y1 = 0.23, -0.07
    w1 = torch.Tensor([[0.2, 0.5], [-0.4, 0.6]])    # 一般情况下，参数无需初始化，让机器自己学习。 有时，初始化参数可提高效率。
    w2 = torch.Tensor([[0.1, -0.3], [-0.5, 0.8]])
    print("输入值：x1, x2；", x, "\n输出值：y1, y2:", y)

    out_o = net(x, w1, w2)
    print("输入值：x1, x2；", x, "\n输出值：y1, y2:", out_o)