在pytorch中手动分配权重、偏差和激活函数

Question

我正在尝试在 pytorch 中使用 nn.module 构建一个神经网络。我想实现自定义权重、偏差和激活函数。 输入值=5，第一层权重= [[0.2, 0.3]]，第二层权重= [[1.5],[2.5]]，第一层偏差= 2，第二层偏差= 3，激活函数 y=x^ 2 输出值应该获得 2220.765625 但我的代码不计算这个值。你能帮我纠正这个代码吗？

import torch
import torch.nn as nn


class NeuralNet(nn.Module):
    def __init__(self):    
        super().__init__()  
        self.fc1 = nn.Linear(in_features=1, out_features=2)
        self.output = nn.Linear(in_features=2, out_features=1)
        self.bias1 = torch.tensor([[2.]])
        self.bias2 = torch.tensor([[3.]])
        

    def act(self, x):
        return x**2

    def forward(self, x):
        x = self.act(self.fc1(x)) + self.bias1 
        x = self.act(self.output(x)) + self.bias2

        return x

    def weights_initialization(self):
        with torch.no_grad():
            self.fc1.weight.copy_(torch.tensor([[0.2, 0.3]]))
            self.output.weight.copy_(torch.tensor([[1.5],
                                                   [2.5]]))
 

net = NeuralNet()
input_data = torch.tensor([[5.]])
output = net(input_data)
print(output)

Answer 1

创建

weights_initialization

对象后，需要调用

net

方法。另外，手动添加

bias

值时，将线性图层内的

False

设置为

bias

。以下代码中还有一些其他修复L

import torch
import torch.nn as nn


class NeuralNet(nn.Module):
    def __init__(self):    
        super().__init__()  
        self.fc1 = nn.Linear(in_features=1, out_features=2, bias=False)
        self.output = nn.Linear(in_features=2, out_features=1, bias=False)
        self.bias1 = torch.tensor([[2.]])
        self.bias2 = torch.tensor([[3.]])
        

    def act(self, x):
        return x**2

    def forward(self, x):
        x = self.act(self.fc1(x) + self.bias1)
        x = self.act(self.output(x) + self.bias2)

        return x

    def weights_initialization(self):
        with torch.no_grad():
            self.fc1.weight.copy_(torch.tensor([[0.2],
                                                [0.3]]))
            self.output.weight.copy_(torch.tensor([[1.5, 2.5]]))
    

net = NeuralNet()
net.weights_initialization()
input_data = torch.tensor([[5.]])
output = net(input_data)
print(output)