设计一个多层感知机mlp网络。1该网络执行如下操作：将输入32×32的灰度图像拉伸为1×1024;将拉伸后的数据传入第一个隐藏层该隐藏层为全连接层包含2048个隐藏单元并使用Sigmoid激活函数;将第一个隐藏层的输出传入第二个隐藏层第二个隐藏层为全连接层包含512个隐藏单元使用ReLU激活函数;将第二个隐藏层的输出传入最后一层最后一层也为全连接层输出20维特征不使用激活函数。2该网络的全连接层权

import torch.nn as nn import torch.nn.functional as F

class MLP(nn.Module): def init(self): super(MLP, self).init() self.linear1 = nn.Linear(32*32, 2048) self.linear2 = nn.Linear(2048, 512) self.linear3 = nn.Linear(512, 20)

def forward(self, x):
    x = x.view(x.size(0), -1)
    x = self.linear1(x)
    x = F.sigmoid(x)
    x = self.linear2(x)
    x = F.relu(x)
    x = self.linear3(x)
    return x

if name == 'main': net = MLP() # initialize weights and biases for name, param in net.named_parameters(): if 'bias' in name: nn.init.constant_(param, 0.0) elif 'weight' in name: nn.init.uniform_(param, 0.0, 1.0) print(f'{name}'s mean: \t {param.mean()}') # test with a random input x = torch.randn(1, 1, 32, 32) out = net(x) print(f'Flatten output shape: \t {x.view(x.size(0), -1).shape}') print(f'Linear output shape: \t {net.linear1(out).shape}') print(f'Sigmoid output shape: \t {net.linear2(F.sigmoid(net.linear1(out))).shape}') print(f'ReLU output shape: \t {net.linear3(F.relu(net.linear2(F.sigmoid(net.linear1(out))))).shape}') print(f'Linear output shape: \t {out.shape}'