VAE网络结构较AE只有部分改变

import torch
import numpy as np
from torch import nn




class VAE(nn.Module):

    def __init__(self):
        super(VAE, self).__init__()

        # [b,784] => [b,20]
        # u: [b,10]
        # sigma: [b,10]
        self.encoder = nn.Sequential(
            nn.Linear(784,256),
            nn.ReLU(),
            nn.Linear(256,64),
            nn.ReLU(),
            nn.Linear(64,20),
            nn.ReLU()
        )
        # [b,20] => [b,784]
        self.decoder = nn.Sequential(
            nn.Linear(10,64),
            nn.ReLU(),
            nn.Linear(64,256),
            nn.ReLU(),
            nn.Linear(256,784),
            nn.Sigmoid()  # 压缩到0-1
        )


    def forward(self,x):
        """

        :param x:[b,1,28,28]
        :return:
        """
        batchsz = x.size(0)
        # flatten
        x=x.view(batchsz,784)
        # encoder
        # [b,20] 包含了mean 和 sigma
        h_ = self.encoder(x)

        # VAE 和 AE 的不同之处
        # 把mu和sigma拆分出来，用chunk(拆分的个数，位置）
        # [b,20]-——>[b,10] and [b,10]
        mu,sigma = h_.chunk(2,dim=1)
        # reparametrize trick ，epison~N（0,1）
        h = mu + sigma * torch.randn_like(sigma)  # 后边这个是sigma的正态分布
        # decoder
        x_hat = self.decoder(h)
        # reshape  因为是打平过的，还需要再变回照片
        x_hat = x_hat.view(batchsz,1,28,28)

        # 计算KL divergence，网上可以查它的公式,这里u2=0，sigma2=1
        kld = 0.5 * torch.sum(
            torch.pow(mu,2)+
            torch.pow(sigma,2)-
            torch.log(1e-8 + torch.pow(sigma,2))-1
        ) / (batchsz*28*28)





        return x_hat , kld

只是多个这个kld

主函数部分变化有

import torch
from torchvision import transforms,datasets  # datasets自带数据集MNIST
from torch.utils.data import DataLoader
from AE import AE
from VAE import VAE
from torch import nn,optim
import visdom





def main():
    # 把MNIST数据集加载进来
    mnist_train = datasets.MNIST('mnist',True,transform=transforms.Compose([
        transforms.ToTensor()
    ]),download=True)

    # 把数据集加载到DataLoader中
    mnist_train = DataLoader(mnist_train,batch_size=32,shuffle=True)


    # 把MNIST数据集加载进来
    mnist_test = datasets.MNIST('mnist',True,transform=transforms.Compose([
        transforms.ToTensor()
    ]),download=True)

    # 把数据集加载到DataLoader中
    mnist_test = DataLoader(mnist_test,batch_size=32,shuffle=True)

    # 构建一个迭代器
    x,_ = iter(mnist_train).next()  # 不返回label，因为这是无监督学习
    print('x:',x.shape)  # x:torch.Size([32, 1, 28, 28])

    device = torch.device('cuda')
    model = VAE().to(device)
    criteon = nn.MSELoss()
    optimizer = optim.Adam(model.parameters(),lr=1e-3)
    print(model)

    viz = visdom.Visdom()


    for epoch in range(1000):
        for batchidx,(x,_) in enumerate(mnist_train):
            # [b,1,28,28]
            x=x.to(device)

            # x_hat表示重建过的x
            x_hat,kld = model(x)
            loss = criteon(x_hat,x)

            # VAE才有的
            if kld is not None:
                elbo = -loss - 1.0 * kld
                loss = -elbo

            # backprop
            optimizer.zero_grad() #第一步梯度清零
            loss.backward()  # 第二步backward
            optimizer.step()  # 第三步更新梯度


        print(epoch,'loss',loss.item(),'kld',kld.item())

        # 从test中取一些图片进行重构
        x, _ = iter(mnist_test).next()
        x=x.to(device)
        with torch.no_grad():
            x_hat,kld = model(x)
        # 可视化
        viz.images(x,nrow=8,win='x',opts=dict(title='x'))
        viz.images(x_hat,nrow=8,win='x_hat',opts=dict(title='x_hat'))








if __name__ == '__main__':
    main()

得到结果

具体原因可能是任务太简单了，体现不出VAE的好用。