Pytorch VAE Implementation for MNIST
Loss部分必须给予KL一个较小的权重,模型才可以收敛输出,为什么呢?
生成模型GAN+VAE Collection
https://github.com/hwalsuklee/tensorflow-generative-model-collections/blob/master/README.md
Loss Function & Reparemerization:
http://www.cnblogs.com/wzyj/p/9766655.html
Generic AutoEncoder:
https://v.youku.com/v_show/id_XMjc2MDUxNzUyOA==.html?spm=a2h0j.11185381.listitem_page1.5!25~A&&f=49718057
https://github.com/ffzs/ml_pytorch/blob/master/ml_pytorch_VAE.py
import torch
import torch.nn as nn
import torch.utils.data as Data
import torchvision
from torch.autograd import Variable
import matplotlib.pyplot as plt
from mpl_toolkits.mplot3d import Axes3D
from matplotlib import cm
import numpy as np
# torch.manual_seed(1) # reproducible
# Hyper Parameters
EPOCH = 100
BATCH_SIZE = 50
LR = 0.001 # learning rate
DOWNLOAD_MNIST = False
N_TEST_IMG = 5
train_data = torchvision.datasets.MNIST(
root='./mnist',
train=True,
transform=torchvision.transforms.Compose([
torchvision.transforms.ToTensor(),
torchvision.transforms.Normalize((0.1307,), (0.3081,))
]),
download=DOWNLOAD_MNIST
)
test_data = torchvision.datasets.MNIST(
root='./mnist',
train=False,
transform=torchvision.transforms.Compose([
torchvision.transforms.ToTensor(),
torchvision.transforms.Normalize((0.1307,), (0.3081,))
]),
download=DOWNLOAD_MNIST
)
print(train_data.train_data.numpy().shape)
print(test_data.test_data.numpy().shape)
train_loader = Data.DataLoader(
dataset=train_data,
batch_size=BATCH_SIZE,
shuffle=True,
num_workers=2
)
plt.imshow(train_data.train_data[2].numpy(), cmap='gray')
plt.title('%i' % train_data.train_labels[2])
plt.show()
# Data Loader for easy mini-batch return in training, the image batch shape will be (50, 1, 28, 28)
train_loader = Data.DataLoader(dataset=train_data, batch_size=BATCH_SIZE, shuffle=True)
class Net(nn.Module):
def __init__(self, gpu_status):
super(Net, self).__init__()
self.gpu_status = gpu_status
self.hidden = 10
self.en_conv_1 = nn.Sequential(
nn.Conv2d(1, 16, 4, 2, 1),
nn.BatchNorm2d(16),
nn.Tanh(),
nn.Conv2d(16, 32, 4, 2, 1),
nn.BatchNorm2d(32),
nn.Tanh(),
nn.Conv2d(32, 16, 3, 1, 1),
nn.BatchNorm2d(16),
nn.Tanh()
)
self.en_conv_2 = nn.Sequential(
nn.Conv2d(1, 16, 4, 2, 1),
nn.BatchNorm2d(16),
nn.Tanh(),
nn.Conv2d(16, 32, 4, 2, 1),
nn.BatchNorm2d(32),
nn.Tanh(),
nn.Conv2d(32, 16, 3, 1, 1),
nn.BatchNorm2d(16),
nn.Tanh()
)
self.en_fc_1 = nn.Linear(16 * 7 * 7, self.hidden)
self.en_fc_2 = nn.Linear(16 * 7 * 7, self.hidden)
self.de_fc = nn.Linear(self.hidden, 16 * 7 * 7)
self.de_conv = nn.Sequential(
nn.ConvTranspose2d(16, 16, 4, 2, 1),
nn.BatchNorm2d(16),
nn.Tanh(),
nn.ConvTranspose2d(16, 1, 4, 2, 1),
nn.Sigmoid()
)
def encoder(self, x):
conv_out_1 = self.en_conv_1(x)
conv_out_1 = conv_out_1.view(x.size(0), -1)
conv_out_2 = self.en_conv_2(x)
conv_out_2 = conv_out_2.view(x.size(0), -1)
encoded_fc1 = self.en_fc_1(conv_out_1)
encoded_fc2 = self.en_fc_2(conv_out_2)
return encoded_fc1, encoded_fc2 # 这里分别表示均值和标准差的采样
def sampler(self, mean, std):
var = std.mul(0.5).exp_()
eps = torch.FloatTensor(var.size()).normal_() # 生成一个与输入大小一致的标准正态分布随机数
eps = Variable(eps)
if self.gpu_status:
eps = eps.cuda()
return eps.mul(var).add_(mean)
def decoder(self, x):
out = self.de_fc(x)
out = out.view(-1, 16, 7, 7)
out = self.de_conv(out)
return out
def forward(self, x):
mean, std = self.encoder(x)
code = self.sampler(mean, std)
out = self.decoder(code)
# return encoded_fc, decoded
return out, code, mean, std
use_cuda = torch.cuda.is_available()
device = torch.device("cuda" if use_cuda else "cpu")
autoencoder = Net(gpu_status=use_cuda).to(device)
optimizer = torch.optim.Adam(autoencoder.parameters(), lr=LR)
bce = nn.BCELoss()
mse = nn.MSELoss()
bce.size_average = False
mse.size_average = False
if torch.cuda.is_available():
autoencoder = autoencoder.cuda()
bce = bce.cuda()
mse = mse.cuda()
def loss_f(out, target, mean, std):
# 公式上理解,std实际上是方差的对数
# bceloss = bce(out, target)
mseloss = mse(out, target)
# latent_loss = torch.sum(mean.pow(2).add_(std.exp()).mul_(-1).add_(1).add_(std)).mul_(-0.5)
KLD = -0.5 * torch.sum(1 + std - mean.pow(2) - std.exp())
return mseloss + 0.0002 * KLD
# initialize figure
f, a = plt.subplots(2, N_TEST_IMG, figsize=(5, 2))
plt.ion() # continuously plot
# original data (first row) for viewing
view_data = train_data.train_data[:N_TEST_IMG].view(-1,1,28,28).type(torch.cuda.FloatTensor)
for i in range(N_TEST_IMG):
a[0][i].imshow(np.reshape((view_data.cpu()).data.numpy()[i], (28, 28)), cmap='gray'); a[0][i].set_xticks(()); a[0][i].set_yticks(())
for epoch in range(EPOCH):
for step, (x, b_label) in enumerate(train_loader):
b_x = Variable(x).to(device) # batch x, shape (batch, 28*28)
b_y = Variable(x).to(device) # batch y, shape (batch, 28*28)
output, _, mean, std = autoencoder(b_x)
loss = loss_f(output, b_y, mean, std)
optimizer.zero_grad() # clear gradients for this training step
loss.backward() # backpropagation, compute gradients
optimizer.step() # apply gradients
if step % 100 == 0:
print('Epoch: ', epoch, '| train loss: %.4f' % loss.item())
# plotting decoded image (second row)
decoded_data, _, _, _ = autoencoder(view_data)
for i in range(N_TEST_IMG):
a[1][i].clear()
a[1][i].imshow(np.reshape((decoded_data.cpu()).data.numpy()[i], (28, 28)), cmap='gray')
a[1][i].set_xticks(()); a[1][i].set_yticks(())
plt.draw(); plt.pause(0.05)
plt.ioff()
plt.show()
# visualize in 3D plot
view_data = train_data.train_data[:200].view(-1, 28*28).type(torch.FloatTensor)/255.
encoded_data, _ = autoencoder(view_data)
fig = plt.figure(2); ax = Axes3D(fig)
X, Y, Z = encoded_data.data[:, 0].numpy(), encoded_data.data[:, 1].numpy(), encoded_data.data[:, 2].numpy()
values = train_data.train_labels[:200].numpy()
for x, y, z, s in zip(X, Y, Z, values):
c = cm.rainbow(int(255*s/9)); ax.text(x, y, z, s, backgroundcolor=c)
ax.set_xlim(X.min(), X.max()); ax.set_ylim(Y.min(), Y.max()); ax.set_zlim(Z.min(), Z.max())
plt.show()