1.自编码粗线条理解

通过encoder 编码器压缩得到原数据的精髓, 再创建一个小神经网络学习精髓的数据,减少神经网络的负担, 达到效果。

2.代码

import torch
import torch.nn as nn
import torch.utils.data as Data
import torchvision
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 = 10                                  # 训练整批数据10次
BATCH_SIZE = 64                             # 批训练的数据个数
LR = 0.005                                  # learning rate
DOWNLOAD_MNIST = False                      # 已经下载好了mnist数据就写 False
N_TEST_IMG = 5

# Mnist digits dataset
train_data = torchvision.datasets.MNIST(
    root='./mnist/',                                # 保存或者提取数据位置
    train=True,                                     # training data
    transform=torchvision.transforms.ToTensor(),    # 转化数据Converts a PIL.Image or numpy.ndarray to
                                                    # torch.FloatTensor of shape (C x H x W) and normalize in the range [0.0, 1.0]
    download=DOWNLOAD_MNIST,                        # 下载
)

# plot one example
print(train_data.data.size())     # (60000, 28, 28)
print(train_data.targets.size())   # (60000)
plt.imshow(train_data.data[2].numpy(), cmap='gray')
plt.title('%i' % train_data.targets[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 AutoEncoder(nn.Module):
    def __init__(self):
        super(AutoEncoder, self).__init__()

        self.encoder = nn.Sequential(#该形式快
            nn.Linear(28*28, 128),#添加神经层 28*28长宽
            nn.Tanh(),
            nn.Linear(128, 64),#压缩
            nn.Tanh(),
            nn.Linear(64, 12),#再压缩
            nn.Tanh(),
            nn.Linear(12, 3),   # compress to 3 features which can be visualized in plt
        )
        self.decoder = nn.Sequential(#解码器
            nn.Linear(3, 12),#解压成原始
            nn.Tanh(),
            nn.Linear(12, 64),
            nn.Tanh(),
            nn.Linear(64, 128),
            nn.Tanh(),
            nn.Linear(128, 28*28),
            nn.Sigmoid(),       # 压缩输出值为（0，1）compress to a range (0, 1)
        )

    def forward(self, x):
        encoded = self.encoder(x)
        decoded = self.decoder(encoded)
        return encoded, decoded

#定义自编码
autoencoder = AutoEncoder()
#优化
optimizer = torch.optim.Adam(autoencoder.parameters(), lr=LR)
loss_func = nn.MSELoss()#均方差优化

# 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.data[:N_TEST_IMG].view(-1, 28*28).type(torch.FloatTensor)/255.
for i in range(N_TEST_IMG):
    a[0][i].imshow(np.reshape(view_data.data.numpy()[i], (28, 28)), cmap='gray'); a[0][i].set_xticks(()); a[0][i].set_yticks(())


#开始训练啦
for epoch in range(EPOCH):
    for step, (x, b_label) in enumerate(train_loader):
        b_x = x.view(-1, 28*28)   # batch x, shape (batch, 28*28)
        b_y = x.view(-1, 28*28)   # batch y, shape (batch, 28*28)

        encoded, decoded = autoencoder(b_x)

        loss = loss_func(decoded, b_y)      # 必备手续：mean square error
        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.data.numpy())

            # 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.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.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.targets[:200].numpy()  #warnings.warn("train_labels has been renamed targets")
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()

3.运行结果

torch.Size([60000, 28, 28])
torch.Size([60000])
Epoch:  0 | train loss: 0.2335
Epoch:  0 | train loss: 0.0676
Epoch:  0 | train loss: 0.0648
Epoch:  0 | train loss: 0.0628
Epoch:  0 | train loss: 0.0570
Epoch:  0 | train loss: 0.0577
Epoch:  0 | train loss: 0.0558
Epoch:  0 | train loss: 0.0496
Epoch:  0 | train loss: 0.0477
Epoch:  0 | train loss: 0.0531
Epoch:  1 | train loss: 0.0462
Epoch:  1 | train loss: 0.0507
Epoch:  1 | train loss: 0.0484
Epoch:  1 | train loss: 0.0460
Epoch:  1 | train loss: 0.0460
Epoch:  1 | train loss: 0.0456
Epoch:  1 | train loss: 0.0422
Epoch:  1 | train loss: 0.0454
Epoch:  1 | train loss: 0.0447
Epoch:  1 | train loss: 0.0408
Epoch:  2 | train loss: 0.0419
Epoch:  2 | train loss: 0.0425
Epoch:  2 | train loss: 0.0411
Epoch:  2 | train loss: 0.0434
Epoch:  2 | train loss: 0.0391
Epoch:  2 | train loss: 0.0426
Epoch:  2 | train loss: 0.0449
Epoch:  2 | train loss: 0.0414
Epoch:  2 | train loss: 0.0420
Epoch:  2 | train loss: 0.0394
Epoch:  3 | train loss: 0.0401
Epoch:  3 | train loss: 0.0406
Epoch:  3 | train loss: 0.0382
Epoch:  3 | train loss: 0.0424
Epoch:  3 | train loss: 0.0431
Epoch:  3 | train loss: 0.0429
Epoch:  3 | train loss: 0.0453
Epoch:  3 | train loss: 0.0388
Epoch:  3 | train loss: 0.0410
Epoch:  3 | train loss: 0.0442
Epoch:  4 | train loss: 0.0371
Epoch:  4 | train loss: 0.0388
Epoch:  4 | train loss: 0.0401
Epoch:  4 | train loss: 0.0376
Epoch:  4 | train loss: 0.0375
Epoch:  4 | train loss: 0.0390
Epoch:  4 | train loss: 0.0405
Epoch:  4 | train loss: 0.0376
Epoch:  4 | train loss: 0.0376
Epoch:  4 | train loss: 0.0351
Epoch:  5 | train loss: 0.0368
Epoch:  5 | train loss: 0.0350
Epoch:  5 | train loss: 0.0383
Epoch:  5 | train loss: 0.0374
Epoch:  5 | train loss: 0.0394
Epoch:  5 | train loss: 0.0356
Epoch:  5 | train loss: 0.0337
Epoch:  5 | train loss: 0.0387
Epoch:  5 | train loss: 0.0391
Epoch:  5 | train loss: 0.0369
Epoch:  6 | train loss: 0.0365
Epoch:  6 | train loss: 0.0347
Epoch:  6 | train loss: 0.0376
Epoch:  6 | train loss: 0.0385
Epoch:  6 | train loss: 0.0367
Epoch:  6 | train loss: 0.0362
Epoch:  6 | train loss: 0.0376
Epoch:  6 | train loss: 0.0369
Epoch:  6 | train loss: 0.0366
Epoch:  6 | train loss: 0.0372
Epoch:  7 | train loss: 0.0354
Epoch:  7 | train loss: 0.0349
Epoch:  7 | train loss: 0.0346
Epoch:  7 | train loss: 0.0389
Epoch:  7 | train loss: 0.0394
Epoch:  7 | train loss: 0.0360
Epoch:  7 | train loss: 0.0356
Epoch:  7 | train loss: 0.0377
Epoch:  7 | train loss: 0.0364
Epoch:  7 | train loss: 0.0337
Epoch:  8 | train loss: 0.0369
Epoch:  8 | train loss: 0.0398
Epoch:  8 | train loss: 0.0349
Epoch:  8 | train loss: 0.0368
Epoch:  8 | train loss: 0.0333
Epoch:  8 | train loss: 0.0346
Epoch:  8 | train loss: 0.0378
Epoch:  8 | train loss: 0.0361
Epoch:  8 | train loss: 0.0343
Epoch:  8 | train loss: 0.0342
Epoch:  9 | train loss: 0.0346
Epoch:  9 | train loss: 0.0379
Epoch:  9 | train loss: 0.0343
Epoch:  9 | train loss: 0.0349
Epoch:  9 | train loss: 0.0355
Epoch:  9 | train loss: 0.0380
Epoch:  9 | train loss: 0.0329
Epoch:  9 | train loss: 0.0364
Epoch:  9 | train loss: 0.0373
Epoch:  9 | train loss: 0.0347

青青不熬夜

原创文章 23 获赞 1 访问量 719

关注私信

pytorch白话入门笔记1.12-自编码/非监督性学习

1.自编码粗线条理解

2.代码

3.运行结果

猜你喜欢