PyTorch实现用CNN识别手写数字

程序来自莫烦Python，略有删减和改动。

import os
import torch
import torch.nn as nn
import torch.utils.data as Data
import torchvision
import matplotlib.pyplot as plt
 
torch.manual_seed(1)    # reproducible
 
# Hyper Parameters
EPOCH = 1
BATCH_SIZE = 50
LR = 0.001              # learning rate
DOWNLOAD_MNIST = False
 
# Mnist digits dataset
if not(os.path.exists('./mnist/')) or not os.listdir('./mnist/'):    # not mnist dir or mnist is empyt dir. (./表示当前目录)
    DOWNLOAD_MNIST = True
 
train_data = torchvision.datasets.MNIST(
    root='./mnist/',
    train=True,                                     # this is training data
    transform=torchvision.transforms.ToTensor(),    # 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,
)
 
print('train dataset shape: ', train_data.data.size())                 # (60000, 28, 28)
print('train dataset lable shape:', train_data.targets.size())         # (60000)
# plot one example
# plt.imshow(train_data.data[0].numpy(), cmap='gray')
# plt.title('%i' % train_data.targets[0])
# plt.show()
 
# Data Loader for easy mini-batch return in training, the image batch shape will be (BATCH_SIZE, 1, 28, 28)
train_loader = Data.DataLoader(dataset=train_data, batch_size=BATCH_SIZE, shuffle=True)
 
# pick 2000 samples to speed up testing
test_data = torchvision.datasets.MNIST(root='./mnist/', train=False)
test_x = torch.unsqueeze(test_data.data, dim=1).type(torch.FloatTensor)[:2000]/255.   # shape from (2000, 28, 28) to (2000, 1, 28, 28), value in range(0,1)
test_y = test_data.targets[:2000]
 
 
class CNN(nn.Module):
    def __init__(self):
        super(CNN, self).__init__()
        self.conv1 = nn.Sequential(         # input shape (1, 28, 28)
            nn.Conv2d(in_channels=1, out_channels=16, kernel_size=5, stride=1, padding=2), # output shape (16, 28, 28)   
            nn.ReLU(),
            nn.MaxPool2d(kernel_size=2),    # output shape (16, 14, 14)
        )
        self.conv2 = nn.Sequential(         # input shape (16, 14, 14)
            nn.Conv2d(16, 32, 5, 1, 2),     # output shape (32, 14, 14)
            nn.ReLU(),
            nn.MaxPool2d(2),                # output shape (32, 7, 7)
        )
        self.out = nn.Linear(32 * 7 * 7, 10)
 
    def forward(self, x):
        x = self.conv1(x)
        x = self.conv2(x)
        x = x.view(x.size(0), -1)           # flatten the output of conv2 to (batch_size, 32 * 7 * 7)
        output = self.out(x)                # output shape (batch_size, 10)
        return output
 
 
cnn = CNN()
print('CNN architecture:\n ', cnn)
 
optimizer = torch.optim.Adam(cnn.parameters(), lr=LR)   # optimize all cnn parameters
loss_func = nn.CrossEntropyLoss()
 
# training and testing
for epoch in range(EPOCH):
    for iteration, (b_x, b_y) in enumerate(train_loader):
        output = cnn(b_x)               # cnn output, the size of b_x is ([batchsize, channel, height, width)
        loss = loss_func(output, b_y)   # cross entropy loss
        optimizer.zero_grad()           # clear gradients for this training step
        loss.backward()                 # back propagation, compute gradients
        optimizer.step()                # apply gradients
 
        if iteration % 100 == 0:
            test_output = cnn(test_x)
            pred_y = torch.max(test_output, 1)[1].data.numpy()
            accuracy = float((pred_y == test_y.data.numpy()).sum()) / float(test_y.size(0))
            print('Epoch:{:<2d} | Iteration:{:<4d} | Train loss: {:6.3f} | Test accuracy: {:4.2f}'.format(epoch, iteration, loss.data.numpy(), accuracy))
 
# print 10 predictions from test data
test_output = cnn(test_x[:10])
pred_y = torch.max(test_output, 1)[1].data.numpy()
print(pred_y, 'prediction number')
print(test_y[:10].numpy(), 'real number')