21- PyTorch通过CNN实现手写数字识别 (PyTorch系列) (项目二十一)

项目要点

• torch 版本: torch.__version__      # ‘1.13.1+cpu’

• 设置GPU: device = torch.device(‘cuda:0’ if torch.cuda.is_available() else ‘cpu’)

• train_ds = datasets.MNIST(‘./’, train = True, transform=transformation, download= True)  # 数据导入  transformation = transforms.Compose([transforms.ToTensor()])

• train_d1 = torch.utils.data.DataLoader(train_ds, batch_size=64, shuffle=True)  # 转换为dataloader

• 通过iter转换为迭代器: images, labels = next(iter(train_d1))

• 数据转换为numpy: img = img.numpy()

• 显示图片: plt.imshow(img, cmap=’gray’)

• 创建卷积模型:

import torch.nn as nn
import torch.nn.functional as F
class Model(nn.Module):
    def __init__(self):
        super().__init__()
        self.conv1 = nn.Conv2d(1, 32, 3)  # 3表示3*3卷积  
        # in 64 , 1, 28, 28 -> 64, 32, 26, 26
        self.pool = nn.MaxPool2d((2, 2))  # 池化 , # in : 64, 32, 13, 13
        self.conv2 = nn.Conv2d(32, 64, 3) # in: 64, 32, 13, 13 -> out: 64,64,11,11
        # 再加一层池化, input: 64, 64, 11, 11   ->  out: 64, 64, 5, 5 
        self.linear_1 = nn.Linear(64* 5* 5, 256)  # 计算
        self.linear_2 = nn.Linear(256, 10)  # 10个数字的one_hot编码
        
    def forward(self, input):
        x = F.relu(self.conv1(input))
        # 再加池化
        x = self.pool(x)
        # 卷积
        x = F.relu(self.conv2(x))
        x=  self.pool(x)
        # flatten
        x = x.view(-1, 64 * 5 * 5)  
        # 卷积
        x = F.relu(self.linear_1(x))
        x = self.linear_2(x)
        return x

• 定义损失函数: loss_fn = torch.nn.CrossEntropyLoss()
• optimizer 优化器: optimizer = optim.Adam(model.parameters(), lr=0.001)     # 防止过拟合
• 数据位置调整: x, y = x.to(device), y.to(device)
• 梯度清零: optimizer.zero_grad()
• backward 反向传播: loss.backward()

---

一 手写数字识别

1.1 导包

import torch
import torch.nn as nn
import torch.nn.functional as F
import torch.optim as optim
import numpy as np
import matplotlib.pyplot as plt

• 查看torch版本

# torch 版本
torch.__version__   # '1.13.1+cpu'

1.2 定义GPU设置

• 使用GPU进行训练
• 把模型转移到GPU上
• 将每一批次的训练数据转移到GPU上

device = torch.device('cuda:0' if torch.cuda.is_available() else 'cpu')
device     # device(type='cpu')

1.3 导入数据

transforms.ToTensor:

• 1.把数据转化为tensor
• 2.数据的值转化为0到1之间
• 3.会把channel放到第一个维度上

# torchvision 内置了常用的数据集和常见的模型.
import torchvision
# transforms 用来做数据增强, 数据预处理的功能
from torchvision import datasets, transforms

transformation = transforms.Compose([transforms.ToTensor(), ])
# 训练数据
train_ds = datasets.MNIST('./',train = True,transform=transformation,download= True)
# 测试数据
test_ds = datasets.MNIST('./',train = False,transform=transformation,download= True)

• 转换成dataloader

# 转换成dataloader
train_d1 = torch.utils.data.DataLoader(train_ds, batch_size=64, shuffle=True)
test_d1 = torch.utils.data.DataLoader(test_ds, batch_size=256)

• 通过 iter 转换为迭代器

# 通过iter转换为迭代器
images, labels = next(iter(train_d1))
# pytorch中图片的表现形式[batch, channel, highet, width]
images.shape   # torch.Size([64, 1, 28, 28])
labels

img = images[0]
img.shape    # torch.Size([1, 28, 28])
img = img.numpy()
img.shape    # (1, 28, 28)

img = np.squeeze(img)  # 去掉1所在的维度
img.shape  # (28, 28)
plt.imshow(img, cmap='gray')

 1.4 创建模型

class Model(nn.Module):
    def __init__(self):
        super().__init__()
        self.conv1 = nn.Conv2d(1, 32, 3)  # 3表示3*3卷积  
        # in 64 , 1, 28, 28 -> 64, 32, 26, 26
        self.pool = nn.MaxPool2d((2, 2))  # 池化 , # in : 64, 32, 13, 13
        self.conv2 = nn.Conv2d(32, 64, 3) # in: 64, 32, 13, 13 -> out: 64,64,11,11
        # 再加一层池化, input: 64, 64, 11, 11   ->  out: 64, 64, 5, 5 
        self.linear_1 = nn.Linear(64* 5* 5, 256)  # 计算
        self.linear_2 = nn.Linear(256, 10)  # 10个数字的one_hot编码
        
    def forward(self, input):
        x = F.relu(self.conv1(input))
        # 再加池化
        x = self.pool(x)
        # 卷积
        x = F.relu(self.conv2(x))
        x=  self.pool(x)
        # flatten
        x = x.view(-1, 64 * 5 * 5)  
        # 卷积
        x = F.relu(self.linear_1(x))
        x = self.linear_2(x)
        return x
    
model = Model()
# 把model拷贝到GPU 
model.to(device)

1.5 定义训练过程

# 定义损失函数
loss_fn = torch.nn.CrossEntropyLoss()
# optimizer 优化器, 防止过拟合
optimizer = optim.Adam(model.parameters(), lr=0.001)

# 训练过程
def fit(epoch, model, train_loader, test_loader):
    correct = 0
    total = 0
    running_loss = 0
    
    for x, y in train_loader:
        # 把数据放到GPU上
        x, y = x.to(device), y.to(device)
        y_pred = model(x)
        loss = loss_fn(y_pred, y)
        # 梯度清零
        optimizer.zero_grad()
        loss.backward()  # backward 反向传播
        optimizer.step()
        
        # 计算损失过程
        with torch.no_grad():
            y_pred = torch.argmax(y_pred, dim=1)
            correct += (y_pred == y).sum().item()
            total += y.size(0)
            running_loss += loss.item()
            
        # 循环完一次后, 计算损失
    epoch_loss = running_loss / len(train_loader.dataset)
    epoch_acc = correct / total

    # 测试数据的代码
    test_correct = 0
    test_total = 0
    test_running_loss = 0
    with torch.no_grad():
        for x, y in test_loader:
            x, y = x.to(device), y.to(device)
            y_pred = model(x)
            loss = loss_fn(y_pred, y)

            # 计算损失
            y_pred = torch.argmax(y_pred, dim=1)
            test_correct += (y_pred == y).sum().item()
            test_total += y.size(0)
            test_running_loss += loss.item()

    # 计算平均损失
    test_epoch_loss = test_running_loss /len(test_loader.dataset)
    test_epoch_acc = test_correct / test_total

    # 打印输出
    print('epoch:', epoch,
          'loss:', round(epoch_loss, 3),
          'accuracy:', round(epoch_acc, 3),
          'test_loss:', round(test_epoch_loss, 3),
          'test_accuracy:', round(test_epoch_acc, 3))
        
    return epoch_loss, epoch_acc, test_epoch_loss, test_epoch_acc

# 执行操作  # 可以打包一个history
epochs = 20
train_loss = []
train_acc = []
test_loss = []
test_acc = []

for epoch in range(epochs):
    epoch_loss, epoch_acc, test_epoch_loss, test_epoch_acc = fit(epoch, model,
                                                                 train_d1, test_d1)
    train_loss.append(epoch_loss)
    train_acc.append(epoch_acc)
    test_loss.append(test_epoch_loss)
    test_acc.append(test_epoch_acc)

1.6 添加dropout 和 BN层

# 定义模型
class Net(nn.Module):
    def __init__(self):
        super().__init__()
        self.conv1 = nn.Conv2d(3, 16, 3)   # 16 * 94 * 94
        self.pool = nn.MaxPool2d(2, 2)     # 16 * 47 * 47
        self.conv2 = nn.Conv2d(16, 32, 3)  # 32 * 45 * 45  -> pooling -> 32 * 22 * 22
        self.conv3 = nn.Conv2d(32, 64, 3)  # 64 * 20 * 20  -> pooling -> 64 * 10 * 10
        self.dropout = nn.Dropout()
        
        # batch , channel, height, width, 64, 
        self.fc1 = nn.Linear(64 * 10 * 10, 1024)
        self.fc2 = nn.Linear(1024, 256)
        self.fc3 = nn.Linear(256, 4)
        
    def forward(self, x):
        x = self.pool(F.relu(self.conv1(x)))
        x = self.pool(F.relu(self.conv2(x)))
        x = self.pool(F.relu(self.conv3(x)))
        # x.view(-1, 64 * 10 * 10)
        x = nn.Flatten()(x)
        x = F.relu(self.fc1(x))
        x = self.dropout(x)
        x = F.relu(self.fc2(x))
        x = self.dropout(x)
        x = self.fc3(x)
        return x

# 添加BN层.  # 定义模型
class Net(nn.Module):
    def __init__(self):
        super().__init__()
        self.conv1 = nn.Conv2d(3, 16, 3)   # 16 * 94 * 94
        self.bn1 = nn.BatchNorm2d(16)
        self.pool = nn.MaxPool2d(2, 2)     # 16 * 47 * 47
        
        self.conv2 = nn.Conv2d(16, 32, 3)  # 32 * 45 * 45  -> pooling -> 32 * 22 * 22
        self.bn2 = nn.BatchNorm2d(32)
        self.conv3 = nn.Conv2d(32, 64, 3)  # 64 * 20 * 20  -> pooling -> 64 * 10 * 10
        self.bn3 = nn.BatchNorm2d(64)
        self.dropout = nn.Dropout()
        
        # batch , channel, height, width, 64, 
        self.fc1 = nn.Linear(64 * 10 * 10, 1024)
        self.bn_fc1 = nn.BatchNorm1d(1024)
        self.fc2 = nn.Linear(1024, 256)
        self.bn_fc2 = nn.BatchNorm1d(256)
        self.fc3 = nn.Linear(256, 4)
        
    def forward(self, x):
        x = self.pool(F.relu(self.conv1(x)))
        x = self.bn1(x)
        x = self.pool(F.relu(self.conv2(x)))
        x = self.bn2(x)
        x = self.pool(F.relu(self.conv3(x)))
        x = self.bn3(x)
        # x.view(-1, 64 * 10 * 10)
        x = nn.Flatten()(x)
        x = F.relu(self.fc1(x))
        x = self.bn_fc1(x)
        x = self.dropout(x)
        x = F.relu(self.fc2(x))
        x = self.bn_fc2(x)
        x = self.dropout(x)
        x = self.fc3(x)
        return x