【深度学习】-pytorch新手入门经验，完整代码+详细注释

第一阶段：拿来主义

1.学会如何用torch

方式：看大量博文，将手写数字识别用torch做出来

成果：成功做出效果

import torch
from torchvision import datasets,transforms
import torchvision
from torch.autograd import  Variable
import numpy as np
import matplotlib.pyplot as plt
from torch.utils.data import DataLoader

# 超参数
# epoch的数量定义了我们将循环整个训练数据集的次数，而learning_rate和momentum是我们稍后将使用的优化器的超参数
n_epochs = 3
batch_size_train = 64
batch_size_test = 1000
learning_rate = 0.01
momentum = 0.5
log_interval = 10
random_seed = 1
torch.manual_seed(random_seed)

# 数据集
train_loader = torch.utils.data.DataLoader(
  torchvision.datasets.MNIST('./data/', train=True, download=True,
                             transform=torchvision.transforms.Compose([
                               torchvision.transforms.ToTensor(),
                               torchvision.transforms.Normalize(
                                 (0.1307,), (0.3081,))
                             ])),
  batch_size=batch_size_train, shuffle=True)
test_loader = torch.utils.data.DataLoader(
  torchvision.datasets.MNIST('./data/', train=False, download=True,
                             transform=torchvision.transforms.Compose([
                               torchvision.transforms.ToTensor(),
                               torchvision.transforms.Normalize(
                                 (0.1307,), (0.3081,))
                             ])),
  batch_size=batch_size_test, shuffle=True)

# 查看图片
examples = enumerate(test_loader)
batch_idx, (example_data, example_targets) = next(examples)
# print(example_targets)
# print(example_data.shape)
import matplotlib.pyplot as plt
fig = plt.figure()
for i in range(6):
  plt.subplot(2,3,i+1)
  plt.tight_layout()
  plt.imshow(example_data[i][0], cmap='gray', interpolation='none')
  plt.title("Ground Truth: {}".format(example_targets[i]))
  plt.xticks([])
  plt.yticks([])
plt.show()

# 构建网络
import torch.nn as nn
import torch.nn.functional as F
import torch.optim as optim

class Net(nn.Module):
    def __init__(self):
        super(Net, self).__init__()
        self.conv1 = nn.Conv2d(1, 10, kernel_size=5)
        self.conv2 = nn.Conv2d(10, 20, kernel_size=5)
        self.conv2_drop = nn.Dropout2d()
        self.fc1 = nn.Linear(320, 50)
        self.fc2 = nn.Linear(50, 10)
    def forward(self, x):
        x = F.relu(F.max_pool2d(self.conv1(x), 2))
        x = F.relu(F.max_pool2d(self.conv2_drop(self.conv2(x)), 2))
        x = x.view(-1, 320)
        x = F.relu(self.fc1(x))
        x = F.dropout(x, training=self.training)
        x = self.fc2(x)
        return F.log_softmax(x)

# 初始化网络和优化器
network = Net()
optimizer = optim.SGD(network.parameters(), lr=learning_rate,momentum=momentum)

# 模型训练
# 存储数据，数据可视化
train_losses = []
train_counter = []
test_losses = []
test_counter = [i*len(train_loader.dataset) for i in range(n_epochs + 1)]


def train(epoch):
    network.train()
    for batch_idx, (data, target) in enumerate(train_loader):
        optimizer.zero_grad()
        output = network(data)
        loss = F.nll_loss(output, target)
        loss.backward()
        optimizer.step()
        if batch_idx % log_interval == 0:
            print('Train Epoch: {} [{}/{} ({:.0f}%)]\tLoss: {:.6f}'.format(
                epoch, batch_idx * len(data), len(train_loader.dataset),
                       100. * batch_idx / len(train_loader), loss.item()))
            train_losses.append(loss.item())
            train_counter.append(
                (batch_idx * 64) + ((epoch - 1) * len(train_loader.dataset)))
            # 模型保存
            # torch.save(network.state_dict(), './Model/main03/model.pth')
            # torch.save(optimizer.state_dict(), './Model/main03/optimizer.pth')

# train(1)



def test():
    network.eval()
    test_loss = 0
    correct = 0
    with torch.no_grad():
        for data, target in test_loader:
            output = network(data)
            test_loss += F.nll_loss(output, target, size_average=False).item()
            pred = output.data.max(1, keepdim=True)[1]
            correct += pred.eq(target.data.view_as(pred)).sum()
    test_loss /= len(test_loader.dataset)
    test_losses.append(test_loss)
    print('\nTest set: Avg. loss: {:.4f}, Accuracy: {}/{} ({:.0f}%)\n'.format(
        test_loss, correct, len(test_loader.dataset),
        100. * correct / len(test_loader.dataset)))


# test()
for epoch in range(1, n_epochs + 1):
  train(epoch)
  test()
  
import matplotlib.pyplot as plt
fig = plt.figure()
plt.plot(train_counter, train_losses, color='blue')
plt.scatter(test_counter, test_losses, color='red')
plt.legend(['Train Loss', 'Test Loss'], loc='upper right')
plt.xlabel('number of training examples seen')
plt.ylabel('negative log likelihood loss')
plt.show()

但不会读取保存模型，下一步通过分析代码找到读取模型的方式

# 加载保存模型预测图片值
def model_output():
    # 加载保存模型
    continued_network = Net()
    continued_optimizer = optim.SGD(network.parameters(), lr=learning_rate,
                                    momentum=momentum)
    network_state_dict = torch.load('./Model/main03/model.pth')
    continued_network.load_state_dict(network_state_dict)
    optimizer_state_dict = torch.load('./Model/main03/optimizer.pth')
    continued_optimizer.load_state_dict(optimizer_state_dict)

    # 加载数据
    dataiter = iter(test_loader)
    images, labels = dataiter.next()

    # 查看数据
    # print(images[0],'\n',len(images),'\n',labels[0],'\n',len(labels))
    # 打印图片
    # plt.imshow(images[0].numpy().squeeze(), cmap='Greys_r')
    # plt.show()
    print('GroundTruth:',labels[0].numpy())

    # 输出预测值
    continued_network.eval()
    output = continued_network(images)  # 必须与网络输入通道数一致，1000个,得到预测值
    print(output)
    _, predicted = torch.max(output,1)  # 解析预测值
    print(predicted)

    # 确认准确率
    acc = 0
    predicted = predicted.numpy()
    print(predicted)
    labels = labels.numpy()
    print(labels)
    for i in range(len(labels)):
        if labels[i]==predicted[i]:
            acc+=1
    print("准确率:{:.2%}".format(acc/1000))

成功将模型输出，不过还不清楚如何继续训练，下一步解决如何继续训练

# 模型加载
# coding:utf-8

import torch
from torchvision import datasets,transforms
import torchvision
from torch.autograd import  Variable
import numpy as np
import matplotlib.pyplot as plt
from torch.utils.data import DataLoader
# from main_03 import *

# 超参数
# epoch的数量定义了我们将循环整个训练数据集的次数，而learning_rate和momentum是我们稍后将使用的优化器的超参数
n_epochs = 3
batch_size_train = 64
batch_size_test = 1000
learning_rate = 0.01
momentum = 0.5
log_interval = 10
random_seed = 1
torch.manual_seed(random_seed)

# 数据集
train_loader = torch.utils.data.DataLoader(
  torchvision.datasets.MNIST('./data/', train=True, download=True,
                             transform=torchvision.transforms.Compose([
                               torchvision.transforms.ToTensor(),
                               torchvision.transforms.Normalize(
                                 (0.1307,), (0.3081,))
                             ])),
  batch_size=batch_size_train, shuffle=True)
test_loader = torch.utils.data.DataLoader(
  torchvision.datasets.MNIST('./data/', train=False, download=True,
                             transform=torchvision.transforms.Compose([
                               torchvision.transforms.ToTensor(),
                               torchvision.transforms.Normalize(
                                 (0.1307,), (0.3081,))
                             ])),
  batch_size=batch_size_test, shuffle=True)


# 构建网络
import torch.nn as nn
import torch.nn.functional as F
import torch.optim as optim

class Net(nn.Module):
    def __init__(self):
        super(Net, self).__init__()
        self.conv1 = nn.Conv2d(1, 10, kernel_size=5)
        self.conv2 = nn.Conv2d(10, 20, kernel_size=5)
        self.conv2_drop = nn.Dropout2d()
        self.fc1 = nn.Linear(320, 50)
        self.fc2 = nn.Linear(50, 10)
    def forward(self, x):
        x = F.relu(F.max_pool2d(self.conv1(x), 2))
        x = F.relu(F.max_pool2d(self.conv2_drop(self.conv2(x)), 2))
        x = x.view(-1, 320)
        x = F.relu(self.fc1(x))
        x = F.dropout(x, training=self.training)
        x = self.fc2(x)
        return F.log_softmax(x)

# 初始化网络和优化器
# network = Net()
# optimizer = optim.SGD(network.parameters(), lr=learning_rate,momentum=momentum)


# 模型训练

# 加载保存模型
continued_network = Net()
continued_optimizer = optim.SGD(continued_network.parameters(), lr=learning_rate,
                                momentum=momentum)
network_state_dict = torch.load('./Model/main03/model.pth')
continued_network.load_state_dict(network_state_dict)
optimizer_state_dict = torch.load('./Model/main03/optimizer.pth')
continued_optimizer.load_state_dict(optimizer_state_dict)

# 加载保存模型预测图片值
def model_output():
    # 加载数据
    dataiter = iter(test_loader)
    images, labels = dataiter.next()

    # 查看数据
    # print(images[0],'\n',len(images),'\n',labels[0],'\n',len(labels))
    # 打印图片
    # plt.imshow(images[0].numpy().squeeze(), cmap='Greys_r')
    # plt.show()
    print('GroundTruth:',labels[0].numpy())

    # 输出预测值
    continued_network.eval()
    output = continued_network(images)  # 必须与网络输入通道数一致，1000个,得到预测值
    print(output)
    _, predicted = torch.max(output,1)  # 解析预测值
    print(predicted)

    # 确认准确率
    acc = 0
    predicted = predicted.numpy()
    print(predicted)
    labels = labels.numpy()
    print(labels)
    for i in range(len(labels)):
        if labels[i]==predicted[i]:
            acc+=1
    print("准确率:{:.2%}".format(acc/1000))

# 继续训练
def train(epoch):
    continued_network.train()
    for batch_idx, (data, target) in enumerate(train_loader):
        continued_optimizer.zero_grad()  # 先将梯度清零
        output = continued_network(data)  # 正向传播得到预测值
        loss = F.nll_loss(output, target)  # 计算loss，target是标签
        loss.backward()  # 计算梯度
        continued_optimizer.step()  # 反向传播更新参数
        # 交互
        if batch_idx % log_interval == 0:  # log_interval==10，每十次就记录一下
            print('Train Epoch: {} [{}/{} ({:.0f}%)]\tLoss: {:.6f}'.format(
                epoch, batch_idx * len(data), len(train_loader.dataset),
                       100. * batch_idx / len(train_loader), loss.item()))
            
            # 模型保存
            # torch.save(network.state_dict(), './Model/main03/model_save.pth')
            # torch.save(continued_optimizer.state_dict(), './Model/main03/optimizer.pth')

model_output()

train(1)

# 另外一种加载选择
# path='../save_model/ResNet18/1.pth'
# model = torch.load(path)
# model.eval()



# fig = plt.figure()
# plt.plot(train_counter, train_losses, color='blue')
# plt.scatter(test_counter, test_losses, color='red')
# plt.legend(['Train Loss', 'Test Loss'], loc='upper right')
# plt.xlabel('number of training examples seen')
# plt.ylabel('negative log likelihood loss')
# plt.show()

第二阶段：改进——使用本地图片自己处理

数据集：将numpy数据利用plt转为图片储存

自己建立标签txt

bug：数据集标签处理时部分\t不能分开，特殊处理

成功输出

继续训练

两种保存并加载模型的方法：

1.保存整个最终模型，预测，不能继续训练

torch.save(model, './model_save/2_params.pth')

# 加载保存模型
model_weight_path = './model_save/2_params.pth'
model = torch.load(model_weight_path)
model.eval()

2.保存模型参数与优化器参数，可以继续训练

torch.save(model.state_dict(), './model_save/2_params.pth')
torch.save(optimizer.state_dict(), './model_save/2_optimizer.pth')

# 加载保存模型
model = Net()
optimizer = optim.SGD(continued_network.parameters(), lr=learning_rate,
                                momentum=momentum)
network_state_dict = torch.load('./model_save/1.pth')
model.load_state_dict(network_state_dict)
optimizer_state_dict = torch.load('./model_save/1_params.pth')
optimizer.load_state_dict(optimizer_state_dict)

第三阶段：应用——叶菜病虫害图像识别挑战赛

构建数据集

将官方训练数据集拆开

采用训练集：测试集=8：2

无法训练，自己电脑算力太差

收工

告一段落

开始学习数学建模

---

后面系统学习了深度学习的理论知识——修改batch-size为8（调小），可以降低所需显存大小

并且修改模型参数

成功运行