mao/torch-train-study2

import torch as t
import torchvision as tv
import torchvision.transforms as transforms
import torch.nn as nn
import torch.nn.functional as F
 
#错误的还未学习
########## 超参数设置 ##########
epochs = 200
learning_rate = 0.001
batch_size = 256
gpu_ids = [0, 1, 2]
 
 
########## 一、数据加载与预处理 ##########
transform = transforms.Compose([
    transforms.ToTensor(),
    transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))
])
trainset = tv.datasets.CIFAR10(root='./data/', train=True, download=True, transform=transform)
testset = tv.datasets.CIFAR10(root='./data/', train=False, download=True, transform=transform)
trainloader = t.utils.data.DataLoader(trainset, batch_size=batch_size, shuffle=True, num_workers=4)
testloader = t.utils.data.DataLoader(testset, batch_size=batch_size, shuffle=False, num_workers=4)
 
 
########## 二、定义神经网络 ##########
class LeNet(nn.Module):
    def __init__(self):
        super(LeNet, self).__init__()
        # (in_channels, out_channels, kernel_size, stride, padding)
        self.conv1 = nn.Conv2d(3, 6, 5, 1, 0)
        self.conv2 = nn.Conv2d(6, 16, 5, 1, 0)
        self.fc1 = nn.Linear(16 * 5 * 5, 120)
        self.fc2 = nn.Linear(120, 84)
        self.fc3 = nn.Linear(84, 10)
 
    def forward(self, x):
        x = F.max_pool2d(F.relu(self.conv1(x)), 2, 2)
        x = F.max_pool2d(F.relu(self.conv2(x)), 2, 2)
        x = x.view(x.size()[0], -1)
        x = F.relu(self.fc1(x))
        x = F.relu(self.fc2(x))
        x = self.fc3(x)
        return x
 
 
########## 三、训练神经网络 ##########
# 神经网络实例化
net = LeNet()
if (len(gpu_ids) == 0) or (not t.cuda.is_available()):
    # gpu_ids列表为空, 或者没GPU可用，则设置为GPU模式。
    device = t.device("cpu")
    print("Train Mode : CPU")
elif t.cuda.is_available() and len(gpu_ids) > 1:
    # gpu_ids列表大于1，表明想用多个GPU训练
    device = t.device("cuda:0")
    net = nn.DataParallel(net, device_ids=gpu_ids)
    print("Train Mode : Multi GPU;", gpu_ids)
else:
    # gpu_ids列表等于1，表明想用一个GPU训练
    device = t.device("cuda:" + str(gpu_ids[0]) if t.cuda.is_available() else "cpu")
    print("Train Mode : One GPU;", device)
net = net.to(device)
print("\n", "##" * 10, "  NetWork  ", "##" * 10, "\n", net, "\n", "##" * 26, "\n")
# 定义损失函数和优化器
criterion = nn.CrossEntropyLoss()
optimizer = t.optim.SGD(net.parameters(), lr=learning_rate, momentum=0.9)
# 开始训练
for epoch in range(epochs):
    running_loss = 0.0
    for i, (inputs, labels) in enumerate(trainloader):
        inputs, labels = inputs.to(device), labels.to(device)
        optimizer.zero_grad()
        outputs = net(inputs)
        loss = criterion(outputs, labels)
        loss.backward()
        optimizer.step()
        running_loss += loss.item()
    print("Epoch%03d: Training_loss = %.5f" % (epoch + 1, running_loss))
 
    # 保存模型并进行验证
    if (epoch + 1) % 10 == 0:
        if len(gpu_ids) > 1:
            t.save(net.module.state_dict(), "Cifar10_LeNet_Epoch" + str(epoch + 1) + ".pth")
        else:
            t.save(net.state_dict(), "Cifar10_LeNet_Epoch" + str(epoch + 1) + ".pth")
        correct = 0
        all = 0
        with t.no_grad():
            for (inputs, labels) in testloader:
                inputs, labels = inputs.to(device), labels.to(device)
                outputs = net(inputs)
                _, predicted = t.max(outputs, 1)
                all += inputs.size()[0]
                correct += (predicted == labels).sum().item()
        print("###" * 15)
        print("Epoch%03d: TestSet_Accuracy = %.3f" % (epoch + 1, correct / all))
        print("###" * 15)
 
print("Train Done!")