DL初挑戦 pytorchを使ってSTL10の学習させてみたけどむずすぎた

transform_train = transforms.Compose([
  transforms.RandomHorizontalFlip(),
  transforms.ToTensor(),
  transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5)),
])

transform_test = transforms.Compose([
  transforms.ToTensor(),
  transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5)),
])

trainset = datasets.STL10('./data', split='train', download=True, transform=transform_train)
testset = datasets.STL10('./data', split='test', download=True, transform=transform_test)

# 5000枚
train_loader = DataLoader(
    trainset, batch_size=20, shuffle=True,
    num_workers=4, drop_last=True,
)
# 8000枚
test_loader = DataLoader(
    testset, batch_size=20, shuffle=False,
    num_workers=4, drop_last=False,
)

classes = ('plane', 'bird', 'car', 'cat', 'deer', 'dog', 'horse', 'monkey', 'ship', 'truck')

net = models.vgg16(pretrained=True)

net.classifier[6] = nn.Linear(in_features=4096, out_features=10, bias=True)

criterion = nn.CrossEntropyLoss()
optimizer = optim.SGD(net.parameters(), lr=0.001, momentum=0.9, weight_decay=0.005)

# GPU仕様に変更
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
net = net.to(device)
# 学習結果の保存用
history = {
    'train_loss': [],
    'test_loss': [],
    'test_acc': [],
}
# 学習する
for epoch in range(10):
  running_loss = 0.0
  # 学習開始
  net.train()
  for i, data in enumerate(train_loader, 0):
    inputs, labels = data
    inputs, labels = Variable(inputs), Variable(labels)
    # GPU仕様に変更
    inputs = inputs.to(device)
    labels = labels.to(device)

    optimizer.zero_grad()
    outputs = net(inputs)
    # 損失の計算
    loss = criterion(outputs, labels)
    loss.backward()
    optimizer.step()

    running_loss += loss.data
    if i % 50 == 49:
      print('[%d, %5d] loss: %.3f' % (epoch+1, i+1, running_loss / 100))
      if (i == 249):
        history['train_loss'].append(float(running_loss / 100))
      running_loss = 0.0


  # テストデータで検証
  net.eval()
  correct = 0
  total = 0
  test_loss = 0
  with torch.no_grad():
    for data in test_loader:
        images, labels = data
        images = Variable(images)
        images = images.to(device)
        labels = labels.to(device)
        outputs = net(images)
        _, predicted = torch.max(outputs.data, 1)
        total += labels.size(0)
        correct += (predicted == labels).sum()
        loss = criterion(outputs, labels)
        test_loss += loss.item()

  print('test loss: %.3f' % (test_loss / len(test_loader)))
  print('test Accuracytest: %.3f %%' % (100 * correct / total))
  history['test_loss'].append(float(test_loss / len(test_loader)))
  history['test_acc'].append(float(100 * correct / total))

print('Finished Training')

correct = 0.0
total = 0.0
for data in test_loader:
    images, labels = data
    images = Variable(images)
    images = images.to(device)
    labels = labels.to(device)
    outputs = net(images)
    _, predicted = torch.max(outputs.data, 1)
    total += labels.size(0)
    correct += (predicted == labels).sum()

print('Accuracy of the network on the 8000 test images: %3f %%' % float(100 * correct / total))
class_correct = list(0. for i in range(10))
class_total = list(0. for i in range(10))
for data in test_loader:
    images, labels = data
    images = Variable(images)
    images = images.to(device)
    labels = labels.to(device)
    #print("images type : ", type(images))
    #print("images.shape : ", images.shape)
    # 学習
    outputs = net(images)
    # 一番確率が高いの
    _, predicted = torch.max(outputs.data, 1)

    c = (predicted == labels).squeeze()
    for i in range(10):
        label = labels[i]
        class_correct[label] += c[i]
        class_total[label] += 1


for i in range(10):
    print('Accuracy of %5s : %3f %%' % (
        classes[i], 100 * class_correct[i] / class_total[i]))