python - Pytorch:如何将数据转换为张量

Question

我是 Pytorch 的初学者。我正在尝试引用 Pytorch 教程编写 CNN 代码。下面是代码的一部分，但它显示错误“RuntimeError:变量数据必须是张量，但得到了列表”。我尝试将输入数据转换为张量，但效果不佳。如果有人知道解决方案，请帮助我......

    def read_labels(file):
      dic = {}
      with open(file) as f:
        reader = f
        for row in reader:
            dic[row.split(",")[0]]  = row.split(",")[1].rstrip() #rstrip(): eliminate "\n"
      return dic

    image_names= os.listdir("./train_mini")
    label_dic = read_labels("labels.csv")


    names =[]
    labels = []
    images =[]

    for name in image_names[1:]:
        images.append(cv2.imread("./train_mini/"+name))
        labels.append(label_dic[os.path.splitext(name)[0]])

    """
    Data distribution
    """
    N = len(images)
    N_train = int(N * 0.7)
    N_test = int(N*0.2)

    X_train, X_tmp, Y_train, Y_tmp = train_test_split(images, labels, train_size=N_train)
    X_validation, X_test, Y_validation, Y_test = train_test_split(X_tmp, Y_tmp, test_size=N_test)

    """
    Model Definition
    """

    class CNN(nn.Module):
        def __init__(self):
            super(CNN, self).__init__()
            self.head = nn.Sequential(
                nn.Conv2d(in_channels=1, out_channels=10,
                          kernel_size=5, stride=1),
                nn.MaxPool2d(kernel_size=2),
                nn.ReLU(),
                nn.Conv2d(10, 20, kernel_size=5),
                nn.MaxPool2d(kernel_size=2),
                nn.ReLU())
            self.tail = nn.Sequential(
                nn.Linear(320, 50),
                nn.ReLU(),
                nn.Linear(50, 10))

        def forward(self, x):
            x = self.head(x)
            x = x.view(-1, 320)
            x = self.tail(x)
            return F.log_softmax(x)

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


    """
    Training
    """
    batch_size = 50
    for epoch in range(2):  # loop over the dataset multiple times
        running_loss = 0.0
        for i in range(N / batch_size):
        #for i, data in enumerate(trainloader, 0):
            batch = batch_size * i

            # get the inputs
            images_batch = X_train[batch:batch + batch_size]
            labels_batch = Y_train[batch:batch + batch_size]

            # wrap them in Variable
            images_batch, labels_batch = Variable(images_batch), Variable(labels_batch)

            # zero the parameter gradients
            optimizer.zero_grad()

            # forward + backward + optimize
            outputs = CNN(images_batch)
            loss = criterion(outputs, labels_batch)
            loss.backward()
            optimizer.step()

            # print statistics
            running_loss += loss.data[0]
            if i % 2000 == 1999:    # print every 2000 mini-batches
                print('[%d, %5d] loss: %.3f' %
                      (epoch + 1, i + 1, running_loss / 2000))
                running_loss = 0.0

    print('Finished Training')

这里发生错误

# wrap them in Variable
images_batch, labels_batch = Variable(images_batch), Variable(labels_batch)

Answer 1

如果我的猜测是正确的，您可能会在以下行中收到错误。

# wrap them in Variable
images_batch, labels_batch = Variable(images_batch), Variable(labels_batch)

这意味着，images_batch和/或labels_batch是列表。您可以简单地将它们转换为 numpy 数组，然后转换为张量，如下所示。

# wrap them in Variable
images_batch = torch.from_numpy(numpy.array(images_batch))
labels_batch = torch.from_numpy(numpy.array(labels_batch))

应该能解决你的问题。

<小时/>

编辑:如果您在运行上述代码片段时遇到以下错误:

"RuntimeError: can't convert a given np.ndarray to a tensor - it has an invalid type. The only supported types are: double, float, int64, int32, and uint8."

您可以通过指定数据类型来创建 numpy 数组。例如，

images_batch = torch.from_numpy(numpy.array(images_batch, dtype='int32'))

我假设images_batch包含图像的像素信息，所以我使用int32。欲了解更多信息，请参阅official documentation .