IndexError：维度超出范围（预期在 [-1, 0] 范围内，但得到 1）

Question

As I've read through some previous questions, I'm getting this error that probably has something to do with a discrepancy between dimensions of tensors but since this is my very first attempt of running PyTorch so I'm coming here because I have far to little intuition about it. Wanted to run a non-standard dataset (which I'm pretty sure I'm loading fine) on a basic MNIST setup to mess with it and see what moves what.

Traceback (most recent call last):

File "C:/Users/Administrator/Desktop/pytong/proj/pytorch_cnnv2.py", line 110, in <module>
loss = loss_func(output, b_y)
File "C:\Users\Administrator\AppData\Local\Programs\Python\Python36\lib\site-packages\torch\nn\modules\module.py", line 547, in __call__
result = self.forward(*input, **kwargs)
File "C:\Users\Administrator\AppData\Local\Programs\Python\Python36\lib\site-packages\torch\nn\modules\loss.py", line 916, in forward
ignore_index=self.ignore_index, reduction=self.reduction)
File "C:\Users\Administrator\AppData\Local\Programs\Python\Python36\lib\site-packages\torch\nn\functional.py", line 1995, in cross_entropy
return nll_loss(log_softmax(input, 1), target, weight, None, ignore_index, None, reduction)
File "C:\Users\Administrator\AppData\Local\Programs\Python\Python36\lib\site-packages\torch\nn\functional.py", line 1316, in log_softmax
ret = input.log_softmax(dim)

IndexError: Dimension out of range (expected to be in range of [-1, 0], but got 1)

The line making the fuss is the loss function:

loss = loss_func(output, b_y)

The rest of the code, sans imports and load:

class CNNModel(nn.Module):
def __init__(self):
    super(CNNModel, self).__init__()

    # Convolution 1
    self.cnn1 = nn.Conv2d(in_channels=3, out_channels=16, kernel_size=5, stride=1, padding=0)
    self.relu1 = nn.ReLU()

    # Max pool 1
    self.maxpool1 = nn.MaxPool2d(kernel_size=2)

    # Convolution 2
    self.cnn2 = nn.Conv2d(in_channels=16, out_channels=32, kernel_size=5, stride=1, padding=0)
    self.relu2 = nn.ReLU()

    # Max pool 2
    self.maxpool2 = nn.MaxPool2d(kernel_size=2)

    # Fully connected
    self.fc1 = nn.Linear(338 * 4 * 4, 5)

def forward(self, x):
    # Convolution 1
    out = self.cnn1(x)
    print(out.shape)
    out = self.relu1(out)
    print(out.shape)

    # Max pool 1
    out = self.maxpool1(out)
    print(out.shape)

    # Convolution 2
    out = self.cnn2(out)
    print(out.shape)
    out = self.relu2(out)
    print(out.shape)

    # Max pool 2
    out = self.maxpool2(out)
    print('++++++++++++++ out')
    print(out.shape)
    # out = out.reshape(-1, 169 * 4 * 4)
    out = out.view(out.size(0), -1)
    print(out.shape)
    print('-----------------------')
    # Linear function (readout)
    out = self.fc1(out)
    print(out.shape)
    print('=======================')

    return out



if __name__ == '__main__':

print("Number of train samples: ", len(train_data))
print("Number of test samples: ", len(test_data))
print("Detected Classes are: ", train_data.class_to_idx) # classes are detected by folder structure

model = CNNModel()
optimizer = torch.optim.Adam(model.parameters(), lr=LEARNING_RATE)
loss_func = nn.CrossEntropyLoss()

# Training and Testing
for epoch in range(EPOCHS):
    print(enumerate(train_data_loader))
    for step, (x, y) in enumerate(train_data_loader):
        b_x = Variable(x)   # batch x (image)
        b_y = Variable(y)   # batch y (target)
        # print('============ b_x')
        # print(len(b_x))
        # print(b_x.data)
        # print('============ b_y')
        # print(len(b_y))
        # print(b_y.data)

        output = model(b_x)[0]
        loss = loss_func(output, b_y)
        optimizer.zero_grad()
        loss.backward()
        optimizer.step()

        if step % 50 == 0:
            test_x = Variable(test_data_loader)
            test_output, last_layer = model(test_x)
            pred_y = torch.max(test_output, 1)[1].data.squeeze()
            accuracy = sum(pred_y == test_y) / float(test_y.size(0))
            print('Epoch: ', epoch, '| train loss: %.4f' % loss.data[0], '| test accuracy: %.2f' % accuracy)

Plus the output of my prints that I've tried to use for diagnostics:

torch.Size([100, 16, 60, 60])
torch.Size([100, 16, 60, 60])
torch.Size([100, 16, 30, 30])
torch.Size([100, 32, 26, 26])
torch.Size([100, 32, 26, 26])
++++++++++++++ out
torch.Size([100, 32, 13, 13])
torch.Size([100, 5408])
-----------------------
torch.Size([100, 5])
=======================

Answer 1

The problem is in this line:

output = model(b_x)[0]

The [0] changes the shape from [100, 5] to [5] , and the loss expects the other way. Just remove it:

output = model(b_x)

Answer 2

Its most probably a mismatch between the output and truth label format. You would need to include more details like loss function used, the truth label format etc. in order to be able to debug this