CODEWITHSUNDEEP
pythondeep-learningpytorch
Lintang Gilang Pratama
Lintang Gilang Pratama

Reputation: 97

how to solve this (Pytorch RuntimeError: 1D target tensor expected, multi-target not supported)

i am newbie in pytorch and deep learning

my data set 53502 x 58,

i have problem this my code

model = nn.Sequential(
    nn.Linear(58,64),
    nn.ReLU(),
    nn.Linear(64,32),
    nn.ReLU(),
    nn.Linear(32,16),
    nn.ReLU(),
    nn.Linear(16,2),
    nn.LogSoftmax(1)
)

criterion = nn.NLLLoss()
optimizer = optim.AdamW(model.parameters(), lr = 0.0001)

epoch = 500
train_cost, test_cost = [], []
for i in range(epoch):
    model.train()
    cost = 0
    for feature, target in trainloader:
        output = model(feature)          #feedforward
        loss = criterion(output, target) #loss
        loss.backward()                  #backprop
        
        optimizer.step()                 #update weight
        optimizer.zero_grad()            #zero grad
        
        cost += loss.item() * feature.shape[0]
    train_cost.append(cost / len(train_set))    
    
    with torch.no_grad():
        model.eval()
        cost = 0 
        for feature, target in testloader:
            output = model(feature)          #feedforward   
            loss = criterion(output, target) #loss

            cost += loss.item() * feature.shape
        test_cost.append(cost / len(test_set))                
    
    print(f'\repoch {i+1}/{epoch} | train_cost: {train_cost[-1]} | test_cost : {test_cost[-1]}', end = "")

and then i get problem like this

   2262                          .format(input.size(0), target.size(0)))
   2263     if dim == 2:
-> 2264         ret = torch._C._nn.nll_loss(input, target, weight, _Reduction.get_enum(reduction), ignore_index)
   2265     elif dim == 4:
   2266         ret = torch._C._nn.nll_loss2d(input, target, weight, _Reduction.get_enum(reduction), ignore_index)

RuntimeError: 1D target tensor expected, multi-target not supported

whats wrong? how to solve this problem? why this happend?

Thank you very much in advance!

Upvotes: 4

Views: 8692

Answers (2)

Hamzah Al-Qadasi
Hamzah Al-Qadasi

Reputation: 9786

I got the same error message and the reason is that the targets are like multi-D tensor more than 1D:

tensor([[0],
        [0],
        [0],
        ...,
        [9],
        [9],
        [9]], dtype=torch.int32)

and using torch.flatten(targets) solved my problem. The target now has 1D tensor shape:

tensor([0, 0, 0,  ..., 9, 9, 9], dtype=torch.int32)

Upvotes: 2

Theodor Peifer
Theodor Peifer

Reputation: 3496

When using NLLLoss the target tensor must contain the index representation of the labels and not one-hot. So for example:

I guess this is what your target looks like:

target = [0, 0, 1, 0]

Just convert it to just the number which is the index of the 1:

[0, 0, 1, 0] -> [2]
[1, 0, 0, 0] -> [0]
[0, 0, 0, 1] -> [3]

And then convert it to long tensor, ie:

target = [2]
target = torch.Tensor(target).type(torch.LongTensor)

It might be confusing, that your output is a tensor with the length of classes and your target is an number but that how it is.

You can check it out yourself here.

Upvotes: 7

Related Questions