Assert all(tensors[0].size(0) == tensor.size(0) for tensor in tensors) AttributeError: 'list' object has no attribute 'size'

[terminator9487]

Hello， I am new in using pytorch to build NN. Recently, I am trying to formalize my data using data_utils.TensorDataset. However, this following error was appeared. I do wrap the data with tensor in my code and put it into the Variable. Unfortunately, it didn’t work at all. Are you familiar with this problem？ I really appreciate for yr comment. Here is the code and error information.

y_vals_Valence = np.array([0 if each==‘Neg’ else 1 if each ==‘Neu’ else 2 for each in data[‘all_data’][:,320]])
y_vals_Arousal = np.array([3 if each==‘Pas’ else 4 if each ==‘Neu’ else 5 for each in data[‘all_data’][:,321]])

DEAP_x_train = x_vals[:-256] #using 80% of whole data for training
DEAP_x_train_torch = DEAP_x_train.astype(float)
DEAP_x_train_torch = torch.from_numpy(DEAP_x_train_torch).double()

DEAP_x_test = x_vals[-256:] #using 20% of whole data for testing
DEAP_x_test_torch = DEAP_x_test.astype(float)
DEAP_x_test_torch = torch.from_numpy(DEAP_x_test_torch).double()

DEAP_y_train = y_vals_Valence[:-256]##Valence
DEAP_y_train_torch = DEAP_y_train.astype(float)
DEAP_y_train_torch = torch.from_numpy(DEAP_y_train_torch).double()

DEAP_y_test = y_vals_Valence[-256:]
DEAP_y_test_torch = DEAP_y_test.astype(float)
DEAP_y_test_torch = torch.from_numpy(DEAP_y_test_torch).double()

DEAP_y_train_A = y_vals_Arousal[:-256] ### Arousal
DEAP_y_train_torch_A = DEAP_y_train_A.astype(float)
DEAP_y_train_torch_A = torch.from_numpy(DEAP_y_train_torch_A).double()

DEAP_y_test_A = y_vals_Arousal[-256:]
DEAP_y_test_torch_A = DEAP_y_test_A.astype(float)
DEAP_y_test_torch_A = torch.from_numpy(DEAP_y_test_torch_A).double()

batch_size =50;

Data Loader (Input Pipeline)

train = data_utils.TensorDataset(DEAP_x_train_torch, [DEAP_y_train_torch]) ##()

train_loader = data_utils.DataLoader(train, batch_size=batch_size, shuffle=True)

test = data_utils.TensorDataset(DEAP_x_test_torch, [DEAP_y_test_torch])
test_loader = data_utils.DataLoader(test, batch_size=batch_size, shuffle=False)

[ptrblck]

Try to remove the bracket enclosing the target tensor.

[terminator9487]

Thanks for yr comment. But, I couldn’t follow u because I am new in this field. what’s exactly the bracket you mentioned？the code that transfer the numpy to torch or？

[ptrblck]

Sorry, I thought I’ve quoted the line of code.
Here it is:

train = data_utils.TensorDataset(DEAP_x_train_torch, [DEAP_y_train_torch]) 

While you are passing the data tensor directly, you enclose the target tensor with brackets.
Could you remove these brackets and try it again?

[terminator9487]

Sure, I see. the reason i used the brackets here is my labels are 2-D. the corresponding information is in this link. python - PyTorch: How to use DataLoaders for custom Datasets - Stack Overflow For now, I was just testing only one label. Anyway, I removed the brackets and tried again. the error is like this:

[ptrblck]

I guess your data is of type torch.DoubleTensor.
If you need double precision, you should cast your model using model = model.double().
The common case is to use float so you could try to cast your data using

train = data_utils.TensorDataset(DEAP_x_train_torch.float(), DEAP_y_train_torch)

[terminator9487]

yes, i couldn’t agree more. and, I wanna use double precision. this error occurred after adding model = model.double(）.

[terminator9487]

So I guess the torch should be changed to long type. another error occurred after adding train = data_utils.TensorDataset(DEAP_x_train_torch.long(), DEAP_y_train_torch.long（）)

Now, I am confused. Could you tell me where is wrong？Thanks a lot!

[ptrblck]

nn.NLLLoss and nn.CrossEntropyLoss both require torch.LongTensors as the target input.
Cast your targets to .long() and try it again.

[terminator9487]

sorry, I couldn’t understand it. I have already cast my input feature and label by train = data_utils.TensorDataset(DEAP_x_train_torch.long(), DEAP_y_train_torch.long（）) but it’s still wrong. Did I get correct understanding what you mentioned before?

[ptrblck]

Sorry for the confusion. I didn’t explained your last error properly.

Your training data should be of type float, while your target should be long.
Could you change your dataset to

 train = data_utils.TensorDataset(DEAP_x_train_torch.float(), DEAP_y_train_torch.long())

and run it again?

[terminator9487]

Thanks for yr patience. I thought this feature data should have the same type with the label, right？ Anyway, I just tried it again. However, it still didn’t work. the error is as follow：

[ptrblck]

No worries.
I forgot you wanted to use your model in double precision.
Try the following:

train = data_utils.TensorDataset(DEAP_x_train_torch.double(), DEAP_y_train_torch.long())
model = model.double()

No, the types of data and target depend on the loss function you are using.
In a classification use case you usually use floats for your data and long for the target.
You can find more information in the docs.

[terminator9487]

Thanks for yr replying. It finally works. But, it still existed the problem. How come the accuracy is so low when using Multi-layer perceptron？ here is the code and result. Do you have any idea the reason of that? BTW， I was also using the SVM for classification model. The result is about 45%.

class Net(nn.Module):
    def __init__(self):
    super(Net, self).__init__()
    self.l1 = nn.Linear(320, 512)
    self.l2 = nn.Linear(512, 1024)
    self.l3 = nn.Linear(1024, 512)
    self.l4 = nn.Linear(512, 512)
    self.l5 = nn.Linear(512, 3)

def forward(self, x):
    x = F.relu(self.l1(x))
    x = F.relu(self.l2(x))
    x = F.relu(self.l3(x))
    x = F.relu(self.l4(x))
    return F.log_softmax(self.l5(x), dim=1)
    #return self.l5(x)

    model = Net()
    model = model.double()

   optimizer = optim.SGD(model.parameters(), lr=0.0001, momentum=0.5)

[ptrblck]

What kind of data do you have and what is your expected accuracy?

[terminator9487]

the feature is extracting from DEAP data set using HOC，and the label is also in this dataset. the expected accuracy is about 60%.

[Ahmad_Altameemi]

Hello ,can someone helps me ,when I run my code I got this error
AttributeError: ‘list’ object has no attribute ‘size’
#rpm
for epoch in range(300):
for x,y in enumerate(train_loader):
y = torch.zeros(y.size(0),4).scatter_(1,y.view(-1,1),1.)
x,y = x.to(device),y.to(device)
output,_ = rpm_nn(x,y) # cnn output
loss = loss_func(output) # cross entropy loss# clear gradients for this training step
optimizer_rpm.zero_grad()
loss.backward()
optimizer_rpm.step()
test_out,_=rpm_nn(test_x,test_y)
pred_z=torch.max(test_out,1)[1].cpu().data.numpy()
accuracy = float((pred_z == test_z.cpu().numpy()).astype(int).sum()) / float(test_y.size(0))
print('Epoch: ', epoch, ‘| train loss: %.4f’ % loss.cpu().data.numpy(), ‘| test accuracy: %.2f’ % accuracy)