RuntimeError: Expected object of scalar type Long but got scalar type Float for argument #3 'mat2' in call to _th_addmm_out

1

Hi everyone, I am getting this error and I am not sure what is happening. I am trying to predict air passenger data with a LTSM. The notebook is available on this link:

https://drive.google.com/file/d/1lk9uKhjbMlLHBvvqmaIuBFvbVfoKAia7/view?usp=sharing

Thanks in advance to anyone willing to help.

2

Have you tried just converting the seq variable to a float in your training loop like this

    for seq, y_train in train_data:
        
        seq = seq.float()
        optimizer.zero_grad()
        model.hidden = (torch.zeros(1,1,model.hidden_size),
                        torch.zeros(1,1,model.hidden_size))

if that doesn’t work can you print out the type of seq and the hidden states.

3

Hello,

This is just because LSTM expects Float in input, and your data is Long. Use this code

for seq, y_train in train_data:
    seq = seq.type(torch.float32)
    optimizer.zero_grad()
    y_pred = model(seq)
    loss = criterion(y_pred, y_train)
    loss.backward()
    optimizer.step()
4

Thank you very much @omarfoq and @Dwight_Foster but I have tried both options getting a confusing message. You can see the updated notebook here:
https://drive.google.com/file/d/1VKpZCkSeEJoP_iyXAi0xBZRlPqE8B1fE/view?usp=sharing

How can I proceed?

5

So now it seems like your loss is the problem. Try this:

        
        loss = criterion(y_pred, y_train).float()
        loss.backward()
        optimizer.step()
6

It throws the same error. It is very strange.