Runtimeerror in train loop.

[GAuravY19]

This is my train loop:-

epochs = 10

for epoch in range(epochs):
    model1.train()
    
    y_pred = model1(X_train)
    y_pred_act = torch.argmax(torch.sigmoid(y_pred), dim = 1).float()
    
    loss = loss_fn(y_pred_act, Y_train)
    
    optimizer.zero_grad()
    
    loss.backward()
    
    optimizer.step()
    
    print("LOss", loss)

What should I do to handle this error.

RuntimeError: element 0 of tensors does not require grad and does not have a grad_fn

[sametucuncu]

The loss is the error between expected output of the model (Y_train) and model's prediction which is y_pred in your case (y_pred = model1(X_train)).

y_pred_act = torch.argmax(torch.sigmoid(y_pred), dim = 1).float() is calculating the label from predicted value of the model. It is not exact output of the model. So, you should use output of the model while calculating the loss. So loss should be:

loss = loss_fn(y_pred, Y_train)