Target 7 is out of bounds

1

outof bounds error
outof bounds error1150×439 22.1 KB

I uploaded the out of bounds error picture here.I used cross entropy loss and running my model on rafdb dataset which contain 7 labels.

` epoch = 0

# Train the model

total_step = len(train_loader)

curr_lr = learning_rate

for epoch in range(num_epochs):

    for i, (images, labels) in enumerate(train_loader):

        #images = images.to(device)

        #labels = labels.to(device)

        print(labels)

        print(labels.size())

        # Forward pass

        #print(images.size)

        outputs = model(images)

        #print(outputs)

        print(outputs.size())

        loss = criterion(outputs, labels)

        print(loss)

        # Backward and optimize

        optimizer.zero_grad()

        loss.backward()

        optimizer.step()`

It is showing error at loss function.Thanks in advance for the help.

2

Hi Jami!

CrossEntropyLoss expects class labels that run from
{0, 1, ..., nClass - 1} (not from 1 to nClass).

So in your case, 6 is the largest legal class label, and 7 is out
of bounds.

By the way, please don’t post screen-shots of textual information.
Doing so breaks screen readers (accessibility), search, and copy-paste.

Best.

K. Frank

3

Hi KFrank,
Thanks a lot for the reply,What to do then,my class labels are starting from 1 and i have class labels from 1 to 7.
And also i can’t see any out of bounds error picture here(which you mentioned you uploaded it).
I posted screen shot because it will be clear about the error.May be next time on wards i will post whole error once by copy pasting it.

4

Hi Jami!

Remember, class labels are really just arbitrary labels that distinguish
one class from another, but nothing else. They have no numerical
meaning.

So just subtract 1 from all of your labels to get a set that runs from
0 to 6. You could either do this by preprocessing your data, or you
could do it on the fly.

Sorry about that. I was quoting you (rather than referring to an upload
of my own), but I messed up the forum quote block. (I’ve edited my
post to fix the quoting and clean that up.)

Thanks, that would be great. Makes for a more functional forum.

Best.

K. Frank

5

Thanks a lot KFrank,Actually that idea of class labels from 0 to 6 worked for me.