Typo in 02_pytorch_classification.ipynb

[Caesar0714]

Hi there,

I just want to mention that there is a typo in the part under the question, I think the activation layer is usually put just before the output layer instead of after.

Best,
Caesar

[mrdbourke]

Hey @Caesar0714 ,

That's a great point!

However, the part about "just after the output layer" refers to activation functions such as the torch.softmax function being used after a model's output layer to convert output logits to prediction probabilities.

For example:

outputs = model(x)
prediction_probabilities = torch.softmax(outputs, dim=1)

But you are right, you can use non-linear activations throughout the neural network before the output layer as well.

[Caesar0714]

Oh, I see. I didn't realise it since the previous code cell puts the ReLU right before the output layer, so I felt a bit confused about the "just after the output layer" in the later content. But anyway, thanks Daniel!