Hi,
in 3.3 - Train the Model there is a description of the NN:
Implements a three-layer tensorflow neural network: LINEAR->RELU->LINEAR->RELU->LINEAR->SOFTMAX.
I can’t find the SOFTMAX layer in the assignment. where is it?
Hi,
in 3.3 - Train the Model there is a description of the NN:
Implements a three-layer tensorflow neural network: LINEAR->RELU->LINEAR->RELU->LINEAR->SOFTMAX.
I can’t find the SOFTMAX layer in the assignment. where is it?
That’s a good point. It turns out that it is not there explicitly, but it’s there implicitly: we need to tell the loss function to calculate it for us. That turns out to be a better way to handle this case. Here’s a thread which discusses why that is. And here’s another version of that same discussion which maybe goes one level deeper.
Thank you.
I got it.
Maybe it’s good to note this in the assignment.
Yes, they don’t actually discuss this in the assignment instructions. I forget whether Prof Ng mentions this in the lectures or not, but it would be a good idea to explicitly mention it in the notebook. I’ll file a git issue about that.
Update: I couldn’t find any mention of this idea in the lectures either, so I went ahead and filed the enhancement request. Note that Prof Ng will do it this way from this point forward as we switch to using TF for most of our work in C4 and C5.