Week 1 Assignment 1 Exercise 3 : len-1? L+1?

Course Q&A Deep Learning Specialization Improving Deep Neural Networks: Hyperparameter tun
1

Hi, I am confused about the He initialization code in the assignment, in exercise 3.

L is set to len(layers_dims) minus 1, but then the for loop goes to L+1

What’s the point of subtracting the length by 1 and then extending the loop range by one. Makes no difference (unless I am missing something). I get that the He adjustment is only supposed to go to the RelU layers, but the code provided doesn’t do that. It does all layers.

2

Hi @Yousif, welcome to Discourse.

Note that the iteration starts at 1, so I think is just to make the code consistent with the notation used in the course material. In this way you can initialize the weights inside the loop going from layer l-1 to layer l.

I hope that helps.

3

Thanks for the welcome and reply @kampamocha but, respectfully, I think you missed my point.

My point is that the len-1 and L+1 are totally pointless together in the code. You can edit the code to remove the -1 and +1 and it will work the same. It just adds a layer of confusion.

If the code really wanted to reflect the course material then there shouldn’t be a L+1 in the for-loop range, to exclude the last layer (usually a sigmoid)

Do you see what I mean?

Thanks

4

Hi @Yousif, I think now I understand what you mean, and looking at the previous exercises I agree that the inconsistency can be confusing. Thank you for pointing this out.