L_model_forward for loop

Course Q&A Deep Learning Specialization Neural Networks and Deep Learning
1

Hello,

This might be a bit of a silly question but i don’t understand why the for loop in L_model_forward is:
for l in range(1, L):

and not
for l in range(1, L-1):

Thanks for the help

2

Hi @SuitaX , welcome to this channel.
Regarding to your question, you can use the “initialize_parameters_deep()” comments as reference. Note that if I use (1,L) and I have only one layer, there is the explanation about what I get. By the other side, if I use (1, L-1), it means (1, 0) or the for loop will never be executed.

1 Like
3

Hello Carlosrl,

Thank you for your help !

So i quote the first “initialize_parameters_deep” instruction :

“The model’s structure is [LINEAR → RELU] × (L-1) → LINEAR → SIGMOID. I.e., it has 𝐿−1 layers using a ReLU activation function followed by an output layer with a sigmoid activation function.”

From what i understand if L=1 there is no relu layer and just the final sigmoid layer and thus i would find it normal for the for loop not to work.
I know i am wrong, i tried with “for l in range(1, L-1):” and it doesn’t work… i will give some more thoughts.

5

Hello @SuitaX !

I think the answer is related to python loops. The behavior of the loop is:

  1. range(n) means it generates numbers from 0 to n-1
  2. range(start_value, end_value) means, it generates numbers from start_value to end_value-1.

Here’s an example:

Screen Shot 2021-07-02 at 11.38.29 PM
Screen Shot 2021-07-02 at 11.38.29 PM882×604 59.9 KB

You can see that we never get to 6 in the above example. The loop terminates at 5.

So, in range(1, L), we are not calculating the activation for the final layer(L). The last activation that the loop calculates is A_L-1(activation of the second to last layer.) The next part of the code, outside the loop, calculates A_L (activation of the last layer.) Thus sigmoid is only applied the A_L, while the rest have ReLu as their activation function.

1 Like
6

hello @shwetank

That’s it… I was thinking from 1 to n.
Thank you, i really appreciate !

1 Like
7

Yes, indexing in python is all “0 based”, which includes both array indices and loop indices. Here’s the analog of the experiment that @shwetank showed us above for loops, but with array indices:

>>> v = np.array([1,2,3,4])
>>> len(v)
4
>>> v[1]
2
>>> v[0]
1
>>> v[3]
4
>>> v[4]
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
IndexError: index 4 is out of bounds for axis 0 with size 4
>>>
2 Likes
8

Thank you, this is helpful. I understand the behavior of the loop better now.