W2_A2_Gradient Descent

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

What’s the problem here? Can you please write the code for dw and db?

{moderator edit: removed assignment code}

2

I dont remember the formulas if they are rights but I would suggest you to you numpy transpose and numpy dot instead of python operators.

3

Hi, did you mean to use the following code ? Well, it’s not working.
dw = (np.dot(X, np.transpose(A-Y)))/m

Web capture_4-11-2022_940_www.coursera.org
Web capture_4-11-2022_940_www.coursera.org866×449 53.3 KB

4

The syntax error in your first post was probably because the line that computed cost had imbalanced parentheses.

Btw, which course or specialization does this assignment belong to? You posted this in the Machine Learning Specialization category but I can move it to the right specialization for you if you tell me the name of it.

Cheers,
Raymond

5

@rmwkwok, hi thank you for your reply. The specialization is Neural Networks and Deep Learning. from DeepLearning.Ai. This is a Coursera Course.

6

I moved it to the DLS category for you. Please post your question here next time.

Good luck with the assignments!

Raymond

7

@rmwkwok @gent.spah thank you fro your answer. I am stuck in this assignment.

Please see the error for “Cost”
### Exercise 5 - propagate;

Excercise 5
Excercise 5900×716 31.7 KB

8

np.dot is matrix multiply (dot product style, not “elementwise”). So it does not work to multiply a 1 x 3 matrix (vector) with a 1 x 3 matrix (vector), right? If you don’t know how matrix multiply works, you are in trouble here. A knowledge of basic Linear Algebra is a prerequisite for this course. There are two ways to implement the two terms of the cost there:

  1. You can use “elementwise” multiply of the two vectors (* or np.multiply) and then you can sum the results with np.sum.
  2. You can do both the multiply and the sum in one operation by using “real” matrix multiply (np.dot). But to get the dimensions to work right, you have to transpose the second operand. If you dot 1 x 3 with 3 x 1, you get 1 x 1, which is a scalar and that’s what we need here, right?

If you use option 2), then you don’t need that np.sum that you show in your code.

Here’s a thread which talks about this and gives some examples.

9

@paulinpaloalto thank you for your reply. Both way, it is not working. If I use np.multiply ( ), the result does not match. Do you mind talking with in private message?

10

Please check your DMs …