Optional lab - Gradient Descent

[tsvika_greener]

Greeting everyone,
I could not find the definition (def) of gradient_fanction(x, y, w, b). Is it my bad sight or it is really missing from the notebook of this lab?
Thanks in advance, T.G.

[rmwkwok]

Hello @tsvika_greener, welcome to our community!

I suppose you are asking about the gradient_function variable used as the input argument here

if so, and since it is an input argument, we need to look at where the function gradient_descent is being called, which for example, can be this one:

and we see that compute_gradient is the name of the variable which content is assigned to be gradient_function, and following this lead, we can finally get the def line for compute_gradient by searching it in notebook.

Cheers,
Raymond

[tsvika_greener]

Hi @rmwkwok and thanks for your rapid response. Nevertheless, according to the comment of gradient_descent() function, gradient_function is a function gradient_function() - please see the attached screen captures. That is later on gradient_descent() function calls this function gradient_function() - - please see the attached screen captures.

Now I am completely confused since I could not find the definition of that function - gradient_function()
Could you please help?
Thanks in advance, T.G.

[rmwkwok]

Hi T.G., if we can focus on your last screenshot, gradient_function is an input argument when you call gradient_descent, is that right?

[tsvika_greener]

Hi @rmwkwok , yes you are right , but then again in the comments of gradient_descent() function in the Args section there is a line where its write that gradient_function is a function to produce gradient (I marked it with blue box in my last screenshot). In addition, the for loop (in my first screenshot) calls gradient_function(x, y, w, b) with the input arguments x, y, w, b (I also marked it with blue box in my first screenshot).
Thanks T.G.

[rmwkwok]

T.G., an input argument can be a value or a function, because both value and function are named by a variable name.

[tsvika_greener]

OK, so because functions in Python are objects we can pass them as arguments to another function - agree
So now back to my original question, where is the (definition) def gradient_function(x, y, w, b) written in the notebook of this lab? I did not find it
Thank you for your patience and understanding

[rmwkwok]

Hi, T.G., can you show here a screenshot of when the function gradient_descent is being called, and highlight the variable that’s passed into that call as the last argument? And do you know why this will help answer your question?

[tsvika_greener]

Sorry for the confusion, I meant gradient_function(x, y, w, b) as I have previously referred in my messages. I could not find where is the (declaration) def gradient_function(x, y, w, b) of this function in the notebook of this lab:
https://www.coursera.org/learn/machine-learning/ungradedLab/lE1al/optional-lab-gradient-descent
Also, please see the attached screenshots:

Thanks T.G.

[rmwkwok]

Hello T.G.,

This is not a confusion. There is a reason I asked you what variable is passed into the function gradient_descent, which is because it will lead you to the answer you want. If you don’t follow me, we can’t make any progress.

So far you have asked your question for 3 times, and you question is pretty clear. Now, please try to answer my following 4 questions, and do not repeat your question for the 4th times.

1. In the following scenario of only 4 lines of code, where do you find the declaration of the variable ‘name’?

student = 'sally'

def print_student_name(name):
    print(name)

print_student_name(student)

2. You ask about gradient_function, but I believe you are also aware that gradient_function is called inside another function named as gradient_descent, do you agree?

3. If you agree, gradient_function is not the only variable being used inside gradient_descent, do you agree?

4. If you agree, then, I would tell you, apart from gradient_function, the following are also used: cost_function, num_iters, alpha, w_in, and b_in, now, my question, can you pick any one of them, and tell me where it is declared?

T.G. @tsvika_greener, there is no confusion, so please don’t repeat your question again, and I ask you please move your attention away from your exact question for a just a little while, and try to answer my 4 questions so that we can make some progress. There is nothing I can do if you keep thinking there is a confusion, and just repeat your question over and over again.

If you don’t answer my questions, and just repeat your question, I am afraid I will have to leave your question to other mentors, so please try, ok?

Raymond

[tsvika_greener]

Dear @rmwkwok I appreciate your your patience and understanding very much.
Regarding to your questions:

1. Agreed
2. Agreed
3. Agreed
4. Yes, you made your point and I think that I got it - thanks
   So just last thing, from the implementation of Gradient Descent (screenshot is attached) it is written that gradient_descent(…) is utilizing compute_gradient(…) and comput_cost(…)
   
   
   
   I am curious to know if the argument cost_function in gradient_decent(…) is actually the function comput_cost(…) that was defined before then and the argument gradient_function in gradient_decent(…) is actually the function compute_gradient(…) that was defined before then?
   Thank you very much and I appreciate your efforts to clarify it to me. T.G.

[rmwkwok]

Hi T.G.

Yes, but I hope you know that you can verify that yourself by examining the code.

I am copying 2 lines of code here and you can find them in the lab.

1. def gradient_descent(x, y, w_in, b_in, alpha, num_iters, cost_function, gradient_function)
2. w_final, b_final, J_hist, p_hist = gradient_descent(x_train ,y_train, w_init, b_init, tmp_alpha,
   iterations, compute_cost, compute_gradient)

The first line is the line that defines gradient_descent, and the second is the line that calls the gradient_descent.

Now, we focus on the arguments:

1. x, y, w_in, b_in, alpha, num_iters, cost_function, gradient_function
2. x_train ,y_train, w_init, b_init, tmp_alpha, iterations, compute_cost, compute_gradient

Do you see that the numbers of arguments are the same? They are actually one-one corresponded. Knowing how to verify it yourself is more important for the future.

Raymond

[tsvika_greener]

Many thanks @rmwkwok

[rmwkwok]

You are welcome T.G.

Raymond