It doesn't need 'deep copy' or even 'shallow copy' of `w_in`

Course Q&A Machine Learning Specialization Supervised ML: Regression and Classification
1

This is not a big problem. But I found unnecessary code in [C1_W2_Lab02_Multiple_Variable_Soln].

def gradient_descent(X, y, w_in, b_in, cost_function, gradient_function, alpha, num_iters): 
    # number of training examples
    m = len(X)
    
    # An array to store cost J and w's at each iteration primarily for graphing later
    J_history = []
    w = copy.deepcopy(w_in)  #avoid modifying global w within function
    b = b_in

We don’t need to deep-copy w_in above. w_in has just immutable items, float.
Actually, there is no need even to shallow-copy it. This function does not mutate w_in at all.
It’s enough to make its alias like below.

    w = w_in
    b = b_in
2

Thank you Jangsea. It is not needed. And welcome to our community!

Raymond

1 Like
3

Thanks for welcoming!

Jangsea

1 Like
4

Hi all,

Nor is it necessary to explicitly load dictionary values into temporary variables such as: dWaa, dWax, dWya, db, dby and then recreate the dictionary.

def clip2(gradients, maxValue):

for gradient in gradients:
np.clip(gradients[gradient], xxx, xxx, out = xxx])

return gradients

I noticed this revising lessons. I imagine they did this for some pedagogical purposes, it is not a python specialization after all.

Bye

5

Hello, @Francis60,

Maybe. I am not sure because gradient clipping is not covered by MLS to which this post belongs, so I do not know which implementation you were referring to.

Cheers,
Raymond

6

I was referring to Dinosaurus_Island_Character_level_language_model.ipynb which doesn’t need deepcopy as well. This old post from jnsp was suggested as I attempted editing a new one. I should have created a specific topic not to confuse readers about a dictionary oversomething usage nonexistent in Lab02.

Cheers

7

Hello, @Francis60,

It’s alright! With the assignment name, I can find it in DLS C5 W1 A2.

For that one, we do not need a copy for optimize. However, we do need a copy for clip_test as it needs both the original values and clipped values, otherwise, the test probably is not going to be useful. There is alternative to keep clip_test useful while leaving deepcopy out of clip, but that does not really matter, because assignment is one thing, how you build your own code is another :wink: . The key is you know and we know what is going on.

Cheers,
Raymond

8

Thank you @rmwkwok to share your deep understanding. Yes, they could have use the deepcopy just inside clip_test.

As somebody said:

1 Like
9

You are welcome!