Hi,

I’ve read all the post i could find but I couldn’t find any help for this so I am forced to post it here.

I am having an issue with the function gradient_check.

Can you help me? I have double checked all the computations and I am quite sure that the difference is computed correctly. At this point I am having doubts on how to compute grad and J_plus/J_minus

Thanks in advance!

N.

1 Like

To compute grad, you have to call `backward_propagation`

function. And, for `J_plus/J_minus`

, you have to call `forward_propagation`

function.

Are you getting any errors? If so, please share that.

Hi saif!,

Yes I did that, indeed I am getting no errors! I am not sure how to deal with it.

The only message I get is

There is a mistake in the backward propagation! difference = 1.0

Maybe i can send you my chunk of code in private?

1 Like

One common mistake students make in this exercise is that they copy a code from `theta_plus`

and paste it to `theta_minus`

without changing the sign. If you are not making this mistake, you can send me your code in a private message. Click my name and message.

PS: You are doing exercise 3, `gradient_check`

; not exercise 4, `gradient_check_n`

, right?

Here is my output for Ex. 3:

```
theta_plus: 4.0000001
theta_minus: 3.9999999
J_plus: 12.0000003
J_minus: 11.9999997
gradapprox: 2.9999999995311555
grad: 3
numerator: 4.688445187639445e-10
denominator: 5.9999999995311555
difference: 7.814075313343006e-11
Your backward propagation works perfectly fine! difference = 7.814075313343006e-11
```

You may compare your result with it and see where you are making mistake.

Update:

The mistake was in computing `grad`

. The correct way is to call `backward_propagation`

only.

Thanks! All solved. I just got confused and thought that grad was alrady checking the difference while that is done when actually computing difference, whereas grad is a variable to which back_prop is assigned!

Thank you!

1 Like