Error in Programming Assignment: Optimization Methods

Course Q&A Deep Learning Specialization Improving Deep Neural Networks: Hyperparameter tun
1

It seems there is an error (or rather a sequence of errors) in the programming assignment of week two. I could easily pass the assignment by stupidly following the instructions, but it seem to be wrong nevertheless:

Formula (3) in exercise 4 says:
π‘£π‘‘π‘Š[𝑙]=π›½π‘£π‘‘π‘Š[𝑙]+(1βˆ’π›½)π‘‘π‘Š[𝑙]

  • If I interpret this as a mthematical equation, it would mean
    (1βˆ’π›½)π‘£π‘‘π‘Š[𝑙]=(1βˆ’π›½)π‘‘π‘Š[𝑙] , i.e. π‘£π‘‘π‘Š[𝑙]=π‘‘π‘Š[𝑙]
    or if I iterpret it as a program assignment, then since π‘‘π‘Š[𝑙] is initialised as zero:
    π‘£π‘‘π‘Š[𝑙]=(1βˆ’π›½)π‘‘π‘Š[𝑙]

The latter is what I’ve implemented and which passes the chcks, however the correct formula should read from the lecture:
π‘£π‘‘π‘Š[𝑙]=π›½π‘£π‘‘π‘Š[𝑙-1]+(1βˆ’π›½)π‘‘π‘Š[𝑙]

(L replaced by l-1 on the rhs).
This error seems to be essentially in all subsequent exercises.

Please advise/correct…

2

I also faced the same problem and have the same question!!
I checked Andrew’s lecture and I found that the formula given is in accordance with the β€œcorrect” answer in assignment.
I thought vdW2 should be somehow linked in vdW1 rather than itself, as what you thought.
I hope someone can answer this for me and I will check the previous questions people post and see if there is an answer

3

I think the issue is answered here:

4

Sorry if I generated confusion, there is no error. The average must be taken over β€œtime”/steps, not over different layers, I got the indices mixed up. In zhe code, the update is performed only once, hence no real average is seen, but if several steps are performed one gets an average. relating different layers makes no sense, even the shapes will not match