Very strange practical exercise

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

How does the “Planar data classification” exercise connected to the course videos?

In my view it is made of code which is not covered in the course, asking question not covered in the course. There was so much theory in the course - why don’t you make an exercise to train what’s was given in the lectures? How is it supposed to work?

2

Hey @Eduard_Cherednik,
Welcome to the community. If we take a look at the Week 3’s content only, we will find topics like:

  1. Neural Networks overview and representation
  2. Computing a neural network’s output
  3. Vectorizing across multiple examples
  4. Explanation for vectorized implementation
  5. Activation functions
  6. Why do you need non-linear activation functions
  7. Derivatives of activation functions
  8. Gradient descent for neural networks
  9. Backpropagation intuition
  10. Random Initialization

Now, if we take a look at the assignment’s exercises:
Exercises 1, 2, 3 and 4 resemble to what we studied in 1., 2., 3. and 4. sections of Week 3. Similarly, exercises 6 and 7 resemble to what we studied in 7., 8. and 9. sections of Week 3, and so on. And I believe that a lot of related code has been discussed in the lecture videos itself. However, still if there is something in the assignment that is tripping you off from your learning journey, the Discourse community will be more than happy to help you whenever you need with whatever you need.

Regards,
Elemento

3

Hi Elemnto, thanks for your effort. I’m afraid you’re not addressing the question. The entire exercise in week 3 is built upon sklearn.linear_model.LogisticRegressionCV() library - which is NOT covered in the class. I don’t doubt sklearn library is able to generate all the required outcome (and is very effective in doing so), but I would rather expect students to hit the road by calculating forward and backward propagation as it’s given in the lectures - from the very basic.

This might be very time-efficient approach from the course-making perspective (just throw a readymade library every time you bring new concepts in), but not the best way for your students to practice new material.

4

Hey @Eduard_Cherednik,
I am a little bit confused whether we are on the same page or not. In the week 3’s assignment, LogisticRegressionCV from the scikit-learn library is just shown for comparison. The students aren’t asked to implement any code related to scikit-learn, and in all the exercises, the students are required to implement code related to a neural network only. In fact, the students are required to implement forward propagation, backward propagation, etc from scratch only.

Is it possible that you might be referring to a different thing than the Week 3’s Assignment?

Regards,
Elemento