How should we think when initializing w & b?

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

I seem to be having a problem while initializing the parameters w and b in the model function. Could you give any tips on how to think about their shapes? Shouldn’t we be just considering the equation transpose of x times w + b?

2

Hi, @Kutay_Eroglu . Welcome to the specialization. When creating a new topic in Discourse, it is a great help to the community if you post the week, assignment number, and Exercise number in your topic heading. Example: W3, A1, Ex 3: How should we think … .

Also, please post a snapshot the “traceback” (i.e. the error log) generated by your code. Do not post the code directly from the function that you are trying to complete. That would be a violation of the course honor code.

Thanks!

1 Like
3

Here is an example of another learner’s post just today:

Week 4 Exercise 5 - L_model_forward shape challenges

1 Like
4

Hi, @Kutay_Eroglu !

Yes, you are right for lineal layers. For convolutional layers, check this post.

If you are using other types of layers or more complex ones, the output size will difer.

5

I’m guessing this question is about the C1 W2 Logistic Regression assignment. There the linear activation formula is:

Z = w^T \cdot X + b

In that case b is a scalar value and w is a column vector with dimensions n_x x 1, where n_x is the number of “features” or elements in each input sample. Each column of X is one input sample, so n_x is the number of rows in the X matrix.

All this was discussed in the lectures and also in the notebook. Please read the material in the notebook again carefully if you still have questions.

1 Like