Hi,
I am a little confused by the first calculation in this exercise. Applying the sigmoid function in Course 1 Week 2 exercise 5
We are given (in the calling module) :
w = a 2 by 1 np.array and X, a 2 by 3 np.array.
We are to transpose w using the .T function and this immediately makes the two arrays incompatible for broadcasting.
Am I reading this wrongly? When I run the code with w.T it fails.
When I run it without the .T it appears to work ( I haven’t got to the second calc yet!)
Thanks
Ian
It sounds like the problem is that you are using “elementwise” multiply where you should be using a dot product. The formula for the “linear activation” (the first step before you apply sigmoid) is this:
Z = w^T \cdot X + b
So the operation between w^T and X is a dot product. w is n_x x 1 and X is n_x x m so that will work. 1 x n_x dotted with n_x x m gives a 1 x m result, right?
It is important to realize the notational conventions that Prof Ng uses. Here’s thread about when to use * versus dot product.
Thank you Paul.
I spent a few hours trying to decipher the term ‘activation’ and assumed I was being asked to apply the sigmoid function. Clearing that up + my confusion over dot product vs elementwise multiplication should set me on the right track.
Regards
Ian
Great! Of course you do apply the sigmoid, but only after doing the “linear” activation. Logistic Regression is the model for how all layers in Neural Networks will work when we get to that in Week 3: you first do a linear transformation, followed by the application of the non-linear activation function (sigmoid in this case, but we’ll see other such functions later).