Problem in the code given for calculation of dz2 and dz1

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

In the programming assignments these are the codes given to us for dz2 and dz1. what is the role of np.int64(A2>0). To me it seems this is the value of g’(z). is it . how it is?

dZ2 = np.multiply(dA2, np.int64(A2 > 0))

dZ1 = np.multiply(dA1, np.int64(A1 > 0))
please help

2

You’re right: that is the derivative of ReLU. Think about it for a second and that should become clear. It is 1 if a > 0 and 0 otherwise, right?

Note that the code here looks quite a bit different than the fully general code that we wrote in Week 4 of Course 1 for the L layer case. Here they just “hard-code” everything to keep the code simple, so they don’t call a function for the derivative of the activations that is parameterized with the function being used: they just literally write it out.

1 Like