Is this a mistake in the Dropout Regularization video of Week 1?

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

At 4:10 Andrew writes;

a3 = np.multiply(a3, d3)

however, shouldn’t this be;

a3 = np.multiply(a3, d3.T)

as when multiplying 2 matrices a and b, the number of columns in a should equal the number of rows in b.

1 Like
2

Just checked it out in Python and it works. so I guess Python sorts out the shape of each matrix automatically before perofrming the computation.

3

Dear @ai_is_cool,

Yes you are right. Can you please provide the dimensions of a3 and d3?

4

Hi @Girijesh,

In the video both a3 and d3 have the same dimensions. But I have done some testing in Python3 and for any two 2-dimensional matrices;

a and b

np.dot(a, b)

equals;

np.matmul(a, b)

and

np.multiply(a, b)

is a differemnt operation and is the element-wise multiplication of;

a and b

which is of course a different matrix result from the previous two operations.

So, there is no mistake in Andrew’s lesson here, although there is a little bit of unfairness in his lectures as there is an implied assumption that the student should know NumPy.

5

Well, numpy was introduced in DLS Course 1 and we’ve been using it ever since. But you may be right that this might be the first time that he has explicitly used np.multiply. Normally he uses * to mean “elementwise multiply”. Here’s a thread about his notation for the two completely different types of matrix multiplication.