Problem writing the dropout logic in course2-W1-exercise on regularization

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

The model defined with layers_dims = [X.shape[0], 20, 3, 1]. So, the dimension of A1=(20,m) but it doesn’t pass the test (as copied below). But passes all tests if I change (20,m) to (2,m).

in forward_propagation_with_dropout(X, parameters, …
42 D1 = np.random.rand(20,m)
43 D1 = (D1<keep_prob).astype(int)
—> 44 A1 = A1*D1
45 A1 = A1/keep_prob
46 # YOUR CODE ENDS HERE

ValueError: operands could not be broadcast together with shapes (2,5) (20,5)

Any help would be appreciated.
Thanks!

2

It is a mistake to “hardcode” the dimensions on the np.random.rand call. The better idea is to use the “shape” attribute of the relevant A matrix (A1 in the case you are showing).

BTW this is a bug in your code, not a bug in the tests or the grader, so I will take the liberty of editing the title of your post.

3

It worked by using shape. Thank you so much for your help.

1 Like
4

I haven’t quite figured out the correct arguments to pass into np.random.rand()… tried a few different approaches and still getting below error. What am I missing?

ValueError Traceback (most recent call last)
in
1 t_X, parameters = forward_propagation_with_dropout_test_case()
2
----> 3 A3, cache = forward_propagation_with_dropout(t_X, parameters, keep_prob=0.7)
4 print ("A3 = " + str(A3))
5

in forward_propagation_with_dropout(X, parameters, keep_prob)
42 D1=np.random.rand(2,A1.shape[0])
43 D1= (D1 < keep_prob).astype(int)
—> 44 A1=A1*D1
45 A1=A1/keep_prob
46

ValueError: operands could not be broadcast together with shapes (2,5) (2,2)

5

As D1 is expected to have the same shape of A1, you need to pass the size of A1 as parameters.
If you pass (2, A1.shape[0]), then, the shape of output is (2, 2). On the other hand, the shape of A1 is (2,5). Where “2” comes from ? You can get dimension of A1 as A1.shape, and extract each size like A1.shape[0], A1.shape[1].

1 Like