Testing of "conv_backward" in week 1 assignment 1

Course Q&A Deep Learning Specialization Convolutional Neural Networks
1

I’m trying to solve this issue for like hours.

{moderator edit - solution code removed}

it gives me the right results

dA_mean = 1.457943469708148
dW_mean = 1.7519334429380524
db_mean = 7.882856704391539

but it gave me this error and I can’t find why

---------------------------------------------------------------------------
AssertionError                            Traceback (most recent call last)
<ipython-input-66-3575597a0654> in <module>
     21 assert dA.shape == (10, 4, 4, 3), f"Wrong shape for dA  {dA.shape} != (10, 4, 4, 3)"
     22 assert dW.shape == (2, 2, 3, 8), f"Wrong shape for dW {dW.shape} != (2, 2, 3, 8)"
---> 23 assert db.shape == (1, 1, 1, 8), f"Wrong shape for db {db.shape} != (1, 1, 1, 8)"
     24 assert np.isclose(np.mean(dA), 1.4524377), "Wrong values for dA"
     25 assert np.isclose(np.mean(dW), 1.7269914), "Wrong values for dW"

AssertionError: Wrong shape for db (2, 2, 3, 8) != (1, 1, 1, 8)
2

Wrong shape for db.
db’s shape should be initialized as b.shape, np.zeros(b.shape), instead of using W shape.

3

Also the above line looks suspect. Python is a case sensitive language, right?

And why use np.random.randn to initialize the variables? I guess it does no harm, since you’re going to overwrite the values, but it’s also a waste of compute. np.zeros would make more sense.

4

image
image743×630 32.3 KB

Why dA mean value gives wrong ans?

5

My guess is that it is because your dA values are wrong. Seems like the operative theory :nerd_face: …

My “mean” values agree with the ones they show. Let’s print some of the contents of dA and see the correct values:

print(dA[0,0,:,:])
[[  0.41601058 -12.12787317  -8.33770594]
 [ -6.90397663  -1.56843937 -17.60125782]
 [ 15.84199049   3.11023817  12.90284339]
 [  3.01845418  20.73395875  -6.09838706]]
print(dA[1,3,:,:])
[[  0.19156246   9.19938463  -2.33883242]
 [ 18.18424878  11.89353277   2.22553913]
 [  9.57071875  -2.16532197 -11.8747392 ]
 [  4.05685555  11.95389449  -2.00952336]]

What do you see when you try that?

Note that they give you a lot of the moving parts of the computation here in the instructions. The math formulas can be a little hard to follow, but compare those and the rest of the instructions to what your code does.

Also note that way the stride interacts with vert_start and horiz_start is exactly the same here as it was in conv_forward. If you passed the tests for that function, make sure you used the same logic here.