Course 4, Week 1, Residual Network assignment , Identity block

Course Q&A Deep Learning Specialization Convolutional Neural Networks
1

image
image1162×621 27 KB

I don’t know what is wrong with the code. I checked it several times, yet it gives me the same error.

2

Here are the results I get for that test cell:

With training=False

[[[  0.        0.        0.        0.     ]
  [  0.        0.        0.        0.     ]]

 [[192.71234 192.71234 192.71234  96.85617]
  [ 96.85617  96.85617  96.85617  48.92808]]

 [[578.1371  578.1371  578.1371  290.5685 ]
  [290.5685  290.5685  290.5685  146.78426]]]
96.85617

With training=True

[[[0.      0.      0.      0.     ]
  [0.      0.      0.      0.     ]]

 [[0.40739 0.40739 0.40739 0.40739]
  [0.40739 0.40739 0.40739 0.40739]]

 [[4.99991 4.99991 4.99991 3.25948]
  [3.25948 3.25948 3.25948 2.40739]]]

So you can see that the training = False values agree with yours, but they don’t in the True case. So that should be a pretty good clue where to look.

The only place that the training variable has any effect is the BatchNorm calls. They give you one example in the template code:

X = BatchNormalization(axis = 3)(X, training = training) # Default axis

Your calls to BatchNorm should look the same as that example. Do they?

3

The test passed . But I dont see the reason why training=false or true should be the case for every batch normalisation statement. Isnt one enough?

4

Apparently not. Each instance of batch normalization is independent. The coefficients are specific to that instance.