Compute layer style cost error

Course Q&A Deep Learning Specialization Convolutional Neural Networks
6

It looks like you instrumented your gram_matrix function to print GA, so I added that along with some other instrumentation in compute_style_layer_cost and here’s what I see:

type(a_G) <class 'tensorflow.python.framework.ops.EagerTensor'>
m is 1
shape(a_S) [1 4 4 3]
shape(a_G) [1 4 4 3]
after reshape+T: shape(a_S) [ 3 16]
after reshape+T: shape(a_G) [ 3 16]
a_S [[ 2.6123514   6.3462267  -2.0568852   1.0848972  -0.34144378  5.9450154
  -0.68249106  6.6380787   4.405425    3.2136106  -0.88850987  8.216282
   1.1940846  -3.8442307   4.486284   -3.2315984 ]
 [-3.3520837   3.8470404  -3.1489944  -1.2055032  -3.17067    -1.7347562
  -3.1652112  -0.90944517  0.31713337  0.23800504 -0.10706711  0.6901974
   1.7071393   6.2297974  -2.2124014  -5.5964684 ]
 [ 0.74761856 -0.95714605 -4.0077353  -5.972679    5.036553    3.6944358
  -1.7189786   9.18924     2.566379    1.4399388  -1.7099016   3.6196625
  -0.9568796   2.8206615  -4.4811783   3.4338741 ]]
GA =
[[277.3386      1.9207706  84.06223  ]
 [  1.9207706 139.91864    -1.1867776]
 [ 84.06223    -1.1867776 244.99716  ]]
GA =
[[277.3386      1.9207706  84.06223  ]
 [  1.9207706 139.91864    -1.1867776]
 [ 84.06223    -1.1867776 244.99716  ]]
J_style_layer 0.0
type(a_G) <class 'tensorflow.python.framework.ops.EagerTensor'>
m is 1
shape(a_S) [1 4 4 3]
shape(a_G) [1 4 4 3]
after reshape+T: shape(a_S) [ 3 16]
after reshape+T: shape(a_G) [ 3 16]
a_S [[-3.404881   -2.5186315   1.3512313  -1.8821764  -0.39341784  5.434381
  -0.2787553   3.5750656  -2.616547    1.2345984   0.25895953 -2.933355
  -5.2402277   0.19823879  1.3253293  -0.41526246]
 [ 7.183007   -3.8986888   0.18695849 -1.5039697  -0.34587932  6.118635
   2.493302    9.585233    6.579515   -0.9685255  -0.5711074   2.555351
   0.36834985  7.452428    6.3523054   0.583993  ]
 [ 2.534576   -2.9244845  -1.2326248  -1.8601038   1.730303    0.91409665
   2.0111642  -2.3005989   5.8995004  -2.2799122  -1.6340904  -3.1489797
  -0.42677724  3.718691    5.739221   -2.0045857 ]]
GA =
[[112.29061    37.05909    -1.9554844]
 [ 37.05909   352.1803    116.86712  ]
 [ -1.9554844 116.86712   136.64961  ]]
GA =
[[277.3386      1.9207706  84.06223  ]
 [  1.9207706 139.91864    -1.1867776]
 [ 84.06223    -1.1867776 244.99716  ]]
J_style_layer 14.01648998260498
J_style_layer = tf.Tensor(14.01649, shape=(), dtype=float32)
All tests passed

You can see that the GA values I get are significantly different than yours. One possible mistake is to directly reshape to the final shape you want for a_S and a_G instead of including a transpose to preserve the channels dimension.

Here’s an earlier thread about this issue. And that thread points to this one, which demonstrates why just the direct reshape doesn’t work.