Course 2, Week 3, compute_cost

Course Q&A Deep Learning Specialization Improving Deep Neural Networks: Hyperparameter tun
1 
cost = tf.reduce_mean(tf.keras.losses.categorical_crossentropy(y_true = labels,y_pred = logits,from_logits=True))

I am trying this code but it keeps showing

AssertionError: Test does not match. Did you get the mean of your cost functions?

can someone spot the error and guide me.
Thanks in advance.

2

Did you transpose the labels and logits first? The instructions tell you what the shapes need to be, which implies a transpose.

2 Likes
3

I did a transpose and still got the same error. Please what am I doing wrong?

{moderator edit - solution code removed}

cost -> tf.Tensor(0.4981846, shape=(), dtype=float32)

Then I got this error:

AssertionError: Test does not match. Did you get the reduce sum of your cost functions?

4

What you need there are transpose operations, not reshapes. Those are not the same thing, even if the shapes end up the same. Try it on a simple example and watch what happens. Transpose is a very specific mathematical operation.

1 Like
5

Hi, please, Iā€™m having a similar problem. I did tf transposes on labels and logits, then crossentropy and then reduce sum, but I get tf.Tensor(1.6142862, shape=(), dtype=float32) and the same message: AssertionError: Test does not match. Did you get the reduce sum of your cost functions?

6

[SOLVED] I tried with different values for the two extra parameters of crossentropy and it worked with:

from_logits=True