DLS Course 5, Week 1, assignment 3,

Course Q&A Deep Learning Specialization Sequence Models
1

Hello,

After running the unit test of the djmodel, I get the following error message:

Test failed
Expected value

[‘Reshape’, (None, 1, 90), 0]

does not match the input value:

[‘TensorFlowOpLayer’, [(None, 90)], 0]

Even though the following code in the djmodel function gives the correct shapes:
x = X[:,0,:]
print(x.shape)
(None, 90)
x = reshaper(x)
(None, 1, 90)

please advise.

Thank you
print(x.shape)

2

please ignore. problem fixed

3

Hi,
Glad that you solved the issue. Maybe it would be great if you share the error and the fix in case other students face the same issue :slight_smile:
Thanks and happy learning

Rosa

1 Like
5

Please share how you fixed this, I have exactly the same problem. I have tried to restart the Kernel as was suggested in another thread.

The problem seems to arise in the last part of the code where I use

model = Model(inputs=[X, a0, c0], outputs=outputs)

is this correct?

Problem solved. I went back and read the hint for part 2D of the exercise carefully.

6

How did you resolve it?

7

Can you please tell me how you fixed this error?

8

Read “Step 2.D” instruction carefully.

9

after reading 2D carefully still could not resolve it. any other hint will be appreciated

1 Like
12

Step 2.D tells you some important things, along with the documentation of tensorflow:
tf.math.argmax  |  TensorFlow Core v2.5.0 tells us there’s an “axis” parameter. there’s also a mention of “axis” in the 2D hints.
tf.one_hot  |  TensorFlow Core v2.5.0 also mentions a depth parameter that could be important for the final shapes

Hope this helps and is vague enough to count as a “hint”

13

I am getting the same error. How did you resolve the error?

14

Thanks for your post. However, under 2.D I have the following hint:

2D. Dense layer

  • Propagate the LSTM’s hidden state through a dense+softmax layer using densor .
15

The answer to look at 2D more carefully is for problem 3 - Generating Music, not problem 2 - Building the Model. My answer is for problem 2 - Building the Model.

I got this exact same error but it had to do with my answer for 2C. Within the LSTM_cell function,
I set the initial state to [a0, c0]. However, that will only work for the first value of Tx and never gets updated after that. The variables a and c were initialized to a0 and c0 before the for loop. Once I changed the initial state to the correct variables, it worked.

2 Likes
16

You need not use the Lambda function of keras. Instead just mention x = (lambda x: x[:,t,:])(X)

This should resolve it

17

Thanks coyne44. This helped me to fix my error. initial state should be initialized using a and c