C4W4_Assignment MSE and MAE is too big

Course Q&A TensorFlow Developer Professional Sequences, Time Series and Prediction
1

Hi guys,
I’ve been stuck with this problem for a few days.
No matter I changed different optimizer, and model architecture, I still cannot slove this problem.

The optimizer I have tried:
optimizer = tf.keras.optimizers.Adam()
optimizer = tf.keras.optimizers.SGD(lr=1e-8, momentum=0.9)
optimizer = tf.keras.optimizers.SGD(lr=1e-5, momentum=0.9)

[code removed - moderator]

2

Please don’t post code in public. Sharing stacktrace is okay.
Try using MaxPooling1D layer after Conv1D to reduce the number of parameters and improve performance.
The kernel_size is usually set to 2 or 3. Setting to a much bigger value is likely to affect learning.

3

Hello , can you send me all your notebook ?

4

Sorry, I got it. Thanks for your advice!

5

Hi there, I have tried all the advice here and I can’t get it to work, my Mae is below two but my mse will not go below 6

1 Like
6

Please see this topic on general tips for designing a NN.

7

Sorry this was not helpful to me

8

Unfortunately, there’s not much of a shortcut for designing NNs.
Look at the ungraded labls for the week for further hints.
If you’ve taken deep learning specialization, that should help with error analysis as well.

9

I had the same issue and using Adam(on both places) fixed it.