link to notebook: https://www.coursera.org/learn/natural-language-processing-tensorflow/programming/GkEhd/predicting-the-next-word
I am on final graded function : create_model
I cannot get accuracy > 75% . The hints are quite specific so it looks like we are limited to just changing the embedding layer. i haave added dropout layer, and added then removed additional dense relu layer. Didnt help. Where would i find more detail on this topic to get past this test? Should i be adding more parameters to embedding layer?
Many thanks.
That’s untrue.
Please read the markdown for the function create_model which asks you to add additional layers. As long as you pass unittests.parameter_count(model), the grader kernel is guaranteed not to crash. Use it as a reference to decide the model architecture.
On a somewhat related note, mentors don’t have access to learner coursera workspaces. So, it’s better for you to share the notebook as an attachment via a direct message when a mentor asks for it.
The link above, was merely the URL of the programming assignment, for reference.
ok i went to sleep for 40 minutes, came back, deleted 2 layers ( a dropout & dense layer)…and it worked. Im sure i previously tried this super simple model but anyways…seems the model only works for me with 4 layers. Thanks for your help!
1 Like