I completed assignment 3 but was surprised to find my model was not producing the correct dates.
It turns out that the code can pass all the unit tests and receive full credit even if the model does not work.
I was able to find the issue ; which was in the post_activation_LSTM_cell ; the problem was that I kept using the same initial states.
Is your output similar to this one? Check the full thread, please.
I was the original author of that post you linked to, that was my exact output 
Yes that is exactly the output I was getting!
Thankfully thinking it through a bit more carefully revealed that it was a very simple fix.
The fix was to correctly set the post_activation_LSTM_cell(inputs=?, initial_state=[?, ?])
s and c are initialized to s0 and c0 ; we don’t want to continually input s0 and c0 or we will keep getting the first predicted character for the entire sequence.
@b0otable! Are you still facing this issue?
@Joshua_Martin_Osborn! Please mention that simple fix (without revealing the code). This way, users in the future can easily get the answer.