My model doesn’t output anything not before or post training but still all unit test case passed and got 100%, please help to find the issue.
@Deepti_Prasad any suggestions ??
Because your output of prediction output is empty, I would first check how are tokenizing the input as you got no word in prediction word, so either there is mix up between the global variables to local variables or not recalling with arguments assigned for the grade cell as the unittest didn’t catch this error.
then check if in your next word prediction the sequence of assigned arguments in form of question, answer and model as your test does show passed.
let me know if you need me to check the codes, then please send a screenshot of the grade cell you got issue I mean next word grade cell by personal DM.
Regards
DP
after seeing your codes, first kindly confirm if you are working on an updated copy of assignment
I am doubting that you are working on an obselete copy.
Check in the lab environment, click on ?, then check log, you will know if you are working on an old copy.
Kindly confirm
to get fresh copy, follow the below link comments 
Regards
DP
after rebooting, click ? again, then Get the latest Version and Update lab, to get the latest updated copy.
Confirm once you got the fresh copy.
Also remember course 3 and course 4 of NLP specialisation were updated in early this year and recently there were some changes made in some assignment, so always update the lab before you start an assignment.
regards
DP
hi @Deepti_Prasad thanks for helping here. The issue was how I was trying to pass the shape to masking function, instead of simply passing the shape, I was checking the shape first and then passing the column dimension which led me to this issue. As you pointed out there was no need of passing tf.shape there.
Although, Please consider including a check in test case in where it should check the length also of the generated sequence where length should be greater than zero, I checked the code for test cases and found It just checks the type and element type of output as prediction can’t be known for any user but still we can make sure that model should predict something atleast by checking len(predicted seq) > 0.
Again, thanks for all the help.
As I far as I understood the reason they didn’t add the length for prediction sequence is because they are looking more simpler perspect of next word prediction. you will find newer assignment where you will find this task.
1 Like