C5W2 Emojify assignment - LSTM model not working although passed all unit tests and grader


After I implemented the V2 model for Emojify assignment, it passed all the unit tests in the notebook and also got 100 points in the grader. But I am not able to get it to correctly classify “I cannot play” (it answers :baseball: instead of expected :disappointed:)

The model got 100% training accuracy but only 73% test accuracy as compared to expected 80%-95% test accuracy.

Anyone else seeing the same? Where should I look at to “fix” this problem? ( May I share my model source code here? I am not sure if this is against the code of conduct. )

Oh, I didn’t read - " * If it didn’t, be aware that Keras’ outputs are slightly random each time, so this is probably why."

But I tried to recreate the model and retrain it 10 times still could not get it to work once…any help appreciated.

You can send me your code via a private message.

Thank you @TMosh. PM’ed you my source.