I created a model (h5) to predict the class of tweet (negative or positive). I can predict based on all test data using this code:
MAX_NB_WORDS = 20000