Yes, all those are global variables.

This code was not working for me, but it gave me the clue. I changed spark dataframes to pandas outside of the image generators and then I moved Image generators inside of the objective function. This way there was no need to pass the dataframes as arguments to the objective function. It is now working with sparktrials parallelism 🙂

As for the score, it is a list of scalars and score[0] is our test_loss which is a float. What did you receive when you cast it?