The unit test (and apparently this includes the grader) assigns y_test and y_val inputs as numpy ndarrays. E.g., in the unit test in the notebook,
example_y = example_df.TRTMT.values
However, further, when the treatment_dataset_split function is put to use, the y_test and y_vals are passed as pandas series rather than as plain ndarray.
This has meant one or the other broke my code to separate the labels between whether the patient was treated or not. If my code treated the input as a pandas series, the unit test and the grader failed on it, while it worked in the subsequent cells using the function; if I treated the input as an ndarray, then the unit test and the grader passed it, but the subsequent code errored out.
In other words, I found that I had to either pass the grader and break the notebook later, or vice versa.
Frankly, this assignment was really overly convoluted to be of value.