Failing combination of stacked ensemble, MOJO, CV fold assignment, and deep learning

Question

My objective is a MOJO containing a stacked ensemble, including deep learning, for models trained when specifying the cross validation fold assignment. The resultant MOJOs f