GitLab

I noticed that the next() method of the JointBatchGenerator class assumes that X is a numpy array. This is true as we load objects from a single column in the HDF5 table. If, however, we load multiple objects (e.g. for training a multi-modal model) by specifying a list of x_field values, the next() method brakes.

This merge request generalizes the implementation to be able to handle those cases in which X is a list of numpy arrays.