Deep neural network Question

I’m recently looking for some deep neural network (DNN) model for choice prediction. There are various models but one trouble is that, in my problem, the number of choice options in the training data is different from the number of choice options in the validation data (Choice-based Conjoint). Does anyone know some DNN model that may be able to handle such data?

Just try to understand your problem, it is classification problem. But the number of classes for training is different from number of classes for prediction (or inference). Is that correct? If so, is the classes of prediction subclass of the classed of training?