The main problem with training a neural network, is generating a representative sample that covers most but not all of the options, so that your Test Samples can include some options that the system will have to infer from its exposure to the training sample.