The easiest way is c). For the other ways, call advise with multiple objectives for training only. For test and trading, use a separate prediction function that loads the current model and returns the prediction array as a vector from R.