This paper introduces a novel method for obtaining increasedpredictive performance from transparent models in situations where productioninput vectors are available when building the model. First, labeledtraining data is used to build a powerful opaque model, calledan oracle. Se…
