Model-agnostic techniques

  • We don’t assume anything about the model structure.

  • We will assume that the model operates on p-dimensional vector of explanatory variables/features and, for a single observation, it returns a single value (score/probability), which is a real number.