Global surrogates

Surrogate model is an overloaded term. It is used in engineering, statistics, economics, and physics, to name a few, often in the context of metamodels, mathematical optimizations, or simulations.

In the context of machine learning interpretation methods, global surrogate model usually refers to a white-box model that you train with the black-box models' predictions. We do this to extract insights from the white-box model's intrinsic parameters, much like we did in Chapter 3, Interpretation Challenges. There is also another way to use surrogate models: to use a black-box model to approximate and evaluate another model that you don't have access to, but you have its predictions. We will do just this in Chapter 7, Anchor and Counterfactual Explanations, but we prefer the term proxy model for this kind of surrogate.

You don't need any fancy libraries to create a global surrogate. You can use any of the white-box models we discussed in Chapter...