Anchors XAI
The explainable AI (XAI) tools we have explored up to now are model-agnostic. They can be applied to any machine learning (ML) model. The XAI tools we implemented come from solid mathematical theory and Python modules. In Chapter 8, Local Interpretable Model-Agnostic Explanations (LIME), we even ran several ML models to prove that LIME, for example, was model-agnostic.
We can represent model-agnostic (ma) tools as a function of ML(x) algorithms in which ma(x) -> Explanations. You can read the function as a model-agnostic tool that will generate explanations for any ML model.
However, the opposite is not true! Explanations(x) -> ma is false. You can read the function as an explanation of any ML model that can be obtained by any model-agnostic tool x. A model-agnostic XAI tool can technically work with an ML model x, but the results may not be satisfactory.
We can even say that an XAI tool might work with an ML algorithm and that ma(x) is...
