ALIBI
ALIBI is another popular XAI framework that supports both local and global explanations for classification and regression models. In Chapter 2, Model Explainability Methods, we did explore this framework for getting counterfactual examples, but ALIBI does include other model explainability methods too, which we will explore in this section. Primarily, ALIBI is popular for the following list of model explanation methods:
- Anchor explanations: An anchor explanation is defined as a rule that sufficiently revolves or anchors around the local prediction. This means that if the anchor value is present in the data instance, the model prediction is almost always the same, irrespective of changes to other feature values.
- Counterfactual Explanations (CFEs): We have seen counterfactuals in Chapter 2, Model Explainability Methods. CFEs indicate which feature values should change, and by how much, to produce a different outcome.
- Contrastive Explanation Methods (CEMs): CEMs...