The contrastive explanations method
IBM Research and the University of Michigan researchers define the CME in the publication you can find in the Further reading section of this chapter.
The title of the publication is self-explanatory: Explanations based on the Missing: Toward Contrastive Explanations with Pertinent Negatives.
The contrastive method can be summed up as follows:
- x is an input to classify
- y is the class the model predicted for x
- are features that are present
- are features that are missing
- x is classified in y because F and M are true
One of the examples the authors of the publication describe in their paper is a classic example used to illustrate a decision-making process: estimating the health condition of a patient.
We will hence make use of the diagnosis process described in Chapter 1, Explaining Artificial Intelligence with Python, and in Chapter 3, Explaining Machine Learning with Facets.
In this section...