Closed zenna closed 3 years ago
As mentioned in the paper, a sufficient condition for a model being counterfactually fair is that the causal descendants of the protected attributes are not used in the classifier.
Implement a graph based check for this
As mentioned in the paper, a sufficient condition for a model being counterfactually fair is that the causal descendants of the protected attributes are not used in the classifier.
Implement a graph based check for this