Inspired by the selectiveinference framework, we propose an approach that enables learning the hyperparameters and testing onthefullsample without data splitting.

Although the fairness community has recognized the importance of data, re-searchers inthe area primarily rely on UCIAdult when itcomes totabular data.