Closed jakwisn closed 4 years ago
After a few tests, I realized that the better option is to leave it as it is and change documentation. Now it suggests that on the left side of fairness check unprivileged subgroups are the deprived ones which is no longer the case.
Equal opportunity now will be TPR as it originally should. This way properties stated before should be satisfied
After the last change the side of Equal opportunity ratio is inverted. Less false negatives is a good thing (at least from perspective of human that might be affected by ai) therefore scale for this metric should be inverted to match other metrics and be easily interpretable.