Home▸AI & Machine Learning▸Fairness Trade-off Landscape: Threshold Parameter Space (2D)

Fairness Trade-off Landscape: Threshold Parameter Space

2D parameter-space view of the algorithmic-fairness impossibility theorem: sweep both groups' decision thresholds over heatmaps of demographic-parity, equalized-odds and calibration gaps and watch the equalized-odds and calibration low-gap zones fail to overlap, alongside a population-pyramid histogram of who gets accepted.

AI & Machine Learning2DAdvanced60 FPS📱 Mobile-adapted⇄ 3D version

This is the 2D parameter-space counterpart to the 3D fairness classifier: instead of rotating a point cloud of individually classified people, it draws the whole decision space at once. A back-to-back population-pyramid histogram shows each group's score distribution split by outcome, while three small heatmaps sweep every possible pair of per-group thresholds and colour each pair by its demographic-parity, equalized-odds, or calibration gap — making it visible, without touching a single slider, that the low-gap (green) zone of the equalized-odds map and the low-gap zone of the calibration map do not overlap whenever the two groups' base rates differ, exactly as proven by Chouldechova (2017).

⚙ Under the hood

2D parameter-space view of the algorithmic-fairness impossibility theorem: a back-to-back population-pyramid histogram of two groups' scored outcomes sits above three live heatmaps that sweep every possible pair of per-group decision thresholds, colouring each pair by its demographic-parity, equalized-odds or calibration gap -- driving the same population-generation and confusion-matrix math as the 3D scatter version.

fairnessalgorithmic biasmachine learningclassificationequalized oddscalibrationparameter spaceheatmap

2D · HTML5 Canvas 2D · 60 FPS target · runs fully client-side, no install