Part V · Chapter 56Version 1.1 · 5 November 202636 min read
Calibration Across Groups
Key message
A probability score is fair to read only if it means the same thing in every group; group-wise recalibration after training restores that meaning, but it cannot repair a biased label and it generally conflicts with equal error rates.
Five chapters or files are free. This one would be your sixth.
Leave an email address and every chapter and every PDF of the compendium opens, in this browser, for a year. Nothing else is asked, and whether we may contact you is your choice.