This is a highly experimental package intended to test whether things other than Platt scaling and isotonic regression are likely to be useful, in practice, when applied to the problem of "fixing" uncalibrated classification probabilities.
In particular it tests the use of an abiility transform, although that's only applicable when one item from a group is to be chosen.
pip install recalibrate
See walmart example
Proper integration or at least comparison with sklean calibration.