scikit-learn-contrib / MAPIE

A scikit-learn-compatible module to estimate prediction intervals and control risks based on conformal predictions.
https://mapie.readthedocs.io/en/latest/
BSD 3-Clause "New" or "Revised" License
1.3k stars 111 forks source link

Conditional CP should re-optimize g at inference time #506

Closed Damien-Bouet closed 2 weeks ago

Damien-Bouet commented 3 months ago

Is your feature request related to a problem? Please describe. The CCP method related to #449 would require to correct the optimised values for each new X_n+1, considering a bound M, to have the expected theoretical guarantees .

Describe the solution you'd like We need to re-optimize the g function, with an augmented dataset composed by the n calibration points and X_n+1. It can still be interesting to compute the optimisation with only the n calibration points, to use this optimised value as starting value for the inference optimizations.