Subject-Conditional (Mondrian) example on a public bio signal dataset — WESAD

[sushan5140]

Hi Henrik,

I've been using crepes on a stress-detection project with WESAD (ECG/EDA/respiration) and hit the problem Mondrian conformal prediction is meant to solve pooled calibration undercovers the harder subjects. Once I calibrate per subject with Mondrian Categorizer keyed on subject ID, that goes away.

I'd like to write a notebook for docs/ showing this on WESAD directly, standard vs. normalized vs. Mondrian, subject ID as the category, with a coverage comparison on real data instead of a toy example.

Would that fit in docs/ or would you rather have it in crepes.extras? I can match whatever the rest of the repo does.