Conformalized TabICL: Prediction Intervals for a State-Of-The-Art Tabular Foundation Model in Python and R

A tutorial demonstrating how to generate conformal prediction intervals for tabular regression using TabICL (a transformer-based tabular foundation model) in both Python and R. The workflow uses the nnetsauce PredictionInterval wrapper around TabICLRegressor and RidgeCV, evaluated on the sklearn diabetes dataset. Results show both models achieve ~95.5% coverage, with RidgeCV producing slightly narrower intervals (avg_width 211.5 vs 226.1), suggesting the dataset may be too simple to showcase TabICL's advantage. R users can access Python libraries via reticulate and rpy2.