A library of probabilistic surrogate models targeting the low-noise, misspecified regime, written in Julia. The name comes from the POPS regression algorithm from Perez & Swinburne (2025).
- Univariate and multivariate POPS hypercube regression for linear models
- Leverage-based filtering of training points for efficient fitting
- Predictive uncertainty quantification: min–max bounds, standard deviations, and differential entropy estimates
- Compliance with the StatsAPI.jl interface
- Unit testing, including sanity checks against the scikit-learn based Python implementation
- Simple examples, including uncertainty quantification for a linear MLIP predictions, and structural quantities from subsequent molecular dynamics simulations.