Core GP functions: gp_train(),
gp_predict(), and formula interface gpss()
with predict() method.
Regression discontinuity design via gp_rdd() and
gp_rdd_plot().
Interrupted time-series design via gp_its() with
optional placebo checks.
Multiple kernel types: Gaussian, Gaussian-linear,
Gaussian-quadratic, Gaussian-periodic-linear, and
Gaussian-periodic-quadratic.
Automatic bandwidth selection via kernel variance maximization.
Optional noise variance optimization via marginal likelihood.
Support for mixed continuous/categorical covariates.
Prior mean support for incorporating external predictions.
Companion paper: Cho, Kim, and Hazlett (2026), “Inference at the
data’s edge: Gaussian processes for modeling and inference under
model-dependency, poor overlap, and extrapolation,” Political
Analysis, doi:10.1017/pan.2026.10032.