Semiparametric kernel primitives (R port)

R port of morie.semipar_bridge. Provides the kernel-based building blocks used by morie's nuisance estimation pipelines (TMLE, AIPW, DML): kernel evaluation, Nadaraya-Watson regression, local linear regression, kernel density estimation, and bandwidth selection.

Details

The Python module loads a C shared library (semipar_kernels.dylib / .so) and falls back to NumPy. The R port is pure R: it implements the same algorithms in vectorised form and additionally wraps mgcv::gam for a high-quality penalised-spline smoother as an alternative to manual bandwidth selection.

Functions

• kernel_eval: evaluate a kernel function.

• nw_regression: Nadaraya-Watson kernel regression.

• local_linear: local linear kernel regression.

• kde: kernel density estimation.

• silverman_bandwidth: rule-of-thumb bandwidth.

• loocv_bandwidth: leave-one-out CV bandwidth for NW regression.

• kernel_cond_moments: kernel-weighted mean and variance.

• gam_smoother: mgcv::gam thin-plate smoother fit + predict.

• SemiparKernels: object-style wrapper.