fastlpr 1.0.0
Initial CRAN Release
Features
- Fast local polynomial regression via NUFFT with O(N + M log M)
complexity
- Kernel density estimation (KDE) for 1D, 2D, and 3D data
- Local polynomial regression with orders 0 (Nadaraya-Watson), 1
(local linear), and 2 (local quadratic)
- Complex-valued response support
- Heteroscedastic variance estimation
- Automatic bandwidth selection via GCV (regression) and LCV (density
estimation)
- 1-SE rule for conservative bandwidth selection
- Confidence interval computation
- OpenMP parallelization via Rcpp/RcppArmadillo (optional)
Main Functions
- cv_fastlpr() - Cross-validated local polynomial regression
- cv_fastkde() - Cross-validated kernel density estimation
- get_hlist() - Generate bandwidth grid
- fastlpr_predict() - Prediction at new data points
- fastlpr_interval() - Confidence interval computation
Notes
- R port of the MATLAB/Python fastLPR toolbox
- Verified against MATLAB reference implementation (MSE <
1e-8)
- Performance: within 8x of MATLAB speed for all test cases