Choose a Package

Quick routing between np, npRmpi, and crs based on the job you want to do.
Keywords

np, npRmpi, crs, choose a package, kernel methods, splines

If you are deciding where to begin, the short answer is this: start with the package that matches the job, not with the longest document.

If you want to… Start with Why
Fit kernel regression, density, distribution, or related tests on one machine np This is the core package and the natural starting point for most users.
Run the same style of kernel workflows on larger jobs using MPI npRmpi The package supports a modern session route on macOS/Linux and attach under mpiexec when that is the appropriate entry path.
Work with categorical regression splines and related constrained spline methods crs This package is the right entry point for spline-based work rather than kernel-based work.

A practical rule of thumb

  • Start with np unless you already know that you need MPI.
  • Move to npRmpi when the workflow is the same but the job is too large or too slow in serial.
  • Use crs when spline bases, shape restrictions, or spline-specific methods are the natural fit.

Current Capabilities

  • Runtime, Memory, and Scaling: np and npRmpi 0.80-1 standardize bootstrap, uncertainty, and dotted plotting controls; retain prepared samples for safer bandwidth-object reuse; add simultaneous variability bands across the principal estimator families; support factor predictors in single-index models; and strengthen bounded-kernel, nearest-neighbour, scaled-bandwidth, plotting, and serial/MPI behavior.
  • Splines: crs 0.15-46 retains the common named plotting controls and crshat() operator workflow, adds formula/data and S3 support for instrumental spline regression, and provides the observed NOMAD interface used by downstream degree searches.
  • Kernel Methods: np and npRmpi support local polynomial capability across the conditional family, together with joint degree and bandwidth selection for mixed data, gradient support throughout the conditional estimators, and derivative estimation of arbitrarily high order, limited only by the selected polynomial degree whether set manually or chosen automatically for the underlying smoothness class.
  • Density, Distribution, Quantiles: boundary-aware unconditional and conditional density routes cover known two-sided or one-sided bounds and practical unknown-bound proxies, with a dedicated worked example for finite- and unbounded-support settings.
  • Density, Distribution, Quantiles: nplsqreg and nplsqregbw provide a first-class location-scale quantile workflow, with formula and bandwidth-object routes, scalar or vector tau, stored-quantile prediction selection, residuals, summaries, plot routes, and the same NOMAD-enabled LP search style where supported.
  • Density, Distribution, Quantiles: npqreg is a first-class estimator with direct formula support, automatic npcdistbw() selection when bws is omitted, vector tau, gradients, standard errors, and nomad = "auto" pass-through for local-polynomial degree-and-bandwidth search.
  • Classification and Modes: npconmode is a first-class estimator with direct formula/data/newdata routes, automatic npcdensbw() selection when bws is omitted, probability and gradient support, mirrored degree/search metadata, and nomad = "auto" pass-through for local-polynomial degree-and-bandwidth search.
  • Density, Distribution, Quantiles: npcopula is a first-class estimator with a direct formula interface, automatic npudistbw() or npudensbw() selection when bws is omitted, and automatic two-dimensional probability-grid construction for the default copula grid route.
  • Semiparametric Models: npreghat, npindexhat, npplreghat, and npscoefhat simplify constrained nonparametric and semiparametric estimation through one-call hat-matrix operator routes, while npudenshat, npudisthat, npcdenshat, and npcdisthat extend the same operator style to the density/distribution family.
  • Plotting and Intervals: pointwise, Bonferroni, and simultaneous variability bounds share the plotting surface across supported regression, density/distribution, quantile, class-probability, and copula targets, and band = "all" shows all three together.
  • Plotting and Intervals: renderer = "rgl" supports interactive 3D rotation/zoom when rgl is installed, while data.rug and data.overlay manage rugs and observed-data overlays on supported routes.
  • MPI and Large Data: npRmpi session and attach routes make np and npRmpi code directly interchangeable, while preserving the profile route for users who want it.

Common routes

I want a simple first success

Go to Install and Get Started, install the package you need, and then use Quickstarts for the smallest runnable scripts before worrying about longer documents.

I already know I want kernel methods

Go to Kernel Methods. That page starts with np and then points you to npRmpi when parallel execution is warranted.

I already know I need MPI

Go directly to MPI and Large Data. The guidance there is organized around current session, attach, and profile usage rather than the older wrapper-heavy style.

I want working scripts

Go first to Quickstarts if you want the shortest path to runnable code. Then use Code Catalog, Worked Examples, or Interactive Demos when you want more context or a broader script library.

First commands to remember

## Install the two serial packages directly from CRAN
install.packages("np")
install.packages("crs")

For npRmpi, installation depends on your MPI setup and operating system, so it is treated separately on MPI and Large Data.

Back to top