FastKRR

The ‘FastKRR’ implements its core computational operations in C++ via ‘RcppArmadillo’, enabling faster performance than pure R, improved numerical stability, and parallel execution with OpenMP where available. On systems without OpenMP support, the package automatically falls back to single-threaded execution with no user configuration required. For efficient model selection, it integrates with ‘CVST’ to provide sequential-testing cross-validation that identifies competitive hyperparameters without exhaustive grid search. The package offers a unified interface for exact kernel ridge regression and three scalable approximations—Nyström, Pivoted Cholesky, and Random Fourier Features—allowing analyses with substantially larger sample sizes than are feasible with exact KRR. It also integrates with the ‘tidymodels’ ecosystem via the ‘parsnip’ model specification ‘krr_reg’, the S3 method ‘tunable.krr_reg()’, and the direct fitting helper ‘fit_krr()’.

Dependencies: Rcpp, RcppArmadillo, CVST, parsnip
This package uses CVST (GPL ≥ 2). Overall license: GPL (≥ 2).

Installation

You can install the development version of FastKRR from GitHub with:

# install.packages("pak")
pak::pak("kybak90/FastKRR")

Example

This is a basic example of fitting a Gaussian kernel ridge regression estimator to a dataset \({(x_i, y_i)}_{i=1}^n\) using (1) exact computation, (2) pivoted Cholesky decomposition, (3) Nyström approximation, and (4) random Fourier features.

library(FastKRR)

# example data set
set.seed(1)
n = 1000; d = 1
rho = 1
X = matrix(runif(n*d, 0, 1), nrow = n, ncol = d)
y = as.vector(sin(2*pi*rowMeans(X)^3) + rnorm(n, 0, 0.1))

# model fitting - exact
model_exact = fastkrr(X, y, kernel = "gaussian", rho = rho, opt = "exact", verbose = FALSE)

# model fitting - pivoted
model_pivoted = fastkrr(X, y, kernel = "gaussian", rho = rho, opt = "pivoted", verbose = FALSE)

# model fitting - nystrom
model_nystrom = fastkrr(X, y, kernel = "gaussian", rho = rho, opt = "nystrom", verbose = FALSE)

# model fitting - rff
model_rff = fastkrr(X, y, kernel = "gaussian", rho = rho, opt = "rff", verbose = FALSE)


# prediction
new_n = 500
new_x = matrix(runif(new_n*d, 0, 1), nrow = new_n, ncol = d)
new_y = as.vector(sin(2*pi*rowMeans(new_x)^3) + rnorm(new_n, 0, 0.1))

pred_exact = pred_krr(model_exact, new_x)
pred_pivoted = pred_krr(model_pivoted, new_x)
pred_nystrom = pred_krr(model_nystrom, new_x)
pred_rff = pred_krr(model_rff, new_x)

The visualization of the fitted results is shown below.

library(ggplot2)
data = data.frame(new_x, new_y)
data$pred_exact = as.numeric(pred_exact)
data$pred_pivoted = as.numeric(pred_pivoted)
data$pred_nystrom = as.numeric(pred_nystrom)
data$pred_rff = as.numeric(pred_rff)

ggplot(data = data, aes("x" = new_x, "y" = new_y)) +
  geom_point("x" = new_x, "y" = new_y) +
  geom_line(aes("x" = new_x, "y" = pred_exact, color = 'opt = "exact"'), linewidth=1.2) + 
  geom_line(aes("x" = new_x, "y" = pred_pivoted, color = 'opt = "pivoted"'),linewidth=1.2) + 
  geom_line(aes("x" = new_x, "y" = pred_nystrom, color = 'opt = "nystrom"'), linewidth=1.2) + 
  geom_line(aes("x" = new_x, "y" = pred_rff, color = 'opt = "rff"'), linewidth=1.2)

OpenMP support on macOS

1) Install OpenMP runtime (Homebrew)

# Ensure Command Line Tools (if not already installed)
xcode-select --install || true

# Install/update Homebrew libomp
brew update
brew install libomp

2) Configure ~/.R/Makevars (works for Apple Silicon & Intel)

Copy the block below into ~/.R/Makevars (create the file if it does not exist).

# OpenMP on macOS with Apple Clang 
# Try to detect Homebrew prefix; fallback covers both Apple Silicon & Intel.
HOMEBREW_PREFIX := $(shell brew --prefix 2>/dev/null)
ifeq ($(strip $(HOMEBREW_PREFIX)),)
  HOMEBREW_PREFIX := /opt/homebrew
endif

# Include & lib paths (also add Intel path as extra fallback)
CPPFLAGS  += -I$(HOMEBREW_PREFIX)/opt/libomp/include -I/usr/local/opt/libomp/include
LDFLAGS   += -L$(HOMEBREW_PREFIX)/opt/libomp/lib     -L/usr/local/opt/libomp/lib

# Standard R variables for OpenMP
SHLIB_OPENMP_CFLAGS   = -Xpreprocessor -fopenmp
SHLIB_OPENMP_CXXFLAGS = -Xpreprocessor -fopenmp
SHLIB_OPENMP_FCFLAGS  = -fopenmp
SHLIB_OPENMP_FFLAGS   = -fopenmp
SHLIB_OPENMP_LIBS     = -lomp

Restart R after editing ~/.R/Makevars so the flags are picked up.

3) Install FastKRR

#pak::pak("kybak90/FastKRR")
library(FastKRR)

4) Example & OpenMP confirmation (inside FastKRR)

# example data set
set.seed(1)
n = 1000; d = 1
rho = 1
X = matrix(runif(n*d, 0, 1), nrow = n, ncol = d)
y = as.vector(sin(2*pi*rowMeans(X)^3) + rnorm(n, 0, 0.1))

# model fitting - nystrom
model_nystrom = fastkrr(X, y, kernel = "gaussian", rho = rho, opt = "nystrom", verbose = FALSE)

model_nystrom$n_threads    # >1 indicates OpenMP used by FastKRR (default 4)
#> [1] 1

The following code chunk inspects the dynamic libraries linked to the compiled FastKRR shared object (.so or .dylib).

is_macos = (Sys.info()[["sysname"]] == "Darwin")

if (is_macos) {
  so_dir = system.file("libs", .Platform$r_arch, package = "FastKRR")
  if (so_dir == "") so_dir = system.file("libs", package = "FastKRR")
  so_files = list.files(so_dir, pattern = "\\.(so|dylib)$", full.names = TRUE)

  so = if (length(so_files)) so_files[[1]] else ""
  cat("FastKRR shared object:", so, "\n")

  if (nzchar(so)) {
    cmd = paste("otool -L", shQuote(so))
    cat(system(cmd, intern = TRUE), sep = "\n")
  } else {
    cat("No shared object found. Is FastKRR installed with compiled code?\n")
  }
} else {
  cat("Skipping otool check (not macOS).\n")
}

When you run this on macOS, a successful OpenMP build will show a line such as:

/Library/Frameworks/R.framework/Versions/4.3-arm64/Resources/lib/libomp.dylib

This confirms that OpenMP is enabled for FastKRR on your system.