Short Course on R Tools

Introduction to Rcpp

Mehdi Maadooliat and Hossein Haghbin

Marquette University
SCoRT - Summer 2025

Outline

  • Motivation & Introduction
  • Getting Started with evalCpp() & cppFunction()
  • Using sourceCpp()
  • Data Types & Conversions
  • Standard Template Library (STL)
  • Using Rcpp in Packages
  • Advanced Topics & Resources

Motivation & Introduction

  • Performance Bottlenecks: loops, recursion, complex operations
  • Rcpp: smooth bridge between R and C++
  • Do things you could not do before

  • Advantages:

    • Low overhead function calls
    • Access to STL (data structures & algorithms)
    • Cleaner, maintainable code

Basic Usage: evalCpp()

  • evalCpp() evaluates a single C++ expression. Includes and dependencies can be declared.

  • This allows us to quickly check C++ constructs.

library(Rcpp)

evalCpp("2 + 2")  # simple test
># [1] 4

evalCpp("std::numeric_limits<double>::max()")
># [1] 1.79769e+308

Simple Example

  • R Version of ‘is this number odd or even’
isOdd_r <- function(num = 10L) {
  result = (num %% 2L == 1L)
  return(result)
}
c(isOdd_r(42L), isOdd_r(43L))
># [1] FALSE TRUE
  • C++ Version of ‘is this number odd or even’
bool isOdd_cpp(int num = 10) {
  bool result = (num % 2 == 1);
  return result;
}

Simple Example - Using cppFunction()

  • Rcpp Version of ‘is this number odd or even’
Rcpp::cppFunction("
bool isOdd_cpp(int num = 10) {
  bool result = (num % 2 == 1);
  return result;
}")
c(isOdd_cpp(42L), isOdd_cpp(43L))
># [1] FALSE TRUE
  • Use cppFunction() in R console or script
  • Fast prototyping & testing
  • Good for small functions

Second Example: VAR(1)

Let’s consider a simple possible \(VAR(1)\) system of \(k\) variables.

For \(k = 2\):

\[ X_t = X_{t-1} B + E_t \]

where \(X_t\) is a row vector of length \(2\), \(B\) is a \(2×2\) matrix, and \(E_t\) is a row of the error matrix of \(2\) columns.

Second Example: VAR(1)

In R (C++) code, given both the coefficient and error matrices:

rSim <- function(B, E) {
  X <- matrix(0, nrow(E), ncol(E))
  for (r in 2:nrow(E)) {
    X[r, ] = X[r-1, ] %*% B + E[r, ]
  }
  return(X)
}

C++ Version:

arma::mat cppSim(arma::mat B, arma::mat E) {
  int m = E.n_rows, n = E.n_cols;
  arma::mat X(m, n);
  X.row(0) = arma::zeros<arma::mat>(1, n);
  for (int r = 1; r < m; r++) {
    X.row(r) = X.row(r-1) * B + E.row(r);
  }
  return X;
}

Second Example: VAR(1)

Rcpp::cppFunction('arma::mat cppSim(arma::mat B, arma::mat E) {
  int m = E.n_rows, n = E.n_cols;
  arma::mat X(m, n);
  X.row(0) = arma::zeros<arma::mat>(1, n);
  for (int r = 1; r < m; r++) {
    X.row(r) = X.row(r-1) * B + E.row(r);
  }
  return X;
}', depends="RcppArmadillo")

a <- matrix(c(0.5, 0.1, 0.1, 0.5), nrow = 2)
e <- matrix(rnorm(10000), ncol = 2)
rbenchmark::benchmark(cppSim(a, e), rSim(a, e), order="relative")[, 1:4]

>#    test           replications elapsed relative
># 1  cppSim(a, e)  100         0.010   1.0
># 2  rSim(a, e)    100         0.728  72.8

Growth of Rcpp

  • Sometimes speed is not the only reason
    • Easy access to C/C++ libraries
    • C & C++ provide numerous libraries + APIs
    • Easy to provide access to as Rcpp eases data transfer
  • Rcpp is currently used by
    • 3053 CRAN packages
    • 250+ BioConductor packages
    • an unknown (but “large”) number of GitHub projects

Pagerank

#remotes::install_github("https://github.com/andrie/pagerank.git")
suppressMessages(library(utils))
library(pagerank)

cran <- "https://cloud.r-project.org"
pr <- compute_pagerank(cran)
round(100 * pr[1:5], 3)

>#   Rcpp    ggplot2     dplyr      MASS   magrittr
># 2.744     1.516       1.274    1.122    0.814

Screenshot of Pagerank

Percentage of Compiled Packages

db <- tools::CRAN_package_db()       # added in R 3.4.0
db <- db[!duplicated(db[, 1]), ]     # rows: number of packages

nTot <- nrow(db)                     # columns: different attributes
nRcpp <- length(tools::dependsOnPkgs("Rcpp", recursive = FALSE, installed = db))
nCompiled <- table(db[, "NeedsCompilation"])[["yes"]]

propRcpp <- nRcpp / nCompiled * 100
data.frame(tot = nTot, totRcpp = nRcpp, totCompiled = nCompiled,
           RcppPctOfCompiled = propRcpp)

>#   tot   totRcpp   totCompiled   RcppPctOfCompiled
># 1 22501    3053        4981          61.29291

Rcpp in Rstudio - Using sourceCpp()

  • Create standalone .cpp files with:

    # "file.cpp" contents

#include <Rcpp.h>
using namespace Rcpp;
// [[Rcpp::export]]
NumericVector timesTwo(NumericVector x) {
  return x * 2;
}

  • Compile with sourceCpp('file.cpp')

    sourceCpp('file.cpp')
timesTwo(42)
># 84

Using sourceCpp()

  • So what just happened?
    • We defined a simple C++ function
    • It operates on a numeric vector argument
    • We ask Rcpp to ‘source it’ for us
    • Behind the scenes Rcpp creates a wrapper
    • Rcpp then compiles, links, and loads the wrapper
    • The function is available in R under its C++ name
  • Benefits:
    • Editor support (syntax highlighting)
    • Easier debugging & error tracking

Another Example: Focus on Speed

Consider a function defined as

\[ f(n) = \begin{cases} n & \text{when } n < 2 \\ f(n-1) + f(n-2) & \text{when } n \ge 2 \end{cases} \] that creates Fibonacci sequence. The R implementation and use:

f <- function(n) {
  if (n < 2) return(n)
  return(f(n - 1) + f(n - 2))
}
sapply(0:10, f)  # Using it on first 11 arguments

># [1] 0 1 1 2 3 5 8 13 21 34 55

Another Example: Timing R Implementation

  • The R implementation:
f <- function(n) {
  if (n < 2) return(n)
  return(f(n - 1) + f(n - 2))
}
  • Timing:
library(rbenchmark)
benchmark(f(15), f(20), f(25))[, 1:4]

># test         replications elapsed   relative
># 1 f(15)       100          0.06      1.0
># 2 f(20)       100          0.75     12.5
># 3 f(25)       100          7.92    132.0

Another Example: C++ Implementation

int g(int n) {
  if (n < 2) return(n);
  return(g(n - 1) + g(n - 2));
}

Rcpp::cppFunction('
  int g(int n) {
    if (n < 2) return(n);
    return(g(n - 1) + g(n - 2)); 
  }')

sapply(0:10, g)  # Using it on first 11 arguments

># [1] 0 1 1 2 3 5 8 13 21 34 55

Another Example: Comparing Timing

  • Rcpp implementation:
Rcpp::cppFunction('
  int g(int n) {
    if (n < 2) return(n);
    return(g(n - 1) + g(n - 2)); 
  }')
  • Timing:
library(rbenchmark)
benchmark(f(25), g(25))[, 1:4]

>#   test      replications elapsed   relative
># 1 f(25)      100          8.02     267.333
># 2 g(25)      100          0.03       1.000
  • A nice gain of a few orders of magnitude.

Data Types & Conversions

  • R vectors ↔︎ C++ classes:

    • NumericVector, IntegerVector, CharacterVector, LogicalVector

  • Scalars: double, int, String, bool

  • Key methods:

    • .size(), .begin(), .end()
    • Constructors: NumericVector out(n)

Types

R Type mapping:

  • Standard R types (integer, numeric, list, function, … and compound objects) are mapped to corresponding C++ types:
library(Rcpp)
cppFunction("NumericVector logabs(NumericVector x) {
  return log(abs(x));
}")
logabs(seq(-5, 5, by = 2))
># [1] 1.609438 1.098612 0.000000 0.000000 1.098612 1.609438
  • Vectorized C++! Here log(abs()) runs directly on vectors as R would.

STL Type Mapping

  • code/logabs2.cpp
#include <Rcpp.h>
using namespace Rcpp;

inline double f(double x) { return ::log(::fabs(x)); }

// [[Rcpp::export]]
std::vector<double> logabs2(std::vector<double> x) {
  std::transform(x.begin(), x.end(), x.begin(), f);
  return x;
}

Not vectorized but ‘sweeps’ f() along std::vector<double> via STL std::transform().

library(Rcpp)
sourceCpp("code/logabs2.cpp")
logabs2(seq(-5, 5, by = 2))
># [1] 1.609438 1.098612 0.000000 0.000000 1.098612 1.609438

Hands-on Exercises (30 min)

Using Rcpp in Packages

  • Add to package:

    • LinkingTo: Rcpp in DESCRIPTION
    • useDynLib(pkg) & importFrom(Rcpp, sourceCpp) in NAMESPACE

  • usethis::use_rcpp() automates setup

  • Run Rcpp::compileAttributes() to generate bindings

  • Thirteen Simple Steps Vignette by Dirk Eddelbuettel

Advanced Topics & Resources

  • Attributes vignette: vignette('Rcpp-attributes')
  • Modules: expose C++ classes to R
  • RcppGSL, RcppArmadillo for specialized libraries
  • Books: Effective C++, Effective STL
  • Online: Rcpp homepage, mailing list

Summary & Q&A

  • Key takeaways:
    1. Profile before optimizing
    2. Inline vs file-based workflows
    3. Master data conversions & STL
  • Questions?

Thank You

🔗 http://bit.ly/SCoRT