Rcpp

— Christopher Genovese and Alex Reinhart

Rcpp Basics #

Wrapping #

evalCpp()
evaluate short C++ code snippets, given as string
cppFunction()
defines an R function from a C++ function given as a string
sourceCpp()
compiles and links a C++ source file and exports tagged functions into R

Example:

f <- cppFunction('double weightedMean(NumericVector x, NumericVector w) {
  int n = x.size();
  double numerator = 0.0;
  double denominator = 0.0;
  for ( int i = 0 ; i < n ; ++i ) {
      numerator += x[i] * w[i];
      denominator += w[i];
  }
  // No error checking or assertions in this example, see below
  return numerator/denominator;
}')
f(1:4, rep(1,4))  # => 2.5

Note the structure of the for loop:

for ( initializers; condition; updater ) { BODY }

where the initializer can contain declaration of variables that are then only visible inside the loop.

For error checking we might include:

try {
    if ( is_true(any(w < 0.0)) || denominator <= 0.0 ) {
        throw std::domain_error("Invalid weights");
    }
    return numerator/denominator;
} catch(std::exception &ex) {
    forward_exception_to_r(ex);
} catch(...) {
    ::Rf_error("c++ exception (unknown reason)");
}

sourceCpp() reads its input from a file and creates a shared, dynamically linked library. (It can really also take a string with the code argument, which is how the other two work.)

// File slow-fib.cpp
#include <Rcpp.h>

using namespace Rcpp;

// [[Rcpp::export]]
int fibonacci(const int x) {
    if (x == 0) return(0);
    if (x == 1) return(1);
    return fibonacci(x - 1) + fibonacci(x - 2);
}

Note the export comment tag (the space matters), which marks the function for export to R.

sourceCpp("slow-fib.cpp")
fibonacci(10) # => 55

C++ Features #

Unlike R, C++ is a compiled, statically typed language.

Each variable must be given a specific type, and each function must be declared with the types of its arguments and of its return value.

Static typing lets the compiler optimize effectively, but it puts more constraints on the developer.

C++ is a large, complex language with many features. A few things are worth remembering:

  • Standard C is also legal C++.
  • Syntax has similarities with R, but (non-compound) statements must all be terminated with a ;.
  • C++ (like C) is zero-indexed, not one-indexed like R. Beware.
  • There is no <- operator: use = for assignment.
  • Scalars and vectors (or other aggregate types) are not interchangeable (though a spoonful of Sugar helps).
  • Functions must explicitly return their value.
  • You can use C libraries and functions directly (note: externs).
  • The Standard Template Library (or STL) exposes a wide variety of rich and well-tested data structures and algorithms.
  • The Boost library is a powerful third-party library that goes above and beyond the STL.
  • C++ has evolved, modern versions: C++11 and C++14 offer many nice new features. You may have to configure specially to use those features with Rcpp.

Scalar Types #

The common “scalar” types are bool, int, double, and String. (All but the last of these are C++ primitive types.)

double trim(double x, double threshold) {
    if ( x > threshold ) {
        return threshold;
    } else if ( x < -threshold ) {
        return -threshold;
    } else {
        return x;
    }
}

Exercise: Write a function signum() that takes an integer and returns -1, 0, or 1 if that integer is negative, zero, or positive.

int signum(int x) {
  if (x > 0) {
    return 1;
  }
  if (x == 0) {
    return 0;
  }
  return -1;
}

Vector Types #

Rcpp defines several classes to handle R vectors. These have a nice range of methods and work well with “sugar” as we’ll see below.

NumericVector, IntegerVector, CharacterVector, LogicalVector

For instance, you use the .size() method to get the length of the vector, as illustrated above.

Several ways to create vectors:

SEXP x;
std::vector<double> y(10);

NumericVector xx(x);       // create from a SEXP
NumericVector xx(10);      // of a given size (filled with 0)
NumericVector xx(10, 2.0); // ... with a default for all values
NumericVector xx( y.begin(), y.end() ); // range constructor

// using create
NumericVector xx = NumericVector::create(
    1.0, 2.0, 3.0, 4.0 );
// with names attribute
NumericVector yy = NumericVector::create(
    Named["foo"] = 1.0,
    _["bar"] = 2.0 ); // short for Named

Extracting and assigning values:

double u = xx[0];
double v = xx(1);
double z = yy["foo"] + yy["bar"];

xx[0] = 1.618;
xx(1) = -1.0;
yy["foo"] = 10.0;

yy["foobar"] = 1;  // grow the vector

These vectors support some nice R-like operations:

// [[Rcpp::export]]
NumericVector positives(NumericVector x) {
    return x[x > 0];
}

// [[Rcpp::export]]
NumericVector in_range(NumericVector x, double low, double high) {
    return x[x > low & x < high];
}

// [[Rcpp::export]]
NumericVector no_na(NumericVector x) {
    return x[ !is_na(x) ];
}

// [[Rcpp::export]]
List first_three(List x) {
    IntegerVector idx = IntegerVector::create(0, 1, 2);
    return x[idx];
}

// [[Rcpp::export]]
List with_names(List x, CharacterVector y) {
    return x[y];
}

Returning new vectors

pdistR <- function(x, ys) {
  sqrt((x - ys)^ 2)
}
NumericVector pdistCpp(double x, NumericVector ys) {
  int n = ys.size();
  NumericVector out(n);  // <- note constructor

  for(int i = 0; i < n; ++i) {
    out[i] = sqrt(pow(ys[i] - x, 2.0));
  }
  return out;
}

Matrix Types #

Rcpp supplies various matrix types as well: NumericMatrix, IntegerMatrix, CharacterMatrix, LogicalMatrix

  • Use .nrow() and .ncol() methods to get dimensions
  • Use () not [] for indexing
NumericVector rowSumsCpp(NumericMatrix x) {
  int nrow = x.nrow();
  int ncol = x.ncol();
  NumericVector out(nrow);

  for (int i = 0; i < nrow; i++) {
    double total = 0;
    for (int j = 0; j < ncol; j++) {
      total += x(i, j);
    }
    out[i] = total;
  }
  return out;
}

Functions #

You can pass, use, and return R functions from within C++. Note the _[] construction for named arguments.

Function rnorm("rnorm");

rnorm(100, _["mean"]=10.2, _["sd"]=3.2);

Other Useful Classes #

List, DataFrame, Environment are often directly useful, analogously to how we use them in R.

(Note: DataFrames are not easy to use as input because of static typing.)

There are other specialized classes in the library that are less commonly used but are valuable when you need them: SEXP, DottedPair, ….

STL Interface #

One of the big advantages of C++ is a fantastic and well-tuned run-time library. The STL is the center of this. Rcpp plays nicely with the STL.

An important type in the STL is the iterator over some collection.

double iteratorSum(NumericVector x) {
    double total = 0;
    NumericVector::iterator it;
    for(it = x.begin(); it != x.end(); ++it) {
      total += *it;
    }
    return total;
}

Note operations

.begin()
iterator pointing to beginning of collection
.end()
iterator pointing just past the end
= or !
equality checks (cf. distance)
++
advance (also -- for bidirectional iterators)
*
dereferencing.

Algorithms:

// sum a vector from beginning to end
double s = std::accumulate(x.begin(), x.end(), 0.0);

// prod of elements from beginning to end
int p = std::accumulate(vec.begin(),
                        vec.end(), 1, std::multiplies<int>());

// inner_product to compute sum of squares
double s2 = std::inner_product(res.begin(), res.end(),
                               res.begin(), 0.0);

Another example:

IntegerVector findInterval(NumericVector x, NumericVector breaks) {
    IntegerVector out(x.size());

    NumericVector::iterator it, pos;
    IntegerVector::iterator out_it;

    for(it = x.begin(), out_it = out.begin(); it != x.end();
        ++it, ++out_it) {
      pos = std::upper_bound(breaks.begin(), breaks.end(), *it);
      *out_it = std::distance(breaks.begin(), pos);
    }

    return out;
}

Rcpp Syntactic Sugar #

Rcpp provides R-like ``syntactic sugar’’ for operating on vectors in a concise way.

Commonly types of functions:

  • Math functions: abs(), ceil(), sin(), cos(), …
  • Scalar summaries: min(), max(), sum(), …
  • Vector summaries: cumsum(), diff(), pmin(), pmax()
  • Search: match(), which_max(), duplicates(), unique(), …
  • Distribution functions (d, q, p, and r versions)
  • Vector views: head(), tail(), rev(), seq_along(), seq_len(), rep_each(), rep_len()

And More #

There are many additional deep features in Rcpp that are useful in practice. Check the resources.

There are also many plugins and packages that are easy to include and use:

  • Fast matrix computations (Armadillo)
  • Eigenvalue Problems (Eigen)
  • Optimization
  • Monte Carlo Simulation
  • Numpy interface
  • Boost interfaces

See http://rcpp.org/ for links to these packages.