min and max in Rcpp programs

Question

I was converting an R function into Rcpp function. It is all fine but I have difficulties in using the standard max and min function. The code below:

#include <math.h>
#include <RcppArmadillo.h>
#include <algorithm>
#include <iostream>

// [[Rcpp::depends(RcppArmadillo)]]

using namespace Rcpp;
using namespace arma;
using namespace std;

double f4RandomC(double x, double a, double b) {
  double out, temp;

  temp=(log( (a*(1-x)) / (a+x) )) /log(b) ;
  out= std::min(1,temp );
  return out;
}

returns me error "no matchinf function for call min(int, &double). If possible, I would lioke to use the std:: library min function

Answer 1

Just change std::min(1,temp) to std::min(1.0,temp):

#include <cmath>
#include <Rcpp.h>
// [[Rcpp::export]]
double f4RandomC(double x, double a, double b) {
  double out, temp;

  temp =(log( (a*(1-x)) / (a+x) )) /log(b) ;
  out = std::min(1.0,temp );
  return out;
}

I'm assuming this has to do with the template definition of std::min

template <class T> const T& min (const T& a, const T& b);

which is defined in terms of only one type (T), whereas you were passing it two data types (int and double).

Or since you are only comparing two values, you could do this a little more concisely by replacing std::min with the ternary operator (?:):

double f4RandomC(double x, double a, double b) {
  double temp;

  temp =(log( (a*(1-x)) / (a+x) )) /log(b) ;
  return temp < 1 ? temp : 1;
}

I'm guessing the type deduction is a little more flexible for operator< than std::min.

Two other options with std::min:

// [[Rcpp::export]]
double f4RandomC2(double x, double a, double b) {
  double out, temp;
  int z = 1;
  temp =(log( (a*(1-x)) / (a+x) )) /log(b) ;
  out = std::min( static_cast<double>(z),temp );
  return out;
}

// [[Rcpp::export]]
double f4RandomC3(double x, double a, double b) {
  double out, temp;
  int z = 1;
  temp =(log( (a*(1-x)) / (a+x) )) /log(b) ;
  out = std::min<double>( z,temp );
  return out;
}

Although in this case it's certainly much easier to just change 1 to 1.0 than to (unnecessarily) define int z just to cast it to a double anyways later on.

You can learn a lot by reading through function / class definitions (as with most programming languages) - cplusplus.com and cppreference.com are pretty standard sources - and it will often make compiler errors seem much less cryptic.