Checking Null and NA of a vector in Rcpp

Question

I'm trying to evaluate the sum of a vector (y) conditional on whether the values of a second nullable vector (r) is NA or not. If the second vector r is NULL, all of the values of y should be summed. If all elements of r is NA, function should return NA. Please see the end of the text for the desired output.

I tried the following code first:

library(Rcpp)
cppFunction('double foo(NumericVector y, Rcpp::Nullable<Rcpp::IntegerVector> r = R_NilValue) {
  double output = 0;
  bool return_na = !Rf_isNull(r);
  int y_count = y.size();
  for (int i = 0; i < y_count; i++) {
    if (Rf_isNull(r)  || !R_IsNA(r[i])) {
    //// if (Rf_isNull(r)  || !R_IsNA(as<IntegerVector>(r)[i])) {
      if (!Rf_isNull(r))
        Rcout << R_IsNA(as<IntegerVector>(r)[i]) << " - "<< as<IntegerVector>(r)[i] << std::endl;
      output = output + y[i];
      return_na = false;
    } 
  }
  if (return_na) 
    return NA_REAL;
  return output;
}')

This gave me the following error:

 error: invalid use of incomplete type 'struct SEXPREC'
     if (Rf_isNull(r)  || !R_IsNA(r[i])) {
                                     ^

In order to solve it, I used if (Rf_isNull(r) || !R_IsNA(as<IntegerVector>(r)[i])) { instead. But this time, when converting to an integer vector, NA values are converted to a number and R_IsNA() test gives a false positive.

Here is the expected output that I want.

foo(1:4, NULL) #  <- This should return 10 = 1 + 2 + 3 + 4
foo(1:4, c(1, 1, 1, 1)) #  <- This should return 10 = 1 + 2 + 3 + 4
foo(1:4, c(1, 1, NA, 1)) #  <- This should return 7 = 1 + 2 + 4
foo(1:4, c(NA, NA, NA, NA)) # <- This should return NA

How can I get the function that I want? (This example is simplified, I'm not particularly interested in sum function. Instead, I'm interested in checking NA and NULL simultaneously as given in the example.)

Are you aware of this classic answer by Kevin? https://stackoverflow.com/questions/26241085/rcpp-function-check-if-missing-value/26262984#26262984 — Dirk Eddelbuettel, Mar 06 '20 at 18:36
Thanks for pointing out that post. I reviewed Kevin's answer before. But in this case I cannot find it's use for my particular case. The next post (from Alexis) was more helpful in my case. Previously, instead of `!R_IsNA(as(r)[i])`, I tried `!is_na(r[i])` or `!is_na(as(r)[i])` but both raised errors. But when I specify the type of `NA`, like this `!IntegerVector::is_na(as(r)[i])`, it worked. — HBat, Mar 06 '20 at 19:38

Ralf Stubner · Accepted Answer · 2020-03-07T17:08:47.137

Three suggestions:

Use Rcpp instead of R's C API.
Return early when r is NULL.
Create a LogicalVector before looping through the input vector.

#include <Rcpp.h>

// [[Rcpp::export]]
double foo(Rcpp::NumericVector y, Rcpp::Nullable<Rcpp::IntegerVector> r = R_NilValue) {
    if (r.isNull())
        return Rcpp::sum(y);

    Rcpp::LogicalVector mask = Rcpp::is_na(r.as());
    if (Rcpp::is_true(Rcpp::all(mask))) 
        return NA_REAL;

    double output = 0.0;
    int y_count = y.size();
    for (int i = 0; i < y_count; ++i) {
        if (!mask[i]) {
            output += y[i];
        } 
    }
    return output;
}

/***R
foo(1:4, NULL) #  <- This should return 10 = 1 + 2 + 3 + 4
foo(1:4, c(1, 1, 1, 1)) #  <- This should return 10 = 1 + 2 + 3 + 4
foo(1:4, c(1, 1, NA, 1)) #  <- This should return 7 = 1 + 2 + 4
foo(1:4, c(NA, NA, NA, NA)) # <- This should return NA
*/

Result:

> Rcpp::sourceCpp('60569482.cpp')

> foo(1:4, NULL) #  <- This should return 10 = 1 + 2 + 3 + 4
[1] 10

> foo(1:4, c(1, 1, 1, 1)) #  <- This should return 10 = 1 + 2 + 3 + 4
[1] 10

> foo(1:4, c(1, 1, NA, 1)) #  <- This should return 7 = 1 + 2 + 4
[1] 7

> foo(1:4, c(NA, NA, NA, NA)) # <- This should return NA
[1] NA

Further suggestion:

Use the mask for sub-setting y.

#include <Rcpp.h>

// [[Rcpp::export]]
double foo(Rcpp::NumericVector y, Rcpp::Nullable<Rcpp::IntegerVector> r = R_NilValue) {
    if (r.isNull())
        return Rcpp::sum(y);

    Rcpp::LogicalVector mask = Rcpp::is_na(r.as());
    if (Rcpp::is_true(Rcpp::all(mask))) 
        return NA_REAL;

    Rcpp::NumericVector tmp = y[!mask];
    return Rcpp::sum(tmp);
}

Rather than using a mask, you can use `if(y[i] != NA_INTEGER)` in the loop, and keep track if any elements are not NA with another variable. That should be overall more efficient. — thc, Mar 09 '20 at 19:36
@thc Alternatively one could use `!Rcpp::is_na(y[i])`. I find the last version easier to reason about, which **can** be more important than the last bit of performance. We don't know if it is the case here ... — Ralf Stubner, Mar 09 '20 at 21:05

Checking Null and NA of a vector in Rcpp

1 Answers1