Vectors of NA values are parsed as boolean

[hebu]

We are using an R algorithm that usually should return a vector of doubles or NA values. Sometimes it returns a vector of NA values only ... e.g. c(NA, NA, NA, NA, NA)

For some reason, REngine.GetSymbol() would not return the NA-only vector as a list of doubles being Double.NaN (as I would expect), but as a list of Booleans being all "true".

Is this an issue with RDotNet or an issue with R's SEXPREC type? As RServe has similar behaviour, I suspect R itself.

(If my question seems weird: Sorry, I am new to R and RDotNet)

Versions used: RDotNet 1.5.5 / R 3.1.2

[skyguy94]

R has six types of vectors, but only four are commonly used. These are logical, integer, numeric, and character (plus complex and raw). Since a vector, unlike a list, has to be all of the same type R will coerce down to the most flexible format which is usually character. However, according to the docs, NA is defined to be a logical constant of length 1, so a vector of c(NA, NA, NA, NA) is automatically logical. You can verify this with R studio:

> f <- c(NA, NA, NA, NA)
> typeof(f)
[1] "logical"

However, by the coercion rules, you can go from logical->integer->double->character with logical being least flexible and character being most flexible.

So, via RStudio

> f <- as.double(c(NA, NA, NA, NA))
> typeof(f)
[1] "double"

or via R.Net

using RDotNet;

namespace RDotNetTester
{
    internal class Program
    {
        private static void Main()
        {
            var engine = REngine.GetInstance();
            engine.Evaluate("f <- c(NA, NA, NA, NA)");
            var f = engine.GetSymbol("f");
            var n = f.AsNumeric();
        }
    }
}

[skyguy94]

Also, I think if you use the term SEXPREC, you are miles ahead of many R.Net users.

[hebu]

Thanks for your great explanation, @skyguy94 – I will know how to fix my C# code properly now.

[skyguy94]

I'm happy to help. I don't get to touch R much in my day job. Posting about it helps me internalize the fundamental details. If you want a good book for programmers using R, I recommend Advanced R by Hadley Wickham.