Question

I am using R.Net 1.5 to attempt to do a simple forecast using ARIMA. I’ve tried with R 2.14 and R 2.15. I’m using Visual Studio 2012 target .NET 4 though I’ve also tried .NET 4.5 and Visual Studio 2010.

Here is a piece of the code I’ve written:

string rhome = System.Environment.GetEnvironmentVariable("R_HOME");
        if (string.IsNullOrEmpty(rhome))
            rhome = @"C:\Program Files\R\R-2.14.0";

        System.Environment.SetEnvironmentVariable("R_HOME", rhome);
        System.Environment.SetEnvironmentVariable("PATH", System.Environment.GetEnvironmentVariable("PATH") + ";" + rhome + @"\bin\x64");
        using (REngine engine = REngine.CreateInstance("RDotNet"))
        {

            engine.Initialize();

            NumericVector testGroup = engine.CreateNumericVector(submissions);
            engine.SetSymbol("testGroup", testGroup);
            engine.Evaluate("testTs <- c(testGroup)");
            NumericVector ts = engine.GetSymbol("testTs").AsNumeric();

            engine.Evaluate("tsValue <- ts(testTs, frequency=1, start=c(2010, 1, 1))");
            engine.Evaluate("library(forecast)");
            engine.Evaluate("arimaFit <- auto.arima(tsValue)");
            engine.Evaluate("fcast <- forecast(tsValue, h=36)");
            engine.Evaluate("plot(fcast)");

            NumericVector nv = engine.GetSymbol("fcast").AsNumeric();

It fails when I attempt to retrieve the numeric vector. TI get a few errors here. The first is “Error: object cannot be coerced to type ‘double'” and the second is “Error: caught access violation – continue with care”

If I retrieve the forecast as a GenericVector I get a list of RDotNet.SymbolicExpressions. I’ve looped through those to see what they contain and it does seem to be related to the ARIMA function but I cannot locate the actual forecast output. I can find the inputs and other related values as well as a bunch of lists of numbers that I can’t determine what they are.

If I run the script within Revolution I can see what the outputs SHOULD be and that is how I’m determining whether the outputs from R.Net are accurate. I suppose it’s possible R.Net is performing the forecast differently than Revolution (though unlikely, I think) and one of the outputs in the genericvector is indeed the correct output.

Here is the GenericVector initialization. I wrapped in a blanket try, catch just for debugging purposes: Inside the DynamicVector is where I can actually inspect the details.

GenericVector newVector = engine.GetSymbol("fcast").AsList();

            foreach (var vector in newVector)
            {
                try
                {
                    DynamicVector dv = vector.AsVector();
                }
                catch (Exception)
                {
                }

Answer 1

It is not a R.Net problem, you just try to run a wrong script. Let’s run your code on pure R:

> testTs <- c(1, 2, 3)
> tsValue <- ts(testTs, frequency=1, start=c(2010, 1, 1))
> library("forecast")
> arimaFit <- auto.arima(tsValue)
> fcast <- forecast(tsValue, h=36)
> plot(fcast)

Now class(fcast) equals to forecast:

> class(fcast)
[1] "forecast"
> as.numeric(fcast)
Error: (list) object cannot be coerced to type 'double'

fcast strucure:

> str(fcast)
List of 11
 $ method   : chr "Mean"
 $ level    : num [1:2] 80 95
 $ x        : Time-Series [1:3] from 2010 to 2012: 1 2 3
 $ xname    : chr "object"
 $ mean     : Time-Series [1:36] from 2013 to 2048: 2 2 2 2 2 2 2 2 2 2 ...
 $ lower    : mts [1:36, 1:2] -0.177 -0.177 -0.177 -0.177 -0.177 ...
  ..- attr(*, "dimnames")=List of 2
  .. ..$ : NULL
  .. ..$ : chr [1:2] "80%" "95%"
  ..- attr(*, "tsp")= num [1:3] 2013 2048 1
  ..- attr(*, "class")= chr [1:2] "mts" "ts"
 $ upper    : mts [1:36, 1:2] 4.18 4.18 4.18 4.18 4.18 ...
  ..- attr(*, "dimnames")=List of 2
  .. ..$ : NULL
  .. ..$ : chr [1:2] "80%" "95%"
  ..- attr(*, "tsp")= num [1:3] 2013 2048 1
  ..- attr(*, "class")= chr [1:2] "mts" "ts"
 $ model    :List of 4
  ..$ mu   : num 2
  ..$ mu.se: num 0.577
  ..$ sd   : num 1
  ..$ call : language meanf(x = object, h = h, level = level, fan = fan)
 $ lambda   : NULL
 $ fitted   : Time-Series [1:3] from 2010 to 2012: NA 1 1.5
 $ residuals: Time-Series [1:3] from 2010 to 2012: NA 1 1.5
 - attr(*, "class")= chr "forecast"