RDotNet vs R scripting

Question

When is it an advantage/disadvantage to be using RDotNet for making statistical calculations as opposed to generating an R script text file and running it from the application using e.g. Process.Start? Or is there some other better way?

I need to execute a large number of commands and have a feeling that sending them one by one to R takes a lot of time.

Answer 1

I'd say the following two scenario's are stereotypical:

1. .NET code and R code are quite separate, not a lot of interaction is needed between the R code and the .NET code. For example, the .NET code gathers some information, and launches a processing script on that, after which the .NET code picks up the results. In this case spawning an R process (Process.Start) is a simple way to get this working.
2. A lot of interaction is needed between the .NET code and the R code, the workflow consists of going back and forth between .NET and R often. In this case, a more heavy weight, flexible solution such as RDotNet makes a lot of sense. RDotNet allows more easy integration of the .NET code and the R code, with the price being that it is often harder to learn, harder to debug, and often needs to be updated for new versions of R etc.

Answer 2

R.NET currently can initilize once. Parallel execution will be problematic.

Would suggest use RScript.

Our solution based on this answer on stackoverflow Call R (programming language) from .net

With monor change, we send R code from string and save it to temp file, since user run custom R code when needed.

public static void RunFromCmd(string batch, params string[] args)
{
    // Not required. But our R scripts use allmost all CPU resources if run multiple instances
    lock (typeof(REngineRunner))
    {
        string file = string.Empty;
        string result = string.Empty;
        try
        {
            // Save R code to temp file
            file = TempFileHelper.CreateTmpFile();
            using (var streamWriter = new StreamWriter(new FileStream(file, FileMode.Open, FileAccess.Write)))
            {
                streamWriter.Write(batch);
            }

            // Get path to R
            var rCore = Registry.LocalMachine.OpenSubKey(@"SOFTWARE\R-core") ??
                        Registry.CurrentUser.OpenSubKey(@"SOFTWARE\R-core");
            var is64Bit = Environment.Is64BitProcess;
            if (rCore != null)
            {
                var r = rCore.OpenSubKey(is64Bit ? "R64" : "R");
                var installPath = (string)r.GetValue("InstallPath");
                var binPath = Path.Combine(installPath, "bin");
                binPath = Path.Combine(binPath, is64Bit ? "x64" : "i386");
                binPath = Path.Combine(binPath, "Rscript");
                string strCmdLine = @"/c """ + binPath + @""" " + file;
                if (args.Any())
                {
                    strCmdLine += " " + string.Join(" ", args);
                }
                var info = new ProcessStartInfo("cmd", strCmdLine);
                info.RedirectStandardInput = false;
                info.RedirectStandardOutput = true;
                info.UseShellExecute = false;
                info.CreateNoWindow = true;
                using (var proc = new Process())
                {
                    proc.StartInfo = info;
                    proc.Start();
                    result = proc.StandardOutput.ReadToEnd();
                }
            }
            else
            {
                result += "R-Core not found in registry";
            }
            Console.WriteLine(result);
        }
        catch (Exception ex)
        {
            throw new Exception("R failed to compute. Output: " + result, ex);
        }
        finally
        {
            if (!string.IsNullOrWhiteSpace(file))
            {
                TempFileHelper.DeleteTmpFile(file, false);
            }
        }
    }
}

Full blog post: http://kostylizm.blogspot.ru/2014/05/run-r-code-from-c-sharp.html