Back to Questions
10.6k
views6
comments
Using F# with R
f#serverbook

On the F# list server over at Microsoft Research, there was a question posted by Lars Johnson about using F# with R. For whatever reason, I'm not able to post to that list server. I'm not very good with list servers, so I'll post up my usage here and hope that someone could post this back to the list server or notify Lars.

For those that don't know, R is an analytics tool with an emphasis on rich statistical toolboxes. It was first produced at the old AT&T Bell Labs and was later commercialized by Stat-Sci (later to become Insightful) as S-Plus. R remains free and in the public domain.

In the list server thread, Juergen Van Gael mentioned two approaches for possible integration - RSOAP and R(D)COM. This post is for usage of R with R(D)COM with F#. The RSOAP approach is a little more involved, so I chose the R(D)COM path here. As its name implies, R(D)COM is a wrapper for R for usage with COM and DCOM. It was written by Thomas Baier and Erich Neuwirth.

My environment for use of F# with R on a Windows platform:

1. Windows XP SP2
2. .NET 2.0 - 2.0.50727
3. Visual Studio 2005 with SP1 - 8.0.50727.762 (SP.050727-7600)
4. F# 1.9.2.9
5. R 2.6.0 - [link:cran.us.r-project.org]
6. R(D)COM 2.50_pl1 - [link:rcom.univie.ac.at]

Note that VS2005 is not necessary as this could be done with the .NET Framework and F#.

After the above installation, we need to make an R(D)COM .NET wrapper so that F# or any .NET application can see the R(D)COM server. For this, I opened a VS2005 command prompt (dos window with the VS2005 environment variables properly set up). In that window, perform the following two steps. After you complete these steps, the command window is no longer needed.

1. Change directory to the default installation for the R(D)COM binaries. "cd C:\Program Files\R\(D)COM Server\bin".
2. Create the .NET interop wrapper. "TlbImp STATCONNECTORSRV.exe /out:Interop.STATCONNECTORSRVLib.dll". Note that while this is a standard operation for .NET with COM, Robert Pickering's book "Foundations of F#" covers COM usage with F# (and other interoperability) in Chapter 13.

Next, open up a copy of fsi and interact with the following code.

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28

 

//-----------------------------
#light
// attach a copy of the R(D)COM dll for statistics connections.
#r @"C:\Program Files\R\(D)COM Server\bin\Interop.STATCONNECTORSRVLib.dll"
// open the namespace
open Interop.STATCONNECTORSRVLib

// Create a connection instance
let SCCForR = new StatConnectorClass()
// Initialize that instance for usage with R
SCCForR.Init("R")
// Sample statements to modify the R environment.

// create an R variable named abc and assign it the value of 5
SCCForR.SetSymbol("abc", 5)

// Retrieve the value of the R variable named abc and assign that value to the F# value valueForabc
let valueForabc = SCCForR.GetSymbol("abc")

// Evaluate an expression in R and assign that value to an F# value aTestEvaluation
let aTestEvaluation = SCCForR.Evaluate("8 * sin(4)")

// Close the R connection
SCCForR.Close()
//-----------------------------

Note that the value for valueForabc should be 5. The value for aTestEvaluation should be -6.054419962. Also note that this sample does NOT clean up the COM object assigned to SCCForR. It only closes the connection to R.

This sample interop code was modified from the example given in the R(D)COM package. There are many other possibilities for R(D)COM, especially in the rich graphics libraries that R has to offer.

.

We offer top quality new motorhomes for sale and used motorhomes for sale. Our franchises for Auto-Trail, Bessacarr, Autocruise and Compass guarantee the .

====================================

By on 12/21/2010 2:49 AM ()

人类的世界发展到今天，真是无奇不有。最近看了一些关于计量经济学的贴子，发现了一个对于我们这一圈子的人来说，可能是极其陌生的计算机语言，叫做R。这个R语言的前身是S，是一种函数式脚本语言，主要就是为了做...

By on 5/24/2009 9:38 PM ()

I followed the steps listed above and was able to use the interactive mode (fsi i.e. alt-enter each step of the code) from Visual Studio. However, when I try to compile I get the following error:

System.IO.FileNotFoundException was unhandled

Message: Could not load file or assembly 'Interop.STATCONNECTORSRVLib, Version=1.1.0.0, Culture=neutral, PublicKeyToken=null' or one of its dependencies. The system cannot find the file specified.

Any ideas?

Thanks,

Dan

By on 3/3/2008 12:06 PM ()

I think quotations are a next logical step. Consider the following (naive) scenario "on the R side":

I have a set of observations from the US capital markets at a point in time. If I organize the data as a matrix, I'd place companies on rows, attributes on companies on columns. Examples of companies might be IBM, Citigroup, AT&T, ExxonMobil, etc. Examples of attributes might be total return (TR) (one month backlooking), market capitalization (MC), shares outstanding (SO), book to price (BTP), etc.

I'd like to run a linear model (ordinary least squares regression) on that data in R and see if there is any explanatory power. So the following would be needed:

1. A data.frame construct to place the observations. A data.frame is a "list of objects" construct where the primary object is a matrix and there are at least two other lists: a rownames and colnames list.

2. An expression to describe the model. In this case, I'd like to consider explaining total return from all of the other attributes. In R, if the attribute names are TR, MC, SO, BTP, the expression for the model would be "TR ~ MC SO BTP", i.e. TR is explained by MC SO BTP.

3. Given the model (as an R formula), I'd then attempt a linear fit in R using the lm function on the above mentioned data.frame. The output of the lm function is an object of class "lm".

4. I'd then evaluate the output of that lm object and get summary information, e.g. r squared, Beta, etc.

Now imagine that I have 15+ different models of that nature where some of those models also include a time element, e.g. the above mentioned data frame would need to be constructed for each month end period.

Given the above information, quotations would be a first step in managing much of the above. But a DSL is likely needed for sophisticated usage.

So in small steps, I'd consider some F# code with LINQ to load some of that data from a database. Based on the data available, I'd construct the above data.frame construct as a quotation (shuttling data to and from R is an issue; there is an RODBC package that could be used to do direct loading of data into R from a database; again another detailed topic). I'd also consider the model construction and the lm function call in another quotation. I'm not sure how I would handle the output of the lm model within the context of quotations.

To provide all of these details, I'd have to make another post on that topic or open a forum thread. No promises at the moment given my time available.

---O

By on 11/7/2007 8:34 AM ()

Any thoughts on mapping F# quotations to R?

By on 11/6/2007 6:03 PM ()
By on 11/6/2007 6:49 AM ()

    Topic tags

    IntelliFactory Offices Copyright (c) 2011-2012 IntelliFactory. All rights reserved.
    Home | Products | Consulting | Trainings | Blogs | Jobs | Contact Us | Terms of Use | Privacy Policy | Cookie Policy     		Built with WebSharper