Hi, I have been using R for close to two years now and have grown quite comfortable with the language. I am presently trying to implement an optimization routine in R (Newton Rhapson). I have some R functions that calculate the gradient and hessian (pre requisite matrices) fairly efficiently. Now, I have to call this function iteratively until some convergance criterion is reached. I think the standard method of doing this in most programming languages is a while loop. However, I know R can get pretty slow when you use loops. In order to make this efficient, I want to transfer this part of my code to a more efficient programming language like c++ or c. However, I have been trying to learn this all day without any luck. I found a package called Rcpp that makes this easier. However, it seems some functional knowledge of writing R packages is a pre requisite. I tried to follow the standard manual for doing this, but could not find a simple example to get me started. I know I am supposed to make a cpp file and put it some where before it can be called from R, but I'm confused as to how this can be done. My requirement is to start with a parameter vector, update it according to the gradient and hessian, check if the parameter satisfies some convergance criterion and continue doing this until it does. Is there a way to efficiently do this through an R function (replicate?). The problem is that the number of iterations is not fixed. If there is no function in R, is there a way I can quickly use Rcpp or some thing to have this last part of my code in a C or C++ program which repeatedly calls my R functions for updating the parameters? -- Thanks in advance, Rohit Mob: 91 9819926213 [[alternative HTML version deleted]]
On Feb 6, 2011, at 10:28 AM, Rohit Pandey wrote:> Hi, > > I have been using R for close to two years now and have grown quite > comfortable with the language. I am presently trying to implement an > optimization routine in R (Newton Rhapson). I have some R functions > that > calculate the gradient and hessian (pre requisite matrices) fairly > efficiently. Now, I have to call this function iteratively until some > convergance criterion is reached. I think the standard method of > doing this > in most programming languages is a while loop. However, I know R can > get > pretty slow when you use loops. In order to make this efficient, I > want to > transfer this part of my code to a more efficient programming > language like > c++ or c. However, I have been trying to learn this all day without > any > luck. I found a package called Rcpp that makes this easier. However, > it > seems some functional knowledge of writing R packages is a pre > requisite. I > tried to follow the standard manual for doing this, but could not > find a > simple example to get me started. I know I am supposed to make a cpp > file > and put it some where before it can be called from R, but I'm > confused as to > how this can be done.Dirk Eddelbuettel has 8 vignettes linked from his Rcpp page: http://dirk.eddelbuettel.com/code/rcpp.html And reading further you will see about 20 packages written with Rcpp. And there is a mailing list link which I had originally sought on the main Mailing List page (unsuccessfully): http://lists.r-forge.r-project.org/pipermail/rcpp-devel/ -- David.> > My requirement is to start with a parameter vector, update it > according to > the gradient and hessian, check if the parameter satisfies some > convergance > criterion and continue doing this until it does. Is there a way to > efficiently do this through an R function (replicate?). The problem > is that > the number of iterations is not fixed. If there is no function in R, > is > there a way I can quickly use Rcpp or some thing to have this last > part of > my code in a C or C++ program which repeatedly calls my R functions > for > updating the parameters? > > -- > Thanks in advance, > Rohit > Mob: 91 9819926213 > > [[alternative HTML version deleted]] > > ______________________________________________ > R-help at r-project.org mailing list > https://stat.ethz.ch/mailman/listinfo/r-help > PLEASE do read the posting guide http://www.R-project.org/posting-guide.html > and provide commented, minimal, self-contained, reproducible code.David Winsemius, MD West Hartford, CT
On 02/06/2011 07:28 AM, Rohit Pandey wrote:> Hi, > > I have been using R for close to two years now and have grown quite > comfortable with the language. I am presently trying to implement an > optimization routine in R (Newton Rhapson). I have some R functions that > calculate the gradient and hessian (pre requisite matrices) fairly > efficiently. Now, I have to call this function iteratively until some > convergance criterion is reached. I think the standard method of doing this > in most programming languages is a while loop. However, I know R can get > pretty slow when you use loops. In order to make this efficient, I want toHave you looked at ?optim or ?optimize or perhaps ?nlm ? The bulk of the computation is in repeatedly evaluating the function and its gradient, rather than the machinery to iterate to convergence. Maybe these parts of your code can be written more efficiently? Martin> transfer this part of my code to a more efficient programming language like > c++ or c. However, I have been trying to learn this all day without any > luck. I found a package called Rcpp that makes this easier. However, it > seems some functional knowledge of writing R packages is a pre requisite. I > tried to follow the standard manual for doing this, but could not find a > simple example to get me started. I know I am supposed to make a cpp file > and put it some where before it can be called from R, but I'm confused as to > how this can be done. > > My requirement is to start with a parameter vector, update it according to > the gradient and hessian, check if the parameter satisfies some convergance > criterion and continue doing this until it does. Is there a way to > efficiently do this through an R function (replicate?). The problem is that > the number of iterations is not fixed. If there is no function in R, is > there a way I can quickly use Rcpp or some thing to have this last part of > my code in a C or C++ program which repeatedly calls my R functions for > updating the parameters? >-- Computational Biology Fred Hutchinson Cancer Research Center 1100 Fairview Ave. N. PO Box 19024 Seattle, WA 98109 Location: M1-B861 Telephone: 206 667-2793
On 6 February 2011 at 20:58, Rohit Pandey wrote: | Hi, | | I have been using R for close to two years now and have grown quite | comfortable with the language. I am presently trying to implement an | optimization routine in R (Newton Rhapson). I have some R functions that | calculate the gradient and hessian (pre requisite matrices) fairly | efficiently. Now, I have to call this function iteratively until some | convergance criterion is reached. I think the standard method of doing this | in most programming languages is a while loop. However, I know R can get | pretty slow when you use loops. In order to make this efficient, I want to | transfer this part of my code to a more efficient programming language like | c++ or c. However, I have been trying to learn this all day without any | luck. I found a package called Rcpp that makes this easier. However, it | seems some functional knowledge of writing R packages is a pre requisite. I What gave you that impression? Here is a counter-example, using the packages inline (for cxxfunction) and Rcpp: R> library(inline) R> src <- 'std::cout << "Hello C++_From_R World" << std::endl; return(Rcpp::wrap(42));' R> rohit <- cxxfunction(signature(), src, plugin="Rcpp") R> rohit() Hello C++_From_R World [1] 42 R> This compiled, linked and loaded a C++ routine built from the two-statement program submitted as character variable. The Rcpp documentation, including its eight vignettes, is full of other examples. Start with Rcpp-introduction and maybe the Rcpp-FAQ. | tried to follow the standard manual for doing this, but could not find a | simple example to get me started. I know I am supposed to make a cpp file | and put it some where before it can be called from R, but I'm confused as to | how this can be done. | | My requirement is to start with a parameter vector, update it according to | the gradient and hessian, check if the parameter satisfies some convergance | criterion and continue doing this until it does. Is there a way to | efficiently do this through an R function (replicate?). The problem is that | the number of iterations is not fixed. If there is no function in R, is | there a way I can quickly use Rcpp or some thing to have this last part of | my code in a C or C++ program which repeatedly calls my R functions for | updating the parameters? Give the example above a first try, and then read some more. The archives of the rcpp-devel (CC'ed; post there for follow-ups after subscribing) list are full of examples, and the CRAN page for Rcpp lists almost two dozen other packages using Rcpp giving you plenty of examples should you want to write a package using Rcpp. Several of these packages do optimization giving you examples of you to pass parameters etc pp. Hope this helps, Dirk | | -- | Thanks in advance, | Rohit | Mob: 91 9819926213 | | [[alternative HTML version deleted]] | | ______________________________________________ | R-help at r-project.org mailing list | https://stat.ethz.ch/mailman/listinfo/r-help | PLEASE do read the posting guide http://www.R-project.org/posting-guide.html | and provide commented, minimal, self-contained, reproducible code. -- Dirk Eddelbuettel | edd at debian.org | http://dirk.eddelbuettel.com
Hi Rohit, It is fine if you are using the Newton's method just as an excuse to learn Rcpp. If, however, your main goal is to develop a package to implement Newton's method, then you need to stop and look at the numerous optimization packages available in R. A good place to start would be the "optimx" package, which unifies a large number of optimiaztion tools under one umbrella. Hope this helps, Ravi. ____________________________________________________________________ Ravi Varadhan, Ph.D. Assistant Professor, Division of Geriatric Medicine and Gerontology School of Medicine Johns Hopkins University Ph. (410) 502-2619 email: rvaradhan at jhmi.edu ----- Original Message ----- From: Rohit Pandey <rohitpandey576 at gmail.com> Date: Sunday, February 6, 2011 11:33 am Subject: [R] Help with integrating R and c/c++ To: r-help at r-project.org> Hi, > > I have been using R for close to two years now and have grown quite > comfortable with the language. I am presently trying to implement an > optimization routine in R (Newton Rhapson). I have some R functions that > calculate the gradient and hessian (pre requisite matrices) fairly > efficiently. Now, I have to call this function iteratively until some > convergance criterion is reached. I think the standard method of > doing this > in most programming languages is a while loop. However, I know R can > get > pretty slow when you use loops. In order to make this efficient, I > want to > transfer this part of my code to a more efficient programming > language like > c++ or c. However, I have been trying to learn this all day without any > luck. I found a package called Rcpp that makes this easier. However, > it > seems some functional knowledge of writing R packages is a pre > requisite. I > tried to follow the standard manual for doing this, but could not > find a > simple example to get me started. I know I am supposed to make a cpp > file > and put it some where before it can be called from R, but I'm > confused as to > how this can be done. > > My requirement is to start with a parameter vector, update it > according to > the gradient and hessian, check if the parameter satisfies some convergance > criterion and continue doing this until it does. Is there a way to > efficiently do this through an R function (replicate?). The problem > is that > the number of iterations is not fixed. If there is no function in R, > is > there a way I can quickly use Rcpp or some thing to have this last > part of > my code in a C or C++ program which repeatedly calls my R functions for > updating the parameters? > > -- > Thanks in advance, > Rohit > Mob: 91 9819926213 > > [[alternative HTML version deleted]] > > ______________________________________________ > R-help at r-project.org mailing list > > PLEASE do read the posting guide > and provide commented, minimal, self-contained, reproducible code.
Hi Christopher/ Dirk,
Thank you very much for your replys. I think the idea of using inline as you
suggest is the best way to start off with using c++ with R. I went through
your examples and also plenty I found on the net. I have been trying to run
some of these over the past few days, but have consistently been getting
this error message as soon as I run the 'cfunction' or
'cxxfunction'. For
example, in the syntax that Dirk sent,
R> library(inline) #Runs perfectly
R> src <- 'std::cout << "Hello C++_From_R World"
<< std::endl;
return(Rcpp::wrap(42));' #Runs perfectly
R> rohit <- cxxfunction(signature(), src, plugin="Rcpp")
Now, as soon as I run this line, R spills a whole lot of text out and gives
an error and a warning:
ERROR(s) during compilation: source code errors or compiler configuration
errors!
Program source:
1:
2: // includes from the plugin
3:
4: #include <Rcpp.h>
5:
6:
7: #ifndef BEGIN_RCPP
8: #define BEGIN_RCPP
9: #endif
10:
11: #ifndef END_RCPP
12: #define END_RCPP
13: #endif
14:
15: using namespace Rcpp;
16:
17:
18: // user includes
19:
20:
21: // declarations
22: extern "C" {
23: SEXP file59046688( ) ;
24: }
25:
26: // definition
27:
28: SEXP file59046688( ){
29: BEGIN_RCPP
30: std::cout << "Hello C++_From_R World" << std::endl;
return(Rcpp::wrap(42));
31: END_RCPP
32: }
33:
34:
Error in compileCode(f, code, language = language, verbose = verbose) :
Compilation ERROR, function(s)/method(s) not created!
In addition: Warning message:
running command 'C:\PROGRA~1\R\R-212~1.1/bin/i386/R CMD SHLIB
file59046688.cpp 2> file59046688.cpp.err.txt' had status 1
The "file59046688.cpp 2" changes every time I run a different
function, but
the problem seems to be the same.
I installed and loaded the inline package (0.3.8) and then the Rcpp package
(0.9.0). I also tried reversing the order in which I load these, but still
no luck. I think if I can get just one of these programs to work, I will be
on my way. Can any of you tell me what I might be doing wrong?
For your question on what exacly I require, Christopher - I just need to use
a while loop. I have always been able to substitute for loops with some of
the apply functions in R, but can't seem to be able to replace the while
with a more efficient function. But the things that are required inside the
while loop, I have already implemented in R efficiently. So, I thought of
transfering just the while loop to a language that is faster with loops.
Thanks in advance.
On Mon, Feb 7, 2011 at 6:21 AM, Wray, Christopher <
christopher.wray.10@ucl.ac.uk> wrote:
> As Dirk says, using "inline" makes it real simple to start and to
prototype
> code.
>
> You mention you have R functions you wish to "call" via Rcpp. Im
not
> certain I fully understand what you require here, but it is pretty simple
to
> pass R-side functions to C++ via Rcpp, and similarly its simple to send
> compiled functions back to R-side as external pointers (and reuse them
> elsewhere). Here is a toy example using a simple "user" function
defined on
> the R-side (user_F).
>
> This is evaluated in R, passed as a "function" parameter to
compiled C++
> function and evaluated there, and then a compiled version of the function
is
> passed back to R as an external pointer, which you can send back to the C
> side and evaluate:
>
> <R> user_F=function(v){sum((1-v)*exp(-0.5*v))}
>
> cpp <- '
> NumericVector numvec(xvec);
> NumericVector RetVec;
> Function userR_f(fun);
> List ATR;
> typedef SEXP (*h_ptr)(SEXP);
> RetVec = userR_f(numvec);
> ATR["Fn_ptr"]=XPtr<h_ptr> (new h_ptr(&fme));
> ATR["Fn_VAL"]=RetVec;
> return ATR;
> '
> inc<-'
> using namespace Rcpp;
> SEXP fme(SEXP x){
> NumericVector xx(x);
> NumericVector yy=(1-xx)*exp(-0.5*xx);
> double Ans=sum(yy);
> return wrap(Ans);
> }
> '
> <R> FN_CSide <- cxxfunction(signature(xvec = "numeric",
fun = "function"
> ),cpp, , plugin = "Rcpp", includes = inc)
>
> cppP <- '
> typedef SEXP (*h_ptr)(SEXP);
> XPtr<h_ptr>p(x);
> NumericVector yy(y);
> yy=(*p)(yy);
> return yy;
> '
> <R> px<-cxxfunction(signature(x = "externalptr", y =
"numeric" ),cppP, ,
> plugin = "Rcpp", includes = "using namespace Rcpp; ")
>
> <R> R_side=FN_CSide(seq(1,5),user_F)
> <R> user_F(seq(1,5))
> [1] -1.548486
> <R> R_side
> $Fn_ptr
> <pointer: 0x0964c608>
> $Fn_VAL
> [1] -1.548486
> <R> px(R_side$Fn_ptr,seq(1,5))
> [1] -1.548486
>
> I've broken it out to make the logic explicit. The above is sloppy -
and
> makes no attempt to treat memory de/allocation/gc/protection issues (which
> you should consider) especially if you pass objects around, but it does try
> to show you how the mechanics work.
>
> Aside from looking up the Rcpp::sugar functionality used in places above,
> you might also want to look at package "RcppArmadillo", if
functions and
> calculations involve linear algebra calcs/matrix operations. Hopefully this
> gives you some food-for-thought to delve further into the
> archives/documentation to find what you need.
> chris
>
>
>
> ________________________________________
> From: rcpp-devel-bounces@r-forge.wu-wien.ac.at [
> rcpp-devel-bounces@r-forge.wu-wien.ac.at] on behalf of Dirk Eddelbuettel [
> edd@debian.org]
> Sent: 06 February 2011 18:18
> To: Dirk Eddelbuettel
> Cc: Rohit Pandey; rcpp-devel
> Subject: Re: [Rcpp-devel] [R] Help with integrating R and c/c++
>
> [ Now resending to rcpp-devel as I had said below I would --Dirk ]
>
> On 6 February 2011 at 20:58, Rohit Pandey wrote:
> | Hi,
> |
> | I have been using R for close to two years now and have grown quite
> | comfortable with the language. I am presently trying to implement an
> | optimization routine in R (Newton Rhapson). I have some R functions that
> | calculate the gradient and hessian (pre requisite matrices) fairly
> | efficiently. Now, I have to call this function iteratively until some
> | convergance criterion is reached. I think the standard method of doing
> this
> | in most programming languages is a while loop. However, I know R can get
> | pretty slow when you use loops. In order to make this efficient, I want
> to
> | transfer this part of my code to a more efficient programming language
> like
> | c++ or c. However, I have been trying to learn this all day without any
> | luck. I found a package called Rcpp that makes this easier. However, it
> | seems some functional knowledge of writing R packages is a pre requisite.
> I
>
> What gave you that impression?
>
> Here is a counter-example, using the packages inline (for cxxfunction) and
> Rcpp:
>
> R> library(inline)
> R> src <- 'std::cout << "Hello C++_From_R World"
<< std::endl;
> return(Rcpp::wrap(42));'
> R> rohit <- cxxfunction(signature(), src, plugin="Rcpp")
> R> rohit()
> Hello C++_From_R World
> [1] 42
> R>
>
> This compiled, linked and loaded a C++ routine built from the two-statement
> program submitted as character variable.
>
> The Rcpp documentation, including its eight vignettes, is full of other
> examples. Start with Rcpp-introduction and maybe the Rcpp-FAQ.
>
> | tried to follow the standard manual for doing this, but could not find a
> | simple example to get me started. I know I am supposed to make a cpp file
> | and put it some where before it can be called from R, but I'm
confused as
> to
> | how this can be done.
> |
> | My requirement is to start with a parameter vector, update it according
> to
> | the gradient and hessian, check if the parameter satisfies some
> convergance
> | criterion and continue doing this until it does. Is there a way to
> | efficiently do this through an R function (replicate?). The problem is
> that
> | the number of iterations is not fixed. If there is no function in R, is
> | there a way I can quickly use Rcpp or some thing to have this last part
> of
> | my code in a C or C++ program which repeatedly calls my R functions for
> | updating the parameters?
>
> Give the example above a first try, and then read some more. The archives
> of
> the rcpp-devel (CC'ed; post there for follow-ups after subscribing)
list
> are
> full of examples, and the CRAN page for Rcpp lists almost two dozen other
> packages using Rcpp giving you plenty of examples should you want to write
> a
> package using Rcpp. Several of these packages do optimization giving you
> examples of you to pass parameters etc pp.
>
> Hope this helps, Dirk
>
> |
> | --
> | Thanks in advance,
> | Rohit
> | Mob: 91 9819926213
> |
> | [[alternative HTML version deleted]]
> |
> | ______________________________________________
> | R-help@r-project.org mailing list
> | https://stat.ethz.ch/mailman/listinfo/r-help
> | PLEASE do read the posting guide
>
http://www.R-project.org/posting-guide.html<http://www.r-project.org/posting-guide.html>
> | and provide commented, minimal, self-contained, reproducible code.
>
> --
> Dirk Eddelbuettel | edd@debian.org | http://dirk.eddelbuettel.com
> _______________________________________________
> Rcpp-devel mailing list
> Rcpp-devel@lists.r-forge.r-project.org
> https://lists.r-forge.r-project.org/cgi-bin/mailman/listinfo/rcpp-devel
--
Thanks,
Rohit
Mob: 91 9819926213
[[alternative HTML version deleted]]