Rcpp 0.7.2
By romain francois on Wednesday, January 13 2010, 09:44 - Rcpp - Permalink
Rcpp 0.7.2 is out, checkout Dirk's blog for details
selected highlights from this new version:
character vectors
if one wants to mimic this R code in C
> x <- c( "foo", "bar" )one ends up with this :
SEXP x = PROTECT( allocVector( STRSXP, 2) ) ; SET_STRING_ELT( x, 0, mkChar( "foo" ) ) ; SET_STRING_ELT( x, 1, mkChar( "bar" ) ) ; UNPROTECT(1) ; return x ;
Rcpp lets you express the same like this :
CharacterVector x(2) ; x[0] = "foo" ; x[1] = "bar" ;
or like this if you have GCC 4.4 :
CharacterVector x = { "foo", "bar" } ;
environments, functions, ...
Now, we try to mimic this R code in C :rnorm( 10L, sd = 100 )You can do one of these two ways in Rcpp :
Environment stats("package:stats") ; Function rnorm = stats.get( "rnorm" ) ; return rnorm( 10, Named("sd", 100 ) ) ;
or :
Language call( "rnorm", 10, Named("sd", 100 ) ) ; return eval( call, R_GlobalEnv ) ;
and it will get better with the next release, where you will be able to just call call.eval()
and stats["rnorm"]
.
Using the regular R API, you'd write these liks this :
SEXP stats = PROTECT( R_FindNamespace( mkString("stats") ) ) ; SEXP rnorm = PROTECT( findVarInFrame( stats, install("rnorm") ) ) ; SEXP call = PROTECT( LCONS( rnorm, CONS(ScalarInteger(10), CONS(ScalarReal(100.0), R_NilValue)))) ; SET_TAG( CDDR(call), install("sd") ) ; SEXP res = PROTECT( eval( call, R_GlobalEnv ) ); UNPROTECT(4) ; return res ;
or :
SEXP call = PROTECT( LCONS( install("rnorm"), CONS(ScalarInteger(10), CONS(ScalarReal(100.0), R_NilValue)))) ; SET_TAG( CDDR(call), install("sd") ) ; SEXP res = PROTECT( eval( call, R_GlobalEnv ) ); UNPROTECT(2) ; return res ;
Comments
Wow this makes me look like I want to try out Rcpp as playing in C looks a lot simpler.
Out of curiosity, suppose I were to do a simulation with linear regression and record the coefficients and standard error. Would doing such a loop using Rcpp be a lot faster than the loop in R? That is, all the benefits of doing things in C are still there. I'm wondering because the linear regression function lm() is still being but in C instead of R now. Thanks!
Vinh
Hi,
you might want to start with using the much faster lm.fit when doing simulations, it is a bit more work than just using lm or glm but it pays off.
For the speed, I don't know, it is worth a shot. Note that Rcpp has its mailing list where more people (at least one more) would be able to answer this.
Romain
Oh yes, I keep forgetting about lm.fit and glm.fit. I guess the only extra step would be model.matrix(), something along the line of:
fit <- lm.fit(y, x=model.matrix(y~ x1+...+xn), weights=myweights)
Anyhow, my hypothetical question is meant to see the benefits of calling R functions in Rcpp. I was lazy and wanted to know the added benefits without trying it myself =).