You have a point about push_back being amortized constant. In my code I do everything with Lists, which may explain why I needed to preallocate them. Unfortunately I forgot the motivation for using Lists instead of std:vector, which I must have considered at some point. In fact if you look at this commit ( https://github.com/RevolutionAnalytics/rmr2/commit/f8d4b0766ea71fe056a4b3eb0a4ef5ed0b7c6eec for the html-challenged) I had the push_back in a previous version, but it didn't work. I was using a vector<vector<RObject>> to hold the data, and if I can remember that hit some performance wall. The commit message doesn't really refresh my memory, but it says something about protecting SEXPs. That may be why I didn't want to use a vector<SEXP>. I may have had stability issues and, right or wrong, thought that change was helpful.
On Mon, May 4, 2015 at 12:23 PM, Tim Keitt <tke...@utexas.edu> wrote: > > > On Sun, May 3, 2015 at 11:29 PM, Antonio Piccolboni < > anto...@piccolboni.info> wrote: > >> Check here >> <https://github.com/RevolutionAnalytics/rmr2/blob/master/pkg/src/t-list.cpp> >> for something similar to Tim's solution that preallocates all vectors to >> avoid the costly push_back. Still needs the unlists in R, so it's expensive >> in that dimension, the number of lists in the output. >> > > I may have a way around the unlist part; still needs testing. > > push_back is amortized constant so only a little costly. > > THK > > >> >> >> Antonio >> >> On Sun, May 3, 2015 at 4:22 PM, Tim Keitt <tke...@utexas.edu> wrote: >> >>> A slightly improved version: >>> >>> library(Rcpp) >>> >>> code = ' >>> SEXP test(List a) >>> { >>> auto l = Rf_length(a[0]); >>> using svec = std::vector<SEXP>; >>> std::vector<svec> x(l); >>> for (List b : a) >>> { >>> if (b.size() != l) >>> stop("Ragged input"); >>> for (int i = 0; i != l; ++i) >>> x[i].push_back(b[i]); >>> } >>> return wrap(x); >>> }' >>> >>> f = cppFunction(code = code, plugins = "cpp11") >>> >>> res = lapply(f(list(list(T, 1, 'a'), >>> list(F, 2, 'b'))), >>> unlist) >>> >>> >>> On Sun, May 3, 2015 at 11:57 AM, Tim Keitt <tke...@utexas.edu> wrote: >>> >>>> Here's a really bare-bones version. Not very pretty or complete, but it >>>> does do the job. Could the 'unlist' part be converted to Rcpp? >>>> >>>> THK >>>> >>>> library(Rcpp) >>>> >>>> code = ' >>>> SEXP test(List a) >>>> { >>>> int l = Rf_length(a[0]); >>>> typedef std::vector<SEXP> svec; >>>> std::vector<svec> x(l); >>>> for (int i = 0; i != l; ++i) >>>> for (int j = 0; j != a.size(); ++j) >>>> { >>>> List b = a[j]; >>>> x[i].push_back(b[i]); >>>> } >>>> return wrap(x); >>>> }' >>>> >>>> f = cppFunction(code = code) >>>> >>>> res = lapply(f(list(list(1, 'a'), list(2, 'b'))), unlist) >>>> >>>> >>>> On Sat, May 2, 2015 at 1:18 PM, Dirk Eddelbuettel <e...@debian.org> >>>> wrote: >>>> >>>>> >>>>> On 2 May 2015 at 10:49, William Dunlap wrote: >>>>> | Since translation from R to Rcpp is "seamless" I will leave that to >>>>> you. >>>>> >>>>> One problem is with the strongly typed nature of C++. Rearranging >>>>> dynamicly >>>>> growing objects can be done. I think I used Boost's variant type a few >>>>> years. I am sure there are other possibilities. We should collect a >>>>> few and >>>>> compare. But in C++ please :) >>>>> >>>>> Dirk >>>>> >>>>> -- >>>>> http://dirk.eddelbuettel.com | @eddelbuettel | e...@debian.org >>>>> >>>> >>>> >>>> >>>> -- >>>> http://www.keittlab.org/ >>>> >>> >>> >>> >>> -- >>> http://www.keittlab.org/ >>> >>> _______________________________________________ >>> 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 >>> >> >> >> _______________________________________________ >> 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 >> > > > > -- > http://www.keittlab.org/ >
_______________________________________________ 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
