Looking at your pseudo code it appears that you know how many columns
are required in your matrix before run time. This situation lends
itself well to generating the columns of your matrix as vectors as these
can be manipulated with STL. Combining these vectors to a dataframe for
return is very straight forward. Here is some code from Derk pulled
from this list, here a, b, and c are Rcpp::NumericVector objects:
| > | >>> // create a new data frame
| > | >>> Rcpp::DataFrame NDF =
| > | >>> Rcpp::DataFrame::create(Rcpp::Named("a")=a,
| > | >>> Rcpp::Named("b")=b,
| > | >>> Rcpp::Named("c")=c);
Since a dataframe is really a list of vectors, this should be quite
efficient.
A couple of months ago I went through a number of posts to this list
with a subject something like "redimension a matrix". In that problem I
had to use a matrix object in Rcpp, because the number of columns was
only known at run time, based on characteristics identifiable in the
arguments passed to my Rcpp function. At the time of my posts I was
already convinced that I wanted to make an over estimate of the length
of my matrix allocate it by creating an Rcpp::NumericMatrix and then
expect to partially fill it by directly addressing the (row,col). This
result was clearly faster than my use of STL's .push_back() for dynamic
allocation, admittedly my simple mind could only deal with STL on Rcpp's
SEXP objects.
The problem that I then had was how to trim the resultant matrix to its
smaller identified size after the filling loop was complete.
Christian Gunning turned me on to using the RcppArmadillo package, which
gives more flexibility working with a matrix. But in the end I found
that the best result was to return my oversized matrix in a dataframe
(this is done just as if it were a vector as in the above code) which I
trimmed back in my R code.
_______________________________________________
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
