[Rcpp-devel] Non-contiguous submatrix

Simon Zehnder szehnder at uni-bonn.de
Mon Aug 26 13:14:23 CEST 2013


Hi Romain,

thanks for the quick reply! This is an answer I can work with!

Best

Simon


On Aug 26, 2013, at 12:33 PM, Romain Francois <romain at r-enthusiasts.com> wrote:

> Le 26 août 2013 à 12:19, Simon Zehnder <szehnder at uni-bonn.de> a écrit :
> 
>> Dear Rcpp::Users and Rcpp::Devels,
>> 
>> I am right now working on a relabeling algorithm where I have a function taking a permutation matrix of all permutations of the labels. Inside the function the values should be permuted in respect to a row of this matrix. In R I would use the following commands:
>> 
>> values <- matrix(rnorm(4), ncol = 2, nrow = 2, byrow = TRUE)
>> perm <- as.matrix(expand.grid(seq(1, 2), seq(1, 2)))
>> index <- perm[which(perm[, 1] != perm[, 2]),]
>> values[, index[, 1]]
>> values[, index[, 2]]
>> 
>> I found this thread on stackoverflow: http://stackoverflow.com/questions/8435169/selecting-a-non-contiguous-submatrix-in-rcpp. Are there any proceedings for non-contiguous submatrices in Rcpp? I couldn't find a hint in the Rcpp-book (neither in chapter 3 nor in chapter 8). I would like to avoid nested loops and I need the values as Rcpp::NumericMatrix objects for the functions (so using RcppArmadillo (via non-contiguous views) is only a path I would choose if no other simpler possibilities are available). Any suggestions are welcome. 
> 
> Writing loops is what you do in c++. There is no guilty conscience like in R. 
> 
>> @Rcpp::Devels: I try to understand in the doxygen documentation how the operator '()' works. I see that there is always a VECTOR::Proxy underlying each Matrix object and I end up at the traits::r_vector_proxy in the 00_forward_proxy.h file. I assume, that each Matrix object has its type given by the proxy and acts in regard to this characterization. Where can I look for more information about this struct? 
> 
> In the source files. You'll have to jump between several files. Not easy to follow, even for us. Good luck. 
> 
>> Best
>> 
>> Simon
>> _______________________________________________
>> Rcpp-devel mailing list
>> Rcpp-devel at lists.r-forge.r-project.org
>> https://lists.r-forge.r-project.org/cgi-bin/mailman/listinfo/rcpp-devel



More information about the Rcpp-devel mailing list