Assignment inside two for loops - rcpp

Question

I am trying to build a specific matrix but using simply R can take a lot of time considering the size of the entries that I have to use. I write a function in Rcpp with the Armadillo functionality because I need the linear algebra part to work with matrices. My code is the next:
library('Rcpp')
library('inline')
library('RcppArmadillo')
cppFunction("arma::mat GramMat(arma::mat A, double parametro, int n) {
arma::mat resultado=A;
double temp;
for (int i=0; i<n; i++){
for (int j=i; j<n; j++){
resultado(j,i)= exp(-1*parametro*((A.col(i)-A.col(j)).t() * (A.col(i)-A.col(j))));
}
}
for (int i=0; i<n; i++){
for (int j=0; j<i; j++){
resultado(i,j)=resultado(j,i);
}
}
return resultado;}",depends="RcppArmadillo")
and I am getting the next error:
temp= exp(-1*parametro*((A.col(i)-A.col(j)).t() * (A.col(i)-A.col(j))));
^
make: *** [file548914af6578.o] Error 1
The problem is with the assignation, because I tried assigning just a 1 and the assignation is working well. And I tought that maybe the problem was with the right hand side but I print it with Rcout and is delivering well number.

Answer 1

When I tried compiling your code, I saw a more informative error message:
file2f78133e7bc2.cpp: In function ‘arma::mat GramMat(arma::mat,
double, int)’: file2f78133e7bc2.cpp:14:99: error: cannot convert
‘arma::enable_if2,
arma::subview_col, arma::eglue_minus>, arma::op_htrans>,
arma::eGlue, arma::subview_col,
arma::eglue_minus>, arma::glue_times>, arma::eop_scalar_times>,
arma::eop_exp> >::result {aka const
arma::eOp,
arma::subview_col, arma::eglue_minus>, arma::op_htrans>,
arma::eGlue, arma::subview_col,
arma::eglue_minus>, arma::glue_times>, arma::eop_scalar_times>,
arma::eop_exp>}’ to ‘double’ in assignment
resultado(j,i)= exp(-1*parametro*((A.col(i)-A.col(j)).t() * (A.col(i)-A.col(j))));
^ make: *** [file2f78133e7bc2.o] Error 1
This leads us directly to the problem; the operation
(A.col(i)-A.col(j)).t() * (A.col(i)-A.col(j))
returns a type that cannot be directly converted to a double. However, we can just use arma::as_scalar() to fix this (see here in the Armadillo documentation); the following compiled fine for me:
cppFunction("arma::mat GramMat(arma::mat A, double parametro, int n) {
arma::mat resultado=A;
double temp;
for (int i=0; i<n; i++){
for (int j=i; j<n; j++){
resultado(j,i)= arma::as_scalar(exp(-1*parametro*((A.col(i)-A.col(j)).t() * (A.col(i)-A.col(j)))));
}
}
for (int i=0; i<n; i++){
for (int j=0; j<i; j++){
resultado(i,j)=resultado(j,i);
}
}
return resultado;}",depends="RcppArmadillo")
There are quite a few other things that could be improved in this code, of course. For example, as Dirk Eddelbuettel points out, you actually never use temp in your code. You might also want to use arma::dot() to get the dot product of (A.col(i)-A.col(j)) with itself (see here in the Armadillo documentation -- as arma::dot() returns a double, it would also eliminate the need to use arma::as_scalar()), etc.