如何使用Rcpp初始化mappedsparsematrix

Question

I would like to create a sparse matrix using mappedsparsematrix type via Rcpp in R. I choose mappedsparsematrix rather than sparsematrix because I want to use it in R for further computation. Correct me if I am wrong on this point.

Here is my code in cpp form

// [[Rcpp::depends(RcppEigen)]]
# include <RcppEigen.h>
# include <Rcpp.h>
# include <math.h>
# include <list>

using namespace Rcpp;
using Eigen::SparseMatrix;
using Eigen::MappedSparseMatrix;
using Eigen::MatrixXd;
using Eigen::VectorXi;
typedef Eigen::VectorXd Vd;
typedef Eigen::VectorXi Vi;
typedef Eigen::MappedSparseMatrix<double> MSpMat;
typedef MSpMat::InnerIterator InIterMat;
typedef List list;
typedef Eigen::Triplet<double> T;


double euclidean_distance(double lon1, double lat1,double lon2,double lat2){
  double s = pow((lon1 - lon2),2) + pow((lat1 - lat2),2);
    double ed = sqrt(s);
    return(ed);
}


// [[Rcpp::export]]
MSpMat mymain(NumericVector lat, NumericVector lon){
   std::list<T> L;

   int length = lat.size();
   int index = 0;
   for (int i = 0; i < length - 1; i++ ){
       for (int j = i+1; j < length; j++){
         double lon1 = lon[i];
         double lon2 = lon[j];
         double lat1 = lat[i];
         double lat2 = lat[j];
         double dist = euclidean_distance(lon1,lat1,lon2,lat2);
         dist = exp(-dist/0.001);
         if (dist > 0.01 ){
            L.push_back(T(index, i, dist));
            L.push_back(T(index, j, -dist)); 
         }

      }
   }
   int nrows = L.size()/2;
   int ncols = length;

   MSpMat D(nrows, ncols);
   D.setFromTriplets(L.begin(), L.end());
   return(D);
}

However, it returns this error when I try to source the cpp file in R.

no matching constructor for initialization of "MSpMat"

Can anyone help to fix it?

Answer 1

Okay, the full error is:

temp_eigen.cpp:51:10: error: no matching constructor for initialization of 'MSpMat' (aka 'MappedSparseMatrix') MSpMat D(nrows, ncols); ^ ~~~~~~~~~~~~

note: candidate constructor (the implicit copy constructor) not viable: requires 1 argument, but 2 were provided class MappedSparseMatrix ^

note: candidate constructor not viable: requires 6 arguments, but 2 were provided inline MappedSparseMatrix(Index rows, Index cols, Index nnz, Index* outerIndexPtr, Index* innerIndexPtr, Scalar* valuePtr)

Fundamentally, the error here is trying to use a data type that requires a pre-existing memory location and defining it as if it does not (eg a two step load).

Changing:

MSpMat D(nrows, ncols);

to:

typedef Eigen::SparseMatrix< double > SpMat;
SpMat D(nrows, ncols);

yields the desired results.

The alternative is to provide a form of:

MSpMat(Index rows, Index cols, Index nnz, 
       Index* outerIndexPtr, Index* innerIndexPtr,
       Scalar* valuePtr)