1

I need to do a matrix * vector multiplication, which generates only a single number, and pass it to a vector (but it doesn't work in the commented code below). What I have so far is:

cpp code

#include <TMB.hpp>
template<class Type>
Type objective_function<Type>::operator() ()
{
  DATA_MATRIX(U); // id x 2 matrix
  DATA_MATRIX(Z); // n x 2 matrix
  matrix<Type> Z1_m1 = matrix<Type>(Z.row(0))*vector<Type>(U.row(0));
  REPORT(Z1_m1); // works
  vector<Type> ZZ(6);
  // ZZ(0) = matrix<Type>(Z.row(0))*vector<Type>(U.row(0)); // HOW TO FIX IT??
  // (This is a small code, which the real application run inside a for loop, and the row indices will be given by for)
  REPORT(ZZ);
  return 0;
}

R code

require(TMB)
set.seed(232)
model_data = list(U = matrix(c(9,11,2,4), ncol = 2),
                  Z = matrix(c(1,2,3,4,5,6, rpois(6,2)), ncol=2))
model <- "mult"
compile(paste0(model, ".cpp"))
dyn.load(dynlib(model))
m1 = MakeADFun(data=model_data, parameters=list(),type="Fun",
                  checkParameterOrder=FALSE,DLL=model)
print(m1$report())  # Note: order of variables NOT the same as .cpp file

Any help?

I have already posted this question into TMB users group. If the solution appears first there, I will post it here.

1 Answers1

0

I will post here my sollution after @Bob help in TMB Users group:

#include <TMB.hpp>
template<class Type>
Type objective_function<Type>::operator() ()
{
  DATA_MATRIX(U); // id x 2 matrix
  DATA_MATRIX(Z); // n x 2 matrix
  vector<Type> ZZ(6);
  // ZZ(0) = (Z.row(0).array()*U.row(0).array()).sum();
  REPORT(ZZ);
  return 0;
}

This is inefficient because it is an array element-wise multiplication. If it was a matrix multiplication would run faster.