Can someone point if I am doing anything wrong here, or is it just that * and division operations are currently not defined row & column wise?

#include <iostream>
#include <Eigen/Core>
int main()
{
Eigen::ArrayXXf a(2,4);
a << 1, 2, 3, 4,
6, 7, 8, 9;

Eigen::ArrayXf max = a.colwise().maxCoeff();
a.rowwise()-=max;//Compiles & Works
a.rowwise()/=max;//Compilation error
return 0;
}

You can do this as a standard matrix product (the .matrix() are to move to the linear algebra world):

a.matrix() = max.matrix().asDiagonal().inverse() * a.matrix()

The main reason that operator / and * are not yet in row/colwise is because we have to make them available only in the array world.

Note that diagonal matrix product and inversion are optimized.

Hi, I tried this solution. It seems to be 7x slower than writing a for loop per row with llvm-gcc 4.2 on mac. My array has four columns of float. So every row is perfectly aligned.

make sure you bench with optimization enabled -O2 -DNDEBUG

Is there a reason why this wouldn't work in place?

Code: Select all

  Eigen::MatrixXf m(2,3);
  m(0,0) = 1;   m(0,1) = 2;   m(0,2) = 3;
  m(1,0) = 4;   m(1,1) = 5;   m(1,2) = 6;

  std::cout << "m: " << m << std::endl;

  // Want to see that the first row remains the same, while the second row is divided by 2
  Eigen::VectorXf v(2);
  v[0] = 1;
  v[1] = 2;

  // This works
  Eigen::MatrixXf m2 = v.matrix().asDiagonal().inverse() * m;
  std::cout << "m2: " << m2 << std::endl;

  // This does not seem to work properly?
  Eigen::MatrixXf m3 = m;
  std::cout << "m3: " << m3 << std::endl;
  m3 *= v.matrix().asDiagonal().inverse();
  std::cout << "m3: " << m3 << std::endl;

The output is

Code: Select all

m2:   1   2   3
  2 2.5   3
m3:   1   1 inf
  4 2.5 inf


if you compile with -DNDEBUG, you should get your answer: m3 *= d is equivalent to m3 = m3 * d, however here m3 is 2x3 and d 2x2....