Randomly permute rows/columns of a matrix with eigen

Question

I\'m using Eigen and I\'ve got a matrix:

MatrixXi x = MatrixXi::Random(5);

I\'d like to randomly permute the rows and columns using a rando

Answer 1

Using std::random_shuffle is perfectly fine, then you have to use a PermutationMatrix:

PermutationMatrix<Dynamic,Dynamic> perm(size);
perm.setIdentity();
std::random_shuffle(perm.indices().data(), perm.indices().data()+perm.indices().size());
A_perm = A * perm; // permute columns
A_perm = perm * A; // permute rows

Answer 2

As stated here: Stackoverflow:

If you can use C++11 I'd recommend implementing this without using srand() and random_shuffle(); instead you should use the <random> library with std::shuffle.

First, if possible rand should be avoided. Aside from the fact that it isn't usually a very good pRNG, it also has problems with thread safety due to shared state. The <random> library fixes both these problems by giving the programmer explicit control over pRNG state and by providing several options with guaranteed performance, size, and quality characteristics.

Secondly, random_shuffle isn't actually specified to use rand so it's theoretically legal for reseeding using srand not to have the effect you want. To get guaranteed results with random_shuffle you have to write your own generator. Moving to shuffle fixes that, as you can directly use standard engines.

#include <random>       //seed generation
#include <algorithm>    //shuffle()

int main() {

std::random_device r;
std::seed_seq rng_seed{r(), r(), r(), r(), r(), r(), r(), r()};

//create random engines with the rng seed
std::mt19937 eng1(rng_seed);

//create permutation Matrix with the size of the columns
Eigen::PermutationMatrix<Eigen::Dynamic, Eigen::Dynamic> permX(inputX.cols());
permX.setIdentity();
std::shuffle(permX.indices().data(), permX.indices().data()+permX.indices().size(), eng1);
inputX = inputX * permX;   //shuffle column wise

}

If you want to shuffle the rows use inputX.rows() instead for the initialization of the Permutation Matrix. And use inputX = permX * inputX instead.