Eigen: Replicate (broadcast) by rows

问题

I'd like to replicate each row of a matrix M without any copy occurring (i.e. by creating a view):

0  1          0  1
2  3    ->    0  1
              2  3
              2  3

• M.rowwise().replicate(n) is a shorcut for M.replicate(1,n) which seems kind of useless.

• The following snippet does a copy, and cannot work if M is an expression.

    Eigen::Index rowFactor = 2;
    Eigen::MatrixXi M2 = Eigen::Map(M.data(), 1, M.size()).replicate(rowFactor, 1);
    M2.resize(M.rows()*rowFactor, M.cols()) ;

• In some situation, I may use the intermediate view Eigen::Map<Eigen::MatrixXi>(M.data(), 1, M.size()).replicate(rowFactor, 1) by reshaping the other operands, but that's not very satisfying.

Is there a proper way to achieve this broadcast view?

回答1:

What you want is essentially a Kronecker product with a matrix of ones. You can use the (unsupported) KroneckerProduct module for that:

#include <iostream>
#include <unsupported/Eigen/KroneckerProduct>

int main() {
    Eigen::Matrix2i M; M << 0, 1, 2, 3;
    std::cout << Eigen::kroneckerProduct(M, Eigen::Vector2i::Ones()) << '\n';
}

Being 'unsupported' means that the API of the module is not guaranteed to be stable (though this module has not changed since its introduction, I think).

来源：https://stackoverflow.com/questions/45466391/eigen-replicate-broadcast-by-rows