Question

我需要将数据存储在类似3D的结构中，但是我一直依靠Eigen库来处理代码中的矩阵结构，而Eigen不提供3D矩阵。我找到了两种可能的解决方法：

 int x,y,z;
 Eigen::Matrix<Eigen::Matrix<double,Dynamic,Dynamic>, Dynamic,1> M(z);
 for (int i = 0; i < M.rows(); ++i) M(i)=MatrixXd::Zero(x,y);

 // access coefficients with M(z)(x,y)

和

 int x,y,z;
 std::vector<Eigen::Matrix<double,Dynamic,Dynamic> > M(z);
 for (int i = 0; i < M.rows(); ++i) M[i]=MatrixXd::Zero(x,y);

 // access coefficients with M[z](x,y)

我的问题是：使用这两种方法有什么速度/效率优势，还是等效？

Answer 1

试试这段代码：

#include<windows.h>

LARGE_INTEGER startTime, stopTime;
LARGE_INTEGER freq;

int main(int argc, char *argv[])
{
    QueryPerformanceFrequency(&freq);

    // ACCESS TIME 1
    QueryPerformanceCounter(&startTime);

    int x1,y1,z1;
    Eigen::Matrix<Eigen::Matrix<double,Dynamic,Dynamic>, Dynamic,1> M1(z1);
    for (int i = 0; i < M1.rows(); ++i) M1(i)=MatrixXd::Zero(x1,y1);

    QueryPerformanceCounter(&stopTime);
    double msecs1= (double)(stopTime.QuadPart - startTime.QuadPart) / (double)freq.QuadPart;

    // ACCESS TIME 2
    QueryPerformanceCounter(&startTime);

    int x2,y2,z2;
    std::vector<Eigen::Matrix<double,Dynamic,Dynamic> > M2(z2);
    for (int i = 0; i < M2.rows(); ++i) M2[i]=MatrixXd::Zero(x2,y2);

    QueryPerformanceCounter(&stopTime); 
    double msecs2= (double)(stopTime.QuadPart - startTime.QuadPart) / (double)freq.QuadPart;

    // RESULT
    cout<<"t1="<<msecs1<<", t2="<<msecs2;   
}

最有效的方式来保存几个Eigen MatrixXd

1 个答案: