Eigen学习记录

Eigen库由Ubuntu软件源中提供，通过apt命令可以很方便的安装Eigen。

sudo apt-get install libeigen3-dev

Eigen与其他库不同，它是一个由头文件搭建起来的库，Eigen头文件的默认安装位置在“/usr/include/eigen3/”中。我们在使用时，只需引入Eigen头文件，不需要链接它的库文件，在CMakeLists.txt里添加Eigen头文件的目录。

#添加头文件

include_directories("/usr/include/eigen3")

以下参考eigen官网

eigen.cpp

#include <iostream>
#include <Eigen/Dense>
using Eigen::MatrixXd;//using namespace Eigen
int main()
{
  MatrixXd m(2,2);
  m(0,0) = 3;
  m(1,0) = 2.5;
  m(0,1) = -1;
  m(1,1) = m(1,0) + m(0,1);
  std::cout << m << std::endl;
}

CMakeLists.txt

projiect(eigen) inlcude_directories("/usr/include/eigen3") add_executable(eigen eigen.cpp)

When you run the program, it produces the following output:

  3  -1
2.5 1.5

The Matrix class

The three mandatory template parameters of Matrix are:

Matrix<typename Scalar, int RowsAtCompileTime, int ColsAtCompileTime>

//Matrix4f is a 4x4 matrix of floats.
typedef Matrix<float, 4, 4> Matrix4f;

Vectors

typedef Matrix<folat,3,1>Vector3f;//列向量，列为1.
typedef Matrix<int,1,2>RowVector2i;//行向量，行为1,2列。

和MATLAB对比：http://eigen.tuxfamily.org/dox/AsciiQuickReference.txt

参考博客：https://www.cnblogs.com/python27/p/EigenQuickRef.html

Eigen 矩阵定义

#include <Eigen/Dense>

Matrix<double, 3, 3> A;               // Fixed rows and cols. Same as Matrix3d.
Matrix<double, 3, Dynamic> B;         // Fixed rows, dynamic cols.
Matrix<double, Dynamic, Dynamic> C;   // Full dynamic. Same as MatrixXd.
Matrix<double, 3, 3, RowMajor> E;     // Row major; default is column-major.
Matrix3f P, Q, R;                     // 3x3 float matrix.
Vector3f x, y, z;                     // 3x1 float matrix.
RowVector3f a, b, c;                  // 1x3 float matrix.
VectorXd v;                           // Dynamic column vector of doubles
// Eigen // Matlab // comments
x.size()          // length(x) // vector size
C.rows()          // size(C,1) // number of rows
C.cols()          // size(C,2) // number of columns
x(i)              // x(i+1) // Matlab is 1-based
C(i,j)            // C(i+1,j+1) //

Eigen 基础使用

// Basic usage
// Eigen // Matlab // comments
x.size()        // length(x) // vector size
C.rows()        // size(C,1) // number of rows
C.cols()        // size(C,2) // number of columns
x(i)            // x(i+1) // Matlab is 1-based
C(i, j)         // C(i+1,j+1) //

A.resize(4, 4);   // Runtime error if assertions are on.
B.resize(4, 9);   // Runtime error if assertions are on.
A.resize(3, 3);   // Ok; size didn't change.
B.resize(3, 9);   // Ok; only dynamic cols changed.

A << 1, 2, 3,     // Initialize A. The elements can also be
     4, 5, 6,     // matrices, which are stacked along cols
     7, 8, 9;     // and then the rows are stacked.
B << A, A, A;     // B is three horizontally stacked A's.
A.fill(10);       // Fill A with all 10's.

Eigen 特殊矩阵生成

// Eigen // Matlab
MatrixXd::Identity(rows,cols)       // eye(rows,cols)
C.setIdentity(rows,cols)            // C = eye(rows,cols)
MatrixXd::Zero(rows,cols)           // zeros(rows,cols)
C.setZero(rows,cols)                // C = ones(rows,cols)
MatrixXd::Ones(rows,cols)           // ones(rows,cols)
C.setOnes(rows,cols)                // C = ones(rows,cols)
MatrixXd::Random(rows,cols)         // rand(rows,cols)*2-1 // MatrixXd::Random returns uniform random numbers in (-1, 1).
C.setRandom(rows,cols)              // C = rand(rows,cols)*2-1
VectorXd::LinSpaced(size,low,high)  // linspace(low,high,size)'
v.setLinSpaced(size,low,high)       // v = linspace(low,high,size)'

Eigen 矩阵分块

// Matrix slicing and blocks. All expressions listed here are read/write.
// Templated size versions are faster. Note that Matlab is 1-based (a size N
// vector is x(1)...x(N)).
// Eigen // Matlab
x.head(n)                          // x(1:n)
x.head<n>()                        // x(1:n)
x.tail(n)                          // x(end - n + 1: end)
x.tail<n>()                        // x(end - n + 1: end)
x.segment(i, n)                    // x(i+1 : i+n)
x.segment<n>(i)                    // x(i+1 : i+n)
P.block(i, j, rows, cols)          // P(i+1 : i+rows, j+1 : j+cols)
P.block<rows, cols>(i, j)          // P(i+1 : i+rows, j+1 : j+cols)
P.row(i)                           // P(i+1, :)
P.col(j)                           // P(:, j+1)
P.leftCols<cols>()                 // P(:, 1:cols)
P.leftCols(cols)                   // P(:, 1:cols)
P.middleCols<cols>(j)              // P(:, j+1:j+cols)
P.middleCols(j, cols)              // P(:, j+1:j+cols)
P.rightCols<cols>()                // P(:, end-cols+1:end)
P.rightCols(cols)                  // P(:, end-cols+1:end)
P.topRows<rows>()                  // P(1:rows, :)
P.topRows(rows)                    // P(1:rows, :)
P.middleRows<rows>(i)              // P(i+1:i+rows, :)
P.middleRows(i, rows)              // P(i+1:i+rows, :)
P.bottomRows<rows>()               // P(end-rows+1:end, :)
P.bottomRows(rows)                 // P(end-rows+1:end, :)
P.topLeftCorner(rows, cols)        // P(1:rows, 1:cols)
P.topRightCorner(rows, cols)       // P(1:rows, end-cols+1:end)
P.bottomLeftCorner(rows, cols)     // P(end-rows+1:end, 1:cols)
P.bottomRightCorner(rows, cols)    // P(end-rows+1:end, end-cols+1:end)
P.topLeftCorner<rows,cols>()       // P(1:rows, 1:cols)
P.topRightCorner<rows,cols>()      // P(1:rows, end-cols+1:end)
P.bottomLeftCorner<rows,cols>()    // P(end-rows+1:end, 1:cols)
P.bottomRightCorner<rows,cols>()   // P(end-rows+1:end, end-cols+1:end)

Eigen 矩阵元素交换

// Of particular note is Eigen's swap function which is highly optimized.
// Eigen                           // Matlab
R.row(i) = P.col(j);               // R(i, :) = P(:, i)
R.col(j1).swap(mat1.col(j2));      // R(:, [j1 j2]) = R(:, [j2, j1])

Eigen 矩阵转置

// Views, transpose, etc; all read-write except for .adjoint(). // Eigen // Matlab
R.adjoint()                        // R'
R.transpose()                      // R.' or conj(R')
R.diagonal()                       // diag(R)
x.asDiagonal()                     // diag(x)
R.transpose().colwise().reverse(); // rot90(R)
R.conjugate()                      // conj(R)

Eigen 矩阵乘积

// All the same as Matlab, but matlab doesn't have *= style operators.
// Matrix-vector.  Matrix-matrix.   Matrix-scalar.
y  = M*x;          R  = P*Q;        R  = P*s;
a  = b*M;          R  = P - Q;      R  = s*P;
a *= M;            R  = P + Q;      R  = P/s;
                   R *= Q;          R  = s*P;
                   R += Q;          R *= s;
                   R -= Q;          R /= s;

Eigen 矩阵单个元素操作

// Vectorized operations on each element independently
// Eigen // Matlab
R = P.cwiseProduct(Q);    // R = P .* Q
R = P.array() * s.array();// R = P .* s
R = P.cwiseQuotient(Q);   // R = P ./ Q
R = P.array() / Q.array();// R = P ./ Q
R = P.array() + s.array();// R = P + s
R = P.array() - s.array();// R = P - s
R.array() += s;           // R = R + s
R.array() -= s;           // R = R - s
R.array() < Q.array();    // R < Q
R.array() <= Q.array();   // R <= Q
R.cwiseInverse();         // 1 ./ P
R.array().inverse();      // 1 ./ P
R.array().sin()           // sin(P)
R.array().cos()           // cos(P)
R.array().pow(s)          // P .^ s
R.array().square()        // P .^ 2
R.array().cube()          // P .^ 3
R.cwiseSqrt()             // sqrt(P)
R.array().sqrt()          // sqrt(P)
R.array().exp()           // exp(P)
R.array().log()           // log(P)
R.cwiseMax(P)             // max(R, P)
R.array().max(P.array())  // max(R, P)
R.cwiseMin(P)             // min(R, P)
R.array().min(P.array())  // min(R, P)
R.cwiseAbs()              // abs(P)
R.array().abs()           // abs(P)
R.cwiseAbs2()             // abs(P.^2)
R.array().abs2()          // abs(P.^2)
(R.array() < s).select(P,Q);  // (R < s ? P : Q)

Eigen 矩阵化简

// Reductions.
int r, c;
// Eigen                  // Matlab
R.minCoeff()              // min(R(:))
R.maxCoeff()              // max(R(:))
s = R.minCoeff(&r, &c)    // [s, i] = min(R(:)); [r, c] = ind2sub(size(R), i);
s = R.maxCoeff(&r, &c)    // [s, i] = max(R(:)); [r, c] = ind2sub(size(R), i);
R.sum()                   // sum(R(:))
R.colwise().sum()         // sum(R)
R.rowwise().sum()         // sum(R, 2) or sum(R')'
R.prod()                  // prod(R(:))
R.colwise().prod()        // prod(R)
R.rowwise().prod()        // prod(R, 2) or prod(R')'
R.trace()                 // trace(R)
R.all()                   // all(R(:))
R.colwise().all()         // all(R)
R.rowwise().all()         // all(R, 2)
R.any()                   // any(R(:))
R.colwise().any()         // any(R)
R.rowwise().any()         // any(R, 2)

Eigen 矩阵点乘

// Dot products, norms, etc.
// Eigen // Matlab
x.norm()                  // norm(x). Note that norm(R) doesn't work in Eigen.
x.squaredNorm()           // dot(x, x) Note the equivalence is not true for complex
x.dot(y)                  // dot(x, y)
x.cross(y)                // cross(x, y) Requires #include <Eigen/Geometry>

Eigen 矩阵类型转换

//// Type conversion
// Eigen // Matlab
A.cast<double>();                  // double(A)
A.cast<float>();                   // single(A)
A.cast<int>();                     // int32(A)
A.real();                          // real(A)
A.imag();                          // imag(A)
// if the original type equals destination type, no work is done

Eigen 求解线性方程组 Ax = b

// Solve Ax = b. Result stored in x. Matlab: x = A \ b.
x = A.ldlt().solve(b)); // A sym. p.s.d. #include <Eigen/Cholesky>
x = A.llt() .solve(b)); // A sym. p.d. #include <Eigen/Cholesky>
x = A.lu()  .solve(b)); // Stable and fast. #include <Eigen/LU>
x = A.qr()  .solve(b)); // No pivoting. #include <Eigen/QR>
x = A.svd() .solve(b)); // Stable, slowest. #include <Eigen/SVD>
// .ldlt() -> .matrixL() and .matrixD()
// .llt()  -> .matrixL()
// .lu()   -> .matrixL() and .matrixU()
// .qr()   -> .matrixQ() and .matrixR()
// .svd()  -> .matrixU(), .singularValues(), and .matrixV()

Eigen 矩阵特征值

// Eigenvalue problems
// Eigen // Matlab
A.eigenvalues();                  // eig(A);
EigenSolver<Matrix3d> eig(A);     // [vec val] = eig(A)
eig.eigenvalues();                // diag(val)
eig.eigenvectors();               // vec
// For self-adjoint matrices use SelfAdjointEigenSolver<>

Cholesky分解法又叫平方根法，是求解对称正定线性方程组最常用的方法之一。对于一般矩阵，为了消除LU分解的局限性和误差的过分积累，采用了选主元的方法，但对于对称正定矩阵而言，选主元是不必要的。

参考博客：Eigen: C++开源矩阵计算工具——Eigen的简单用法

 MatrixXd::Random(3,3)表示产生一个元素类型为double的3*3的临时矩阵对象。

QR分解
Eigen的QR分解非常绕人，它总共提供了下面这些矩阵的分解方式：

由于我只用到了QR分解，而且Eigen的QR分解开始使用时确实不容易入手，因此这里只提供了householderQR的分解方式的演示代码：

void QR2()
{
    Matrix3d A;
    A<<1,1,1,
        2,-1,-1,
        2,-4,5;

    HouseholderQR<Matrix3d> qr;
    qr.compute(A);
    MatrixXd R = qr.matrixQR().triangularView<Upper>();
    MatrixXd Q =  qr.householderQ();
    std::cout << "QR2(): HouseholderQR---------------------------------------------"<< std::endl;
    std::cout << "A "<< std::endl <<A << std::endl << std::endl;
    std::cout <<"qr.matrixQR()"<< std::endl << qr.matrixQR() << std::endl << std::endl;
    std::cout << "R"<< std::endl <<R << std::endl << std::endl;
    std::cout << "Q "<< std::endl <<Q << std::endl << std::endl;
    std::cout <<"Q*R" << std::endl <<Q*R << std::endl << std::endl;
}

Example:

Matrix3f m = Matrix3f::Random();

Matrix3f y = Matrix3f::Random();

cout << "Here is the matrix m:" << endl << m << endl;

cout << "Here is the matrix y:" << endl << y << endl;

Matrix3f x;

x = m.colPivHouseholderQr().solve(y);

assert(y.isApprox(m*x));

cout << "Here is a solution x to the equation mx=y:" << endl << x << endl;

Here is the matrix m:
  0.68  0.597  -0.33
-0.211  0.823  0.536
 0.566 -0.605 -0.444
Here is the matrix y:
  0.108   -0.27   0.832
-0.0452  0.0268   0.271
  0.258   0.904   0.435
Here is a solution x to the equation mx=y:
 0.609   2.68   1.67
-0.231  -1.57 0.0713
  0.51   3.51   1.05