How to extract the Q & R matrices for the QR decomposition in Eigen?

marvisiyer wrote:How to extract the Q & R matrices for the QR decomposition in Eigen?

Which QR decomposition are you using? For instance, for ColPivHouseholderQR, looking in the documentation at http://eigen.tuxfamily.org/dox-devel/cl ... derQR.html I can see a method .householderQ() for extracting the Q matrix and .matrixR() for extracting the R matrix.

jitseniesen wrote:

marvisiyer wrote:How to extract the Q & R matrices for the QR decomposition in Eigen?

Which QR decomposition are you using? For instance, for ColPivHouseholderQR, looking in the documentation at http://eigen.tuxfamily.org/dox-devel/cl ... derQR.html I can see a method .householderQ() for extracting the Q matrix and .matrixR() for extracting the R matrix.

Thanks. I was using looking at FullPivHouseholderQR(http://eigen.tuxfamily.org/dox-devel/cl ... derQR.html) and HouseholderQR(http://eigen.tuxfamily.org/dox-devel/cl ... derQR.html) and these didn't have the method matrixR(). Thanks again.

but they have a matrixQR() from which you can get the R matrix as the upper triangular part:

Code: Select all

qr.matrixQR().triangularView<Upper>()


ggael wrote:but they have a matrixQR() from which you can get the R matrix as the upper triangular part:

Code: Select all

qr.matrixQR().triangularView<Upper>()


Thanks. But looks like all the three methods return a full QR. My matrix is a very thin matrix and I definitely do not want to do full QR. How should I proceed to get a reduced QR factorization?

they do a reduced QR by default.

There is a thread about the thin QR at viewtopic.php?f=74&t=91271 . The code at the end of that thread may be useful for you.

ggael wrote:they do a reduced QR by default.

The question is precisely how to do that? Given a thin matrix A, how to get its thin QR and store the Q and R factors?

marvisiyer wrote:

ggael wrote:but they have a matrixQR() from which you can get the R matrix as the upper triangular part:

Code: Select all

qr.matrixQR().triangularView<Upper>()


Thanks. But looks like all the three methods return a full QR. My matrix is a very thin matrix and I definitely do not want to do full QR. How should I proceed to get a reduced QR factorization?

I am getting an error Upper was not declared in this scope. Any hints, sorry for being noob.


I have included hearder files Eigen/Dense , Eigen/QR

Edited:

Found the bug, forgot to write in using namespace Eigen;