Matlab's Inverse
in [Matlab]
|
Prev: Matlab GUI help
Next: file names
From: Deb Gruner on 26 Jul 2010 23:54 thank you everyone for trying to help me to answer the last person's question i'm converting the code to c++ and so far its alot easier for me to do using gauss jordan elimination as oposed to lu decomposition although neither are imposibble i do know that the time difference for computation is from n^2 to n^3 difference between the two but im fine with that if the numbers still show the same and then i started wondering how close either are to the numbers produced by matlab since i dont even know which method matlab is using to implement its inv function
From: Walter Roberson on 27 Jul 2010 00:18 Deb Gruner wrote: > thank you everyone for trying to help me to answer the last person's > question i'm converting the code to c++ > and so far its alot easier for me to do using gauss jordan elimination > as oposed to lu decomposition although neither are imposibble i do know > that the time difference for computation is from n^2 to n^3 difference > between the two but im fine with that if the numbers still show the same > and then i started wondering how close either are to the numbers > produced by matlab since i dont even know which method matlab is using > to implement its inv function As I quoted before from the inv reference page, it uses the LAPACK routines DLANGE, DGETRF, DGECON, DGETRI A moment's googling finds: http://www.math.utah.edu/software/lapack/lapack-d/dgetri.html NAME DGETRI - compute the inverse of a matrix using the LU fac- torization computed by DGETRF If you are converting the code to C++ then you can make calls to LAPACK yourself -- it is a standard library readily available.
From: Steve Amphlett on 27 Jul 2010 01:21 Walter Roberson <roberson(a)hushmail.com> wrote in message <n6t3o.30544$xZ2.5377(a)newsfe07.iad>... > Deb Gruner wrote: > > thank you everyone for trying to help me to answer the last person's > > question i'm converting the code to c++ > > and so far its alot easier for me to do using gauss jordan elimination > > as oposed to lu decomposition although neither are imposibble i do know > > that the time difference for computation is from n^2 to n^3 difference > > between the two but im fine with that if the numbers still show the same > > and then i started wondering how close either are to the numbers > > produced by matlab since i dont even know which method matlab is using > > to implement its inv function > > As I quoted before from the inv reference page, it uses the LAPACK routines > > DLANGE, DGETRF, DGECON, DGETRI > > A moment's googling finds: > > http://www.math.utah.edu/software/lapack/lapack-d/dgetri.html > NAME > DGETRI - compute the inverse of a matrix using the LU fac- > torization computed by DGETRF > > > If you are converting the code to C++ then you can make calls to LAPACK > yourself -- it is a standard library readily available. I did precisely this myself (Matlab -> C++) quite recently, using ZGETRF and ZGETRI (my Matrices are complex). You'll need to copy your data out of your C++ Matrix class into regular FORTRAN-style arrays before calling the LAPACK functions and then back into your C++ class(es) after. But the speedup is well worthwhile. My Matrix inversion went from a crippling bottle-neck to mere noise once I'd done this.
From: John D'Errico on 27 Jul 2010 07:04 "Deb Gruner " <gsarah10(a)aol.com> wrote in message <i2ll8t$6l7$1(a)fred.mathworks.com>... > thank you everyone for trying to help me > to answer the last person's question > i'm converting the code to c++ > and so far its alot easier for me to do using gauss jordan elimination as oposed to lu decomposition although neither are imposibble > i do know that the time difference for computation is from n^2 to n^3 difference between the two but im fine with that if the numbers still show the same > No, you DON'T Know that one of these is an n^2 operation, because that simply is not true. John
From: Penny Anderson on 27 Jul 2010 09:09
"Steve Amphlett" <Firstname.Lastname(a)Where-I-Work.com> wrote in message <i2lqc1$6kf$1(a)fred.mathworks.com>... > Walter Roberson <roberson(a)hushmail.com> wrote in message <n6t3o.30544$xZ2.5377(a)newsfe07.iad>... > > Deb Gruner wrote: > > > thank you everyone for trying to help me to answer the last person's > > > question i'm converting the code to c++ > > > and so far its alot easier for me to do using gauss jordan elimination > > > as oposed to lu decomposition although neither are imposibble i do know > > > that the time difference for computation is from n^2 to n^3 difference > > > between the two but im fine with that if the numbers still show the same > > > and then i started wondering how close either are to the numbers > > > produced by matlab since i dont even know which method matlab is using > > > to implement its inv function > > > > As I quoted before from the inv reference page, it uses the LAPACK routines > > > > DLANGE, DGETRF, DGECON, DGETRI > > > > A moment's googling finds: > > > > http://www.math.utah.edu/software/lapack/lapack-d/dgetri.html > > NAME > > DGETRI - compute the inverse of a matrix using the LU fac- > > torization computed by DGETRF > > > > > > If you are converting the code to C++ then you can make calls to LAPACK > > yourself -- it is a standard library readily available. > > I did precisely this myself (Matlab -> C++) quite recently, using ZGETRF and ZGETRI (my Matrices are complex). You'll need to copy your data out of your C++ Matrix class into regular FORTRAN-style arrays before calling the LAPACK functions and then back into your C++ class(es) after. But the speedup is well worthwhile. My Matrix inversion went from a crippling bottle-neck to mere noise once I'd done this. Really? Translating your MATLAB code into C++ (calls into LAPACK) made the inv function go much faster? I find this hard to believe. In general those O(n^3) linear algebra routines perform pretty well in MATLAB. It's usually the other parts of the code around it that may be improved by a translation to C/C++. Penny Anderson MATLAB Math MathWorks |