Regina Henschel wrote:
There is no mathematical problem, but I'm uncertain about coding style. The algorithms work on matrices. They have a lot of parts which are nearly identical but the matrices are transposed. How to handle that?
Hi Regina, I'd suggest using templates to achieve that - this way, you'll get the same runtime performance, but avoid all the (semantically) duplicated code. There are multiple ways to tackle that, to avoid changing your code too much, I'd suggest passing a permutating type as template parameter to each of the methods in need of variation, e.g.: template< class Selector > bool calculateQRdecomposition( ScMatrixRef pMatA, ::std::vector< double>& pVecR, SCSIZE nK, SCSIZE nN, Selector const& rSelector ) { for (SCSIZE major = 0; major < rSelector.getMajor(nK,nN); major+) ... for (SCSIZE minor = major; minor < rSelector.getMinor(nK,nN); minor++) pMatA->PutDouble( pMatA->GetDouble( rSelector.getCol(major,minor), rSelector.getRow(major,minor))/fScale, rSelector.getCol(major,minor), rSelector.getRow(major,minor)); etc - the Selector class then has simple inline double blah(a,b) { return a; } permutation methods - if you like, we could have a look at that during the HackFest next weekend. Otherwise - lovely code! Cheers, -- Thorsten
Attachment:
pgpeDFwYq6PPP.pgp
Description: PGP signature