From: Roman G. <rom...@gm...> - 2010-01-27 22:09:33
|
Hi Dominique Thanks for the explanation. Maybe it would be feasible to store off-diagonal elements twice (above and below the diagonal) when extracting a submatrix from a symmetric matrix. Regards, Roman On 27.01.2010 00:42, Dominique Orban wrote: > On Tue, Jan 26, 2010 at 9:28 PM, Roman Geus<rom...@gm...> wrote: > >> Hello >> >> Triggered by a recent email on the pysparse-users list I tried to >> compile pysparse 1.1 for Windows and Python 2.6. >> >> I used Visual C++ (Express Edition) and the latest ACML for BLAS and >> LAPACK (both are available for free). >> >> Except for the superlu package (which I skipped) only a few changes to >> the source code were necessary (mostly due to the lack of C99 support in >> the MS C compiler) to build the modules. >> >> Unfortunately one of the unit tests in spmatrix_test failed, but >> basically it seemed to work. (Later I saw that the same unit test also >> fails on Linux, so it might be a general problem.) >> >> I have two questions: >> >> Do you know of anybody who has successfully built pysparse using Visual >> C++ recently? >> >> Do you think it would be desirable to be able to build using Visual C++ >> or is build process using MinGW straight-forward and sufficient? >> >> Best regards, >> Roman >> > I will let Windows users comment on what would best suit their needs. > In the meantime I commented out the test that fails. The reason is > that when I implemented fancy indexing (which includes slicing), I > decided that taking a slice on a matrix would result by default in a > non-symmetric matrix, even if the original matrix was symmetric. > Clearly, only symmetric slices would yield a symmetric submatrix > (although special cases may occur but would be costly to check for). > Currently, Pysparse doesn't check for symmetric slices. > > |