Nick and Jason, Thanks for the replies.
> function. In this case, it seems like the best way to do things is to > modify both the function in matrix2.pyx (for general matrices) and the > function in matrix_rational_dense.pyx (for QQ matrices). I was afraid that would be the answer. left_kernel() already has a conditional on the field (is_NumberField( )) to decide if to call Pari, so I've made changes there as well. Fortunately, Pari spits out the basis I want (the one you get from considering the non-pivot columns). IIRC it is an IML call in the matrix_rational_dense class so I'll need to investigate that. I'll poke around some more and survey how many locations need changes to get consistent behavior. On Feb 2, 8:22 pm, Jason Grout <jason-s...@creativetrax.com> wrote: > Rob Beezer wrote: > > I am trying to add some functionality to the matrix kernel routines > > (and I'm learning more about contributing to SAGE along the way). > > Briefly, I want to make some alternative bases possible as output. In > > matrix/matrix2.pyx, there are three methods defined for a matrix: > > kernel(), left_kernel() and right_kernel(). > > > The left_kernel() function contains most of the code. > > The kernel() function is defined by the single line: kernel = > > left_kernel > > The right_kernel() function essentially forms the transpose and then > > calls kernel() on that. > > > I have a new keyword added to the left_kernel() function and new code > > working they way I think it should. However, if I try the following > > (not showing the use of the new keyword) > > > sage: a=matrix(QQ, [[1,2],[3,4]]) > > sage: a.right_kernel() > > > it would appear that that the call to kernel() in right_kernel() uses > > Since the matrix is of type > > sage: a=matrix(QQ, [[1,2],[3,4]]) > sage: type(a) > <type 'sage.matrix.matrix_rational_dense.Matrix_rational_dense'> > > it first tries to call the right_kernel function in > matrix_rational_dense.pyx. That function does not exist, so it falls > back to the right_kernel function in matrix2.pyx. That right_kernel > function in turn calls the kernel function on the (transpose of) the > matrix. First, python looks for the kernel function on the more > specific class (matrix_rational_dense.Matrix_rational_dense). Since > that class has a kernel function, it is the one that is called. > > > the version of kernel() defined for dense rational matrices (as > > perhaps it should), and not the modified version of left_kernel() > > nearby (as the code locally might suggest one to expect). If I edit > > right_kernel() to call left_kernel() on the transpose() then I see the > > effect of my modifications to the left_kernel() code. > > This is because the Matrix_rational_dense class does not have a > "left_kernel" function, so python falls back to the function in matrix2.pyx. > > The key ideas here involve inheritance in object-oriented programming. > > > > > Is it "safe" to modify right_kernel() to call left_kernel()? As I > > chase my way around, I see calls that compute kernels using algorithms > > in Pari, IML, etc, so I'm hesitant to make changes with far-reaching > > effects. Thanks in advance for the advice. > > Probably not. It seems like the purpose for the specialized kernel > function for QQ matrices is that it works much better than the generic > function. In this case, it seems like the best way to do things is to > modify both the function in matrix2.pyx (for general matrices) and the > function in matrix_rational_dense.pyx (for QQ matrices). > > Thanks, > > Jason --~--~---------~--~----~------------~-------~--~----~ To post to this group, send email to sage-devel@googlegroups.com To unsubscribe from this group, send email to sage-devel-unsubscr...@googlegroups.com For more options, visit this group at http://groups.google.com/group/sage-devel URLs: http://www.sagemath.org -~----------~----~----~----~------~----~------~--~---
