On Thu, Jul 10, 2008 at 9:33 AM, Anne Archibald <[EMAIL PROTECTED]> wrote:
> 2008/7/9 Robert Kern <[EMAIL PROTECTED]>: > > > Because that's just what a buffer= argument *is*. It is not a place > > for presenting the starting pointer to exotically-strided memory. Use > > __array_interface__s to describe the full range of representable > > memory. See below. > > Aha! Is this stuff documented somewhere? > > > I was about a week ahead of you. See numpy/lib/stride_tricks.py in the > trunk. > > Nice! Unfortunately it can't quite do what I want... for the linear > algebra I need something that can broadcast all but certain axes. For > example, take an array of matrices and an array of vectors. The > "array" axes need broadcasting, but you can't broadcast on all axes > without (incorrectly) turning the vector into a matrix. I've written a > (messy) implementation, but the corner cases are giving me headaches. > I'll let you know when I have something that works. > I think something like a matrix/vector dtype would be another way to go for stacks of matrices and vectors. It would have to be a user defined type to fit into the current type hierarchy for ufuncs, but I think the base machinery is there with the generic inner loops. Chuck
_______________________________________________ Numpy-discussion mailing list Numpy-discussion@scipy.org http://projects.scipy.org/mailman/listinfo/numpy-discussion
