[Numpy-discussion] triangular matrix fill

Charles R Harris charlesr.harris@gmail....
Mon May 26 13:42:00 CDT 2008


On Mon, May 26, 2008 at 11:06 AM, Tom Waite <twaite@berkeley.edu> wrote:

> Thanks Robert and Chuck.
>
> The matrix calculation is a bottleneck computation and that is my interest
> in only doing the lower triangular part of the outer product. This is part
> of a nonlinear brainwarp for the NIPY project. For a typical anatomical MRI
> volume size this is a 768 x 768 matrix that is inside three nested loops
> that walk the volume. This matrix is actually the curvature matrix
> computation and is "inspired" by the Marquardt routine (mrqcof) in Numerical
> Recipes. Having this lower triangular outer product in numpy could be of
> value for scipy.optimize down the road if Levenberg-Marquardt is added. Is
> this something I should post to be added to numpy or write my own extension
> code as this might be too specialized?
>
> Tom
>

I think your own extension code is the place to start. If something like
this goes into scipy, then it should probably be part of a larger package
containing other relevant optimizations.

Chuck
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://projects.scipy.org/pipermail/numpy-discussion/attachments/20080526/200bab98/attachment.html 


More information about the Numpy-discussion mailing list