[SciPy-Dev] GSoC -- numpy interactions with sparse matrices

Blake Griffith blake.a.griffith@gmail....
Mon Jul 1 22:40:09 CDT 2013


Hello scipy,

Today marks the first day of the next phase of my GSoC proposal, numpy
interactions and overriding ufuncs for sparse matrices.

I've been thinking about how I will handle the ufunc interactions. I'm
still considering how to implement a ufunc override mechanism. However I
think the existing functionality in numpy, (__array_prepare__,
__array_finalize__, etc) would be enough for a few cases:

* unary element-wise ufuncs. Because these could be applied to the .data
attribute using array prepare.

* binary element-wise ufuncs that correspond to python operators (__add__,
__gt__, __mul__, etc.) because there is already functionality in place to
override these.

This of course leave out:
logaddexp
logaddexp2
ones_like
arctan2
hypot
maximum
minimum
copysign
nextafter
ldexp

(However I think it would be possible to override these using the existing
functionality for overriding python overrides but this would be a bit
hackish, and numpy might not like the changes.)

None of these left over ufuncs seem to be the most widely used (I could be
wrong). So going about this without adding an ufunc override attribute
would still add a lot of functionality.

So I think the best course would be to explore adding a override attribute
but if it is extremely difficult I can resort to this.

comment, suggestions, criticisms?
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mail.scipy.org/pipermail/scipy-dev/attachments/20130701/fd765967/attachment.html 


More information about the SciPy-Dev mailing list