[Numpy-discussion] Efficient operator overloading

Eike Welk eike.welk@gmx....
Wed Apr 18 15:56:55 CDT 2007


In the SciPy sandbox there is a module that goes into this direction. 
It was mentioned in the other optimization thread. numexpr:
http://projects.scipy.org/scipy/scipy/browser/trunk/Lib/sandbox/numexpr

This module needs to be combined with a derived array class where the 
operators return unevaluated expressions (lazy evaluation). All 
access methods and asarray must call the evaluation function. Like 
you proposed.
The unevaluated expressions and a compiler for them are in the numexpr 
module.
A very basic version of this would probably be not much work. 

I would like to replace:
import numpy as N
by:
import lazynumpy as N
:-)

Regards Eike.



More information about the Numpy-discussion mailing list