[SciPy-User] Proposal for a new data analysis toolbox
Wed Nov 24 11:50:04 CST 2010
On 11/24/10 9:30 AM, Dag Sverre Seljebotn wrote:
> For the time being, for something like this I'd definitely go with a
> template language to generate Cython code if you are not already.
I've thought for years that one way to really help numpy performance is
an easier way to write C extensions for numpy arrays. If it's easy
enough, folks will write what may well be special-case code, but if the
common cases are covered, we'll get some pretty good performance benefits.
For example, years ago I wrote a "fast_clip" function for Numeric, when
I had some code that was calling clip() a lot. However, it was
hand-written C, and way too much of a pain to write and maintain
(evidenced by the fact that it isn't maintained now...)
Cython gets us a long way, but so far only for the true special cases --
one data type, one dimensionality. That appears to be the kind of thing
this thread was started with.
A good templating system would be a great way to make this all possible.
The <as_yet_unamed_data_analysis_toolbox> could be a great way to
prototype and test such a system.
Christopher Barker, Ph.D.
Emergency Response Division
NOAA/NOS/OR&R (206) 526-6959 voice
7600 Sand Point Way NE (206) 526-6329 fax
Seattle, WA 98115 (206) 526-6317 main reception
More information about the SciPy-User