On Sat, Oct 25, 2008 at 9:34 AM, Gabriel Gellner <[EMAIL PROTECTED]> wrote:
>
> On Sat, Oct 25, 2008 at 12:12:18PM -0400, Gabriel Gellner wrote:
>>
>> On Sat, Oct 25, 2008 at 10:40:33AM -0500, Jason Grout wrote:
>> >
>> > David Joyner wrote:
>> > > (1) Sage also does FFT via the GSL. I guess they didn't try that,
>> > > but I would imaginge they would be fast since they are in C.
>> >
>> >
>> > The current CDF and RDF vectors implement fft using GSL. I'm in the
>> > process of switching the RDF/CDF vector backend to numpy, and then the
>> > fft will be done by numpy. I'm curious where the FFT is done by numpy
>> > currently; I couldn't find the code. Of course, the reviewer used an
>> > extremely old version of Sage (on our timescale), so maybe things were
>> > different then.
>> >
>> > For matrices, switching to numpy made most things faster, so I'm curious
>> > what we'll find when we switch the vectors to numpy as a backend.
>> >
>> Also Sage could use fftw unlike numpy (as it is gpl) which usually beats the
>> pants of fftpack (which is what numpy uses) See:
>> http://www.fftw.org/
>>
>> According to the fftw benchmarks for speed: gsl < fftpack < fftw see:
>> http://www.fftw.org/speed/CoreDuo-3.0GHz-icc/
>>
> I realize for people not used to numpy/scipy might not realize that fftw can
> be used in scipy (if it is installed, which it isn't by default due to
> licensing). What I meant is for sage this should be turned on by default, also
> fftw is what matlab uses, so as it has been said for SVD using lapack, this
> would also remove all real differences, outside of how the package is
> built/wrapped.
>
> Gabriel
>
When I was trying to do FFT's quickly this summer for a project
in Sage, I recall that when I installed the fftw library (easy), then
built scipy, then the scipy FFT used fftw and was quite fast.
I also noticed, somewhat surprisingly, that even without fftw
the scipy fft was faster than numpy's! This is why the code
in the finance.TimeSeries in Sage for doing (real) fft's is:
w = y.numpy(copy=False)
scipy.fftpack.rfft(w, overwrite_x=1)
return y
-- William
--~--~---------~--~----~------------~-------~--~----~
To post to this group, send email to sage-devel@googlegroups.com
To unsubscribe from this group, send email to [EMAIL PROTECTED]
For more options, visit this group at http://groups.google.com/group/sage-devel
URLs: http://www.sagemath.org
-~----------~----~----~----~------~----~------~--~---
