[SciPy-Dev] fftpack test failures for 0.8.0b1

[Ralf Gommers]

On Thu, Jun 17, 2010 at 11:48 PM, Charles R Harris <
charlesr.harris at gmail.com> wrote:

>
>
> On Thu, Jun 17, 2010 at 8:50 AM, Ralf Gommers <ralf.gommers at googlemail.com
> > wrote:
>
>> Below are two test failures in fftpack. The test data is random, so they
>> happen only occasionally. I decreased the precision to decimal=5 and
>> maxulp=1e4, but still got one failure of each test in 200 test runs. The
>> TestSingleIFFT failure always occurs with an array of size 111. This is on
>> OS X 10.6, the same first error was also reported on 10.5.
>>
>> What should be done with these failures?
>>
>>
> I guess the first thing is to figure out what is going on, and that needs
> something repeatable. Maybe the test could be repeated with a set of seeds
> until the problem appears. Also, since 111 = 3*37, I wonder what happens
> with other odd primes besides 37. It might be worth testing with 97 and
> other such.
>
> Attached is a small script to test with larger primes. You can supply a
fixed seed but it's not even necessary. Typical result:
arraysize, max error
   1 0.0
  17 4.76837e-07
  37 2.98023e-06
  97 0.000104427
 313 0.000443935
 701 0.00112867
1447 0.00620008
2011 0.0138307
3469 0.16958

So even decimal=4 would fail for 97 already. For larger primes the FFT
should be slower but not less accurate, right?

If I use the numpy.fft.fft/ifft the result is:
   1 0.0
  17 7.54951656745e-15
  37 2.99760216649e-15
  97 4.99600361081e-15
 313 1.65423230669e-14
 701 5.66213742559e-15
1447 3.0253577421e-14
2011 1.87072579649e-14
3469 1.09912079438e-14

Looks a lot better.

Ralf
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.python.org/pipermail/scipy-dev/attachments/20100619/480cbd98/attachment.html>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: fftpack_failures.py
Type: application/octet-stream
Size: 540 bytes
Desc: not available
URL: <http://mail.python.org/pipermail/scipy-dev/attachments/20100619/480cbd98/attachment.obj>