[SciPy-Dev] DST I precision (was ANN: SciPy 0.11.0 release candidate 1)

[Ralf Gommers]

On Sat, Jul 21, 2012 at 6:06 PM, John Hassler <hasslerjc at comcast.net> wrote:

>
> On 7/21/2012 10:43 AM, Ralf Gommers wrote:
>
>
> On Fri, Jul 20, 2012 at 9:57 PM, John Hassler <hasslerjc at comcast.net>wrote:
>
>  ======================================================================
>> FAIL: test_definition (test_real_transforms.TestDSTIDouble)
>>
>> ----------------------------------------------------------------------
>> Traceback (most recent call last):
>>    File
>> "C:\Python32\lib\site-packages\scipy\fftpack\tests\test_real_transforms.py",
>> line 213, in test_definition
>>     err_msg="Size %d failed" % i)
>>   File "C:\Python32\lib\site-packages\numpy\testing\utils.py", line 800,
>> in assert_array_almost_equal
>>     header=('Arrays are not almost equal to %d decimals' % decimal))
>>   File "C:\Python32\lib\site-packages\numpy\testing\utils.py", line 636,
>> in assert_array_compare
>>     raise AssertionError(msg)
>> AssertionError:
>> Arrays are not almost equal to 15 decimals
>> Size 256 failed
>> (mismatch 3.515625%)
>>  x: array([  1.00000000e+00,  -5.03902743e-01,   3.33300126e-01,
>>         -2.51913719e-01,   1.99940224e-01,  -1.67900640e-01,
>>          1.42771743e-01,  -1.25881543e-01,   1.11000399e-01,...
>>  y: array([  1.00000000e+00,  -5.03902743e-01,   3.33300126e-01,
>>         -2.51913719e-01,   1.99940224e-01,  -1.67900640e-01,
>>          1.42771743e-01,  -1.25881543e-01,   1.11000399e-01,...
>>
>
> This looks like the test precision being a little too high. Could you
> adjust it to decimal=14 or lower, and tell us when the test passes on your
> machine?
>
> Thanks,
> Ralf
>
>
> decimal = 14 passes.
> john
>

This test and the failure look a bit odd, so I just want to make sure: is
it OK to lower the test precision? DST II passes with decimal=15 and DST
III with decimal=16, so not sure why DST I is worse. Looks like this test
could use some improvement:

            # XXX: we divide by np.max(y) because the tests fail otherwise.
We
            # should really use something like assert_array_approx_equal.
The
            # difference is due to fftw using a better algorithm w.r.t error
            # propagation compared to the ones from fftpack.
            assert_array_almost_equal(y / np.max(y), yr / np.max(y),
decimal=self.dec,
                    err_msg="Size %d failed" % i)

Any takers?

Ralf
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.python.org/pipermail/scipy-dev/attachments/20120721/2926ed6e/attachment.html>