[SciPy-User] Some thing wrong in scipy reference
Ralf Gommers
ralf.gommers@googlemail....
Sat Jul 21 09:39:35 CDT 2012
On Sat, Jul 21, 2012 at 1:26 PM, Warren Weckesser <
warren.weckesser@enthought.com> wrote:
>
>
> On Sat, Jul 21, 2012 at 3:02 AM, Ralf Gommers <ralf.gommers@googlemail.com
> > wrote:
>
>>
>>
>> 2012/7/21 王瑞期 <shahuwang@qq.com>
>>
>>>
>>> http://docs.scipy.org/doc/scipy/reference/generated/scipy.cluster.vq.whiten.html#scipy.cluster.vq.whiten
>>>
>>> In the bottom of this page,the result of the whiten function is wrong.
>>>
>>
>> You're right, the current output as documented is off by a constant
>> factor:
>> >>> res = whiten(features)
>> >>> res * 3.41250074 / res[0, 0] # reproduce docstring output
>> array([[ 3.41250074, 2.20300046, 5.88897275],
>> [ 2.69407953, 2.39456571, 7.62102356],
>> [ 1.43684242, 0.57469577, 5.88897275]])
>>
>> Does anyone know why, and is this result correct?
>>
>> Ralf
>>
>>
>
> whiten(obs) returns (in effect) obs / obs.std(axis=0). It appears that
> the docstring shows obs / obs.std(axis=0, ddof=1):
>
> In [14]: obs
> Out[14]:
> array([[ 1.9, 2.3, 1.7],
> [ 1.5, 2.5, 2.2],
> [ 0.8, 0.6, 1.7]])
>
> In [15]: vq.whiten(obs)
> Out[15]:
> array([[ 4.17944278, 2.69811351, 7.21248917],
> [ 3.29956009, 2.93273208, 9.33380951],
> [ 1.75976538, 0.7038557 , 7.21248917]])
>
> In [16]: obs / obs.std(axis=0)
> Out[16]:
> array([[ 4.17944278, 2.69811351, 7.21248917],
> [ 3.29956009, 2.93273208, 9.33380951],
> [ 1.75976538, 0.7038557 , 7.21248917]])
>
> In [17]: obs / obs.std(axis=0, ddof=1)
> Out[17]:
> array([[ 3.41250074, 2.20300046, 5.88897275],
> [ 2.69407953, 2.39456571, 7.62102355],
> [ 1.43684242, 0.57469577, 5.88897275]])
>
>
> The whiten function has not changed in a long time. Has the
> default value of ddof in the numpy std() method changed in
> the last five (or more) years?
>
Four years ago, in 62e99493, a ddof kw was added. But that looks like a
backwards-compatible commit to me. I'll change the output in the whiten
docstring.
Ralf
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mail.scipy.org/pipermail/scipy-user/attachments/20120721/43f415d5/attachment-0001.html
More information about the SciPy-User
mailing list