[SciPy-User] Question about scikits.timeseries.lib.moving_funcs.mov_average

[Wes McKinney]

On Wed, Jun 23, 2010 at 10:40 AM, Pierre GM <pgmdevlist at gmail.com> wrote:
>
> On Jun 23, 2010, at 8:11 AM, Andreas wrote:
>
>> Hi,
>>
>> in the docstring to mov_average() it says:
>>
>>   The result will also be masked at i if any of the input values in the
>>   slice ``[i-span:i+1]`` are masked
>>
>> Is there any way to prevent this behaviour?
>
> Nope. That limits nasty surprises otherwise.
>
>> I have a very patchy
>> timeseries (one value every 3 to 6 days), and I'd like to use the
>> mov_average function to smooth these data.
>> Any idea how to do that?
>> mov_average(data,20) would have been perfect, if not for the masked values
>> ...
>> What I tried so far is
>>
>>   mov_average(data.compressed(),20)
>>
>> but that has the same size as data.compressed(). I would really like to
>> have daily values ..
>
> You could try to fill your missing values beforehand, w/ functions like backward_fill and forward_fill, then passing your series to mov_average. Or the reverse way: compress your data to get rid of the missing values, pass it to mov_average, reconvert it to a daily series w/ fill_missing_dates (to get the right number of dates), then fill it w/ backward_fill or forward_fill.
>
> Let me know how it goes.
> _______________________________________________
> SciPy-User mailing list
> SciPy-User at scipy.org
> http://mail.scipy.org/mailman/listinfo/scipy-user
>

In pandas all my moving window functions accept a "min_periods"
argument so that it will place a value in a data hole assuming there
are sufficient observations in window:

In [4]: arr
Out[4]:
array([  0.,   1.,   2.,   3.,   4.,   5.,   6.,   7.,   8.,   9.,  NaN,
        NaN,  NaN,  NaN,  NaN,  15.,  16.,  17.,  18.,  19.])

In [5]: rolling_mean(arr, 10, min_periods=1)
Out[5]:
array([  0. ,   0.5,   1. ,   1.5,   2. ,   2.5,   3. ,   3.5,   4. ,
         4.5,   5. ,   5.5,   6. ,   6.5,   7. ,   9. ,  11. ,  13. ,
        15. ,  17. ])

Have you thought of adding this functionality to scikits.timeseries?