Poisson (exponential) window for pandas.Series.rolling

[stroobandt]

Problem Description

In pandas.Series.rolling, the SciPy window type exponential has not been implemented. This win_type would be extremely useful for analysing biological processes. It should be pretty easy to add.

Please, note that despite the similar name, the proposed method is mathematically different from the implemented pandas.Series.ewm method.

Enhancement Request

Taking for example df.rolling(51, win_type='exponential', tau=10), the implemented SciPy window would be the following asymmetrical window:

from scipy import signal
window = 51
tau = 10
w = signal.exponential(window, tau=tau, sym=False, center=0)
array([1.        , 0.90483742, 0.81873075, 0.74081822, 0.67032005,
       0.60653066, 0.54881164, 0.4965853 , 0.44932896, 0.40656966,
       0.36787944, 0.33287108, 0.30119421, 0.27253179, 0.24659696,
       0.22313016, 0.20189652, 0.18268352, 0.16529889, 0.14956862,
       0.13533528, 0.12245643, 0.11080316, 0.10025884, 0.09071795,
       0.082085  , 0.07427358, 0.06720551, 0.06081006, 0.05502322,
       0.04978707, 0.0450492 , 0.0407622 , 0.03688317, 0.03337327,
       0.03019738, 0.02732372, 0.02472353, 0.02237077, 0.02024191,
       0.01831564, 0.01657268, 0.01499558, 0.01356856, 0.01227734,
       0.011109  , 0.01005184, 0.00909528, 0.00822975, 0.00744658,
       0.00673795])

Expected Output

>>> df = pd.DataFrame({'B': [1, 0, 0, 0, 0]})
>>> df.rolling(51, win_type='exponential', tau=10)
     B
0  1.0
1  0.90483742
2  0.81873075
3  0.74081822
4  0.67032005

[jreback]

we already support scipy types so this would be very easy to add - a pull request would make this happen
http://pandas.pydata.org/pandas-docs/stable/computation.html#rolling-windows

[stroobandt]

As far as I can see, only these files need to be amended:

• pandas/pandas/core/window.py
• pandas/pandas/tests/test_window.py
• pandas/doc/source/computation.rst