pandas: rolling with periodic boundaries

Question

Is it possible to include the start and end index values in when using the rolling function in pandas such that the values of the series are periodic?

For example, I have the following Series instance:

import pandas as pd
timestamps = [1, 2, 3, 4]
quantities = [1, 16, 9, 4]
series = pd.Series(quantities, index=timestamps)

which when I type:

series.rolling(window=3, min_periods=3, center=True).median()

yields:

1    NaN
2    9.0
3    9.0
4    NaN
dtype: float64

However, I'd like to include either ends of the series in the calculation such that the result is:

1    median([4, 1, 16])   4.0
2    median([1, 16, 9])   9.0
3    median([16, 9, 4])   9.0
4    median([9, 4, 1])    4.0
dtype: float64

Thanks for any help trying to achieve this!

Would make more sense if the last one was `9, 4, 1` – Erfan Apr 04 '19 at 23:48 — Erfan, Apr 04 '19 at 23:48
@Erfan yep, and correct :-) – ajrlewis Apr 04 '19 at 23:56 — ajrlewis, Apr 04 '19 at 23:56

score 1 · Accepted Answer · answered Apr 05 '19 at 00:12

We can extend the list with the first and last element and then calculate the rolling median. After that we dropna and assign the series back

Wont win the price for most beautiful code, but it does the job. Could not think of another way.

quantities2 = quantities.copy()
quantities2.insert(0, quantities2[-1])
quantities2.insert(len(quantities2), quantities2[0])
print(quantities2)

[4, 1, 16, 9, 4, 4]

series = pd.Series(quantities2)
series = pd.Series(series.rolling(window=3, min_periods=3, center=True).median().dropna(), index=timestamps)

print(series)
1    4.0
2    9.0
3    9.0
4    4.0
dtype: float64

But then again, do we have to write cleanest code? :)

pandas: rolling with periodic boundaries

1 Answers1