Create rolling windows in pandas based on window size specified in another column

Question

I have a pandas dataframe that I'd like to sum on a rolling basis where the window is specified by another column.

For example,

values_to_sum	window_size	rolling_sum
1	6	17
2	5	16
1	2	4
3	5	19
4	5	NaN
6	4	NaN
2	3	NaN
4	3	NaN

Trying to call the column window_size within the rolling function results in the error ValueError: window must be an integer .

How can I call the column window_size on a row-by-row basis for the rolling function?

Answer 1

With a list comprehension:

df["rolling_sum"] = [np.nan 
                     if j + ws > len(df.index)
                     else df.values_to_sum.iloc[j: j+ws].sum()
                     for j, ws in enumerate(df.window_size)]

Put np.nan if the current index ( j ) plus window size ( ws ) exceeds the dataframe's length ( len(df.index) ); else get the window with iloc and sum it.

to get

   values_to_sum  window_size  rolling_sum
0              1            6         17.0
1              2            5         16.0
2              1            2          4.0
3              3            5         19.0
4              4            5          NaN
5              6            4          NaN
6              2            3          NaN
7              4            3          NaN

note: you can pre-define df_length = len(df.index) and use it to avoid looking for its length in the comprehension repeatedly.