pandas.DataFrame.rolling summing the current row value and the next 2 row value based on Id and time index, to the 1st row's value

Question

GOAL

I want to sum the current row value and the next 2 row value based on ID and time index to the 1st row value. Time index moves forward with a 3day window while the summed values can only have the same ID

The current incorrect functioning

• 1st row Value SUMMED VALUE = 1st row Value
• 2nd row value SUMMED VALUE = 1st row value + 2nd row Value
• 3rd row Value SUMMED VALUE = 3rd row Value + 2nd row Value + 1st row Value

CODE

gr = current_df.groupby("id").rolling('3D').rolling_sum_3.sum()
final_dataframe = pd.DataFrame(gr)

Output

Goal is the following

• A.) 1st row Value SUMMED VALUE = 3rd row Value + 2nd row Value + 1st row Value
• B.) if there is no 2nd or 3rd row value in this ID and 3 day timeframe just leave this field with a NaN value
• A.) 2nd row value SUMMED VALUE = 2nd row Value + 3rd row Value + 4th row Value
• B.) if there is no 3rd or 4th row value in this ID and 3 day timeframe just leave this field with a NaN value
• etc.

I have tried

• using if and for cycles -> runs for very long
• I need to keep this quick pandas format and have to change maybe the rolling window's win_type property pandas documentation so I can maybe roll the window the why I want but I don't fully understand these options - https://docs.scipy.org/doc/scipy/reference/signal.windows.html#module-scipy.signal.windows
• I can not shift everything up because there are multiple month difference between the happenings and the goal is to just sum values that are in 3 days range to each other
• Nan values are at the top not at the bottom - how to get a continuous rolling mean in pandas?
• No grouping - ValueError: index must be monotonic when applying rolling("2H").mean()
• as I have mentioned before using for and if is not realy an option because that slows down the computation and I have couple hundreds of thousands of rows. Currently it takes 3s. With similar to those methods in the answer even 20 min - Moving average on pandas.groupby object that respects time

Answer 1

• Reversing the series and then taking a sum using a rolling window of 3.
• The backwards looking rolling sum will then be a forward looking rolling sum because you reversed the sorting order