I am trying to do a simple rolling multiplication of a dataframe (each new value should be the product of all the input values in the window). Realizing that rolling does not allow of products, I tried to look for alternate solutions but they seem not to work with variable lenghth windows :
start_date = '2024-05-01'
date_index = pd.date_range(end=start_date, periods=15, freq='B')
df = pd.DataFrame(list(range(1,16)), index = date_index, columns=['A'])
df.rolling('1W').agg({"A": "prod"})
This return the error : ValueError: <Week: weekday=6> is a non-fixed frequency
Is there any clean workarround for that type of rolling ?
Is this what you're looking for, a 7 day window? The documentation says you can provide a pd.Timedelta as parameter to pd.DataFrame.rolling
>>> df.rolling(pd.Timedelta(days=7)).agg(lambda a: np.prod(a))
... # works also: df.rolling(pd.Timedelta('1W')).agg(lambda a: np.prod(a))
... # as well as: df.rolling(pd.Timedelta(weeks=1)).agg(lambda a: np.prod(a))
A
2024-04-11 1.0
2024-04-12 2.0
2024-04-15 6.0
2024-04-16 24.0
2024-04-17 120.0
2024-04-18 720.0
2024-04-19 2520.0
2024-04-22 6720.0
2024-04-23 15120.0
2024-04-24 30240.0
2024-04-25 55440.0
2024-04-26 95040.0
2024-04-29 154440.0
2024-04-30 240240.0
2024-05-01 360360.0
You have to use a lambda it seems; just the method throws an error. Or is your window type the data from the last Friday? If so, use a groupby with offsets (pd.offset.Weekday(weekday=THE_DAY)).