Multiple time series with a different time span approach?

Question

I'm working with 3706 time series loacated in the same file. But the time series time span is different for some of them.

Some of the time series start at 01/01 and end in 12/31 But some straight up start in 11/24 to 12/25 They are daily timeseries.

And the dataframe look somethings like this (I can't provided the real data sorry it's confidential)

Date	Product	Customer
01/01	1	Cust 1
01/01	5	Cust 1
01/02	1	Cust 1
01/02	1	Cust 2

I tried using the dart and use LGBM forecasting it but it seems to error due to the different time span.

Which approach should I used to make it on the same time span. If there's no good option is there any other model that can handle it?

I tried to add the missing date using this code but it just crashes my computer. I assumed due to it using too much memory.

unique_combinations = df[['Product', 'Customer']].drop_duplicates()

date_ranges = unique_combinations.groupby(['Product', 'Customer']).agg({'Date': ['min', 'max']})

start_date = pd.to_datetime('2023-01-01')
end_date = pd.to_datetime('2023-12-31')
date_range = pd.date_range(start_date, end_date, freq='D')
template_df = pd.DataFrame({'Date': date_range})

merged_dfs = []
for combination in unique_combinations.itertuples(index=False):
    product, customer = combination
    date_min, date_max = date_ranges.loc[(product, customer), ('Date', 'min')], date_ranges.loc[(product, customer), ('Date', 'max')]
    merged_df = pd.merge(template_df, df[(df['Product'] == product) & (df['Customer'] == customer)], on='Date', how='left')
    merged_df = merged_df[(merged_df['Date'] >= date_min) & (merged_df['Date'] <= date_max)]  # Optional: Trim the merged DataFrame to the available date range
    merged_dfs.append(merged_df)

final_df = pd.concat(merged_dfs)

final_df.fillna(0, inplace=True)  # Replace missing values with zeros or appropriate fill value

I expected the same Date span for all the times series

Thanks in advance.

Multiple time series with a different time span approach?

0 Answers0