Multiple time series

[metahexane]

I have a data set which contains multiple time series (of the same sort), so there is, besides 'date', another column, indicating the group of the time series. Is there any way to run this with multiple groups?

[Secilia-Cxy]

Could you please provide more information on the size of the dataset? Specifically, how many time series groups are there? Depending on the scale of the groups, we might be able to apply different methods to handle the data more efficiently.

[metahexane]

There are about 3k groups, for a total of about 1M rows, so each time series has like a year's length (365 days).

[hanlu-nju]

There are about 3k groups, for a total of about 1M rows, so each time series has like a year's length (365 days).

It seems to be a problem related to categorical features. You can try training an embedding for each group and add it to the embedding of each series.