Subgrouped lag transformations

[concimuscb]

Is there any way to perform grouped lagged transformations?
Say for example that the unique ID is in the format cat_id@sku_id and I would like to add the expanding mean of each category to all its skus.

How could I accomplish this with MLForecast?

[jmoralez]

Hey. This is the same as what is requested in #194, right?

[concimuscb]

Hello, not sure if it is exactly the same. As far as I understood, #194 references the two "extreme" cases: transformations per unique_id and transformations for the sum of all unique_id.

This would be some kind of middle ground where ids are first grouped and then transformations are applied on a group per group basis.

[jmoralez]

The idea of that issue is being able to provide some columns to aggregate by first and then computing the transformations on the aggregations.

[concimuscb]

Then I believe it would be addressed yes.