Any way to explicitly specify which columns to use as features?

[pchalasani]

This is a simplified version of my scenario. Say I have a df with these columns: groupID, time_idx, x, y. At any time T, I want to use the values of x at times T-3, T-2, T-1, T, to "predict" the value of y as of time T. For example y could be some function of future outcomes or variables.

This is not strictly speaking a "forecasting" problem, perhaps more of a "regression" problem, since we are predicting something other than the input. But this seems like a very basic scenario that I don't see how to specify using any combination of the time_varying_known/unknown_reals or other settings.

More generally it would be nice to have a way to explicitly say which columns are to be used as features to the encoder, and which columns to treat as "target labels", so that a loss can be computed as a function of the decoder output and the "target labels".

Not sure if I am missing something basic. Thanks