I've seen the Databricks documentation on time series here. I'm using forecasts as a feature and those forecasts have both an as-of timestamp (when the forecast was generated) and a time step label (timestamp indicating the time of the forecasted observation).

Schema:

as_of_timestamp

observation_timestamp

value

Does anyone have thoughts on how best to structure this kind of data? Specifically, I'm wondering about how to specify the primary and timestamp keys when creating the feature.