How to structure multivariate timeseries when timeseries metadata varies I am forecasting from multivariate, geographical, timeseries data.
