The ensemble framework is very useful in that it allows us to combine multiple models, but the current code performs the ensemble for each country-commodity pair individually. It would be interesting to see if a "commodity-level" model would provide benefit in the ensemble model. By "commodity-level" model, I mean a model such as EM or a mixed model which imputes values for one commodity across all countries. I could also imagine using EM for "country-level" models (imputing all commodities for one country simultaneously) or even global models (that work on the entire dataset of all countries/commodities simultaneously).
Implementing this framework would also allow us to use ensembles for the yield imputation, in which we could use the mixed model as a base learner.
The ensemble framework is very useful in that it allows us to combine multiple models, but the current code performs the ensemble for each country-commodity pair individually. It would be interesting to see if a "commodity-level" model would provide benefit in the ensemble model. By "commodity-level" model, I mean a model such as EM or a mixed model which imputes values for one commodity across all countries. I could also imagine using EM for "country-level" models (imputing all commodities for one country simultaneously) or even global models (that work on the entire dataset of all countries/commodities simultaneously).
Implementing this framework would also allow us to use ensembles for the yield imputation, in which we could use the mixed model as a base learner.