We have data for 6 years, how should we split in this case? Should we just ignore 2018?
2018 can be use in testing set.
