Train a generalizable model on range of data, evaluate

[nlebovits]

Train and store an RF classifier on a sampling of data from many countries, then see how it performs on a benchmark dataset relative to a classifier trained on data from that country alone

[nlebovits]

Gerald suggests training a model on all of Central America given 1) how comparable the land is and 2) how useful it will be to a range of people