GeoLift

GeoLift copied to clipboard

Added 'stratified' as an alternative sampling method. This stratifies the locations in rank_by_loc and then randomizes the cell assignment within each strata.

The current 'systematic' method only randomizes sample(1:k,k) once, and then repeats that sequence down the rank_by_loc. This produces identical groupings of markets in data_aux for MultiCellMarketSelection each time we run it, even though the cell number labels may differ. This is a potentially significant source of bias when running MultiCellMarketSelection for subsequent experiments with the same number of cells. By randomizing at each strata, we can ensure the market groupings in data_aux will be truly different by setting a different seed.