PFK: Attitude: filter tranches

Daniel Burkhardt Cerigorequested to merge
polarization_footprint_kenya/filter_tranches into main

@andrewsutjahjo note it is configurable in the notebook. For most stuff we should only be using tranches [0, 1], as tranche=2 is an incomplete thus biased sample of the total data, so shouldn't be in any of the calculations. Could use tranche=2 in model training though (just not holdout).

@benjamincerigo I think its ok that included tranche=2 for the norm polarization stuff.

Merge request reports