cookup_train_1.h5: ~10% clients hold ~82% data
cookup_train_2.h5: ~20% clients hold ~90% data
cookup_train_3.h5: ~50% clients hold ~98% data
test.h5: original test set

Can be downloaded from: https://dataverse.harvard.edu/privateurl.xhtml?token=318de241-985a-4c99-8d36-ae0281d086d4