leaf
leaf copied to clipboard
support unbalanced iid data
hi,
Thanks for publishing this lib. AFAIK, this library supports either original, non-i.i.d. data partitions, or i.i.d. equal-sized partitions. Have you considered supporting a third option, where the size of each partition is maintained but the data is being re-samples i.i.d from the dataset? (useful for some experiments, should be a part of the benchmark IMO)