The val_indices argument here https://github.com/apple/pfl-research/blob/develop/pfl/data/dataset.py#L164 is currently not implemented. fix it.
The correct behavior should be if val_indices are provided, then split method should split the user dataset into a train and val partition dataset objects according to those indices and not according to the input arguments of split.
This is useful if you have static partitions within each individual user.
The
val_indices
argument here https://github.com/apple/pfl-research/blob/develop/pfl/data/dataset.py#L164 is currently not implemented. fix it. The correct behavior should be if val_indices are provided, thensplit
method should split the user dataset into a train and val partition dataset objects according to those indices and not according to the input arguments ofsplit
. This is useful if you have static partitions within each individual user.(115345691)