jax_privacy.batch_selection.MinSepRandomAllocation#

class jax_privacy.batch_selection.MinSepRandomAllocation(total_participations, iterations, min_sep=1, random_rotation=False, sampling_prob=1.0)[source]#

Bases: BatchSelectionStrategy

Implements k-out-of-t b-min-sep random allocation.

This implements Algorithm 2 of https://arxiv.org/abs/2610.12712 (and the random-rotation variant in Appendix B when random_rotation=True). Each example independently selects \(k\) steps in [0, iterations) uniformly at random over all subsets of size \(k\) whose consecutive participations are separated by at least min_sep (\(b\)) iterations.

References

Variables:
  • total_participations – The number of steps each example is allocated to participate in before subsampling (k).

  • iterations – The total number of iterations / batches to generate (t).

  • min_sep – The minimum separation between consecutive participations of the same example (b). Defaults to 1.

  • random_rotation – Whether to place participations on a cycle and apply a uniform random rotation for a uniform per-iteration marginal. Defaults to False.

  • sampling_prob – Probability of independently keeping each allocated participation. Defaults to 1.0 (no subsampling).

Methods

__init__

batch_iterator

Yields 1D batches of data indices.

Attributes

total_participations: int#
iterations: int#
min_sep: int = 1#
random_rotation: bool = False#
sampling_prob: float = 1.0#
batch_iterator(num_examples, rng=None)[source]#

Yields 1D batches of data indices.

Return type:

Iterator[ndarray]