jax_privacy.batch_selection.FixedBatchSampling#
- class jax_privacy.batch_selection.FixedBatchSampling(batch_size, iterations, replace=False)[source]#
Bases:
BatchSelectionStrategyImplements fixed-size batch sampling.
Each batch is sampled uniformly at random from the dataset. By default, batches are sampled without replacement within a batch, and with replacement across batches (i.e., the same example can appear in multiple iterations).
References
https://arxiv.org/abs/1808.00087 (sampling without replacement)
- Variables:
batch_size – The number of examples per batch.
iterations – The number of total iterations / batches to generate.
replace – Whether to sample with replacement within each batch.
Methods
__init__Yields 1D batches of data indices.
Attributes
-
batch_size:
int#
-
iterations:
int#
-
replace:
bool= False#