tff.simulation.baselines.ClientSpec

Contains information for configuring clients within a training task.

num_epochs An integer representing the number of passes each client performs over its entire local dataset.
batch_size An integer representing the batch size used when iterating over client datasets.
max_elements An optional positive integer governing the maximum number of examples used by each client. By default, this is set to None in which case clients use their full dataset.
shuffle_buffer_size An optional positive integer specifying the shuffle buffer size to use. If set to None, a default value suitable for the task's dataset will be used. If set to 1, no shuffling occurs.