tff.simulation.datasets.ClientData

Object to hold a federated dataset.

The federated dataset is represented as a list of client ids, and a function to look up the local dataset for each client id.

Each client's local dataset is represented as a tf.data.Dataset, but generally this class (and the corresponding datasets hosted by TFF) can easily be consumed by any Python-based ML framework as numpy arrays:

import tensorflow as tf
import tensorflow_federated as tff
import tensorflow_datasets as tfds

for client_id in sampled_client_ids[:5]:
  client_local_dataset = tfds.as_numpy(
      emnist_train.create_tf_dataset_for_client(client_id))
  # client_local_dataset is an iterable of structures of numpy arrays
  for example in client_local_dataset:
    print(example)

If desiring a manner for constructing ClientData objects for testing purposes, please see the tff.simulation.datasets.TestClientData class, as it provides an easy way to construct toy federated datasets.

client_ids A list of string identifiers for clients in this dataset.
dataset_computation A tff.Computation accepting a client ID, returning a dataset.

element_type_structure The element type information of the client datasets.

elements returned by datasets in this ClientData object.

serializable_dataset_fn A callable accepting a client ID and returning a tf.data.Dataset.

Note that this callable must be traceable by TF, as it will be used in the context of a tf.function.

Methods

create_tf_dataset_for_client

View source

Creates a new tf.data.Dataset containing the client training examples.

This function will create a dataset for a given client, given that client_id is contained in the client_ids property of the ClientData. Unlike create_dataset, this method need not be serializable.

Args
client_id The string client_id for the desired client.

Returns
A tf.data.Dataset object.

create_tf_dataset_from_all_clients

View source

Creates a new tf.data.Dataset containing all client examples.

This function is intended for use training centralized, non-distributed models (num_clients=1). This can be useful as a point of comparison against federated models.

Currently, the implementation produces a dataset that contains all examples from a single client in order, and so generally additional shuffling should be performed.

Args
seed Optional, a seed to determine the order in which clients are processed in the joined dataset. The seed can be any nonnegative 32-bit integer, an array of such integers, or None.

Returns
A tf.data.Dataset object.

datasets

View source

Yields the tf.data.Dataset for each client in random order.

This function is intended for use building a static array of client data to be provided to the top-level federated computation.

Args
limit_count Optional, a maximum number of datasets to return.
seed Optional, a seed to determine the order in which clients are processed in the joined dataset. The seed can be any nonnegative 32-bit integer, an array of such integers, or None.

from_clients_and_tf_fn

View source

Constructs a ClientData based on the given function.

Args
client_ids A non-empty list of strings to use as input to create_dataset_fn.
serializable_dataset_fn A function that takes a client_id from the above list, and returns a tf.data.Dataset. This function must be serializable and usable within the context of a tf.function and tff.Computation.

Raises
TypeError If serializable_dataset_fn is a tff.Computation.

Returns
A ClientData object.

preprocess

View source

Applies preprocess_fn to each client's data.

Args
preprocess_fn A callable accepting a tf.data.Dataset and returning a preprocessed tf.data.Dataset. This function must be traceable by TF.

Returns
A tff.simulation.datasets.ClientData.

Raises
IncompatiblePreprocessFnError If preprocess_fn is a tff.Computation.

train_test_client_split

View source

Returns a pair of (train, test) ClientData.

This method partitions the clients of client_data into two ClientData objects with disjoint sets of ClientData.client_ids. All clients in the test ClientData are guaranteed to have non-empty datasets, but the training ClientData may have clients with no data.

Args
client_data The base ClientData to split.
num_test_clients How many clients to hold out for testing. This can be at most len(client_data.client_ids) - 1, since we don't want to produce empty ClientData.
seed Optional seed to fix shuffling of clients before splitting. The seed can be any nonnegative 32-bit integer, an array of such integers, or None.

Returns
A pair (train_client_data, test_client_data), where test_client_data has num_test_clients selected at random, subject to the constraint they each have at least 1 batch in their dataset.

Raises
ValueError If num_test_clients cannot be satistifed by client_data, or too many clients have empty datasets.