Module: tff.simulation.datasets

Datasets for running TensorFlow Federated simulations.

Modules

celeba module: Libraries for the federated CelebA dataset for simulation.

cifar100 module: Libraries for the federated CIFAR-100 dataset for simulation.

emnist module: Libraries for the federated EMNIST dataset for simulation.

flair module: Libraries for loading the FLAIR dataset.

gldv2 module: Libraries for the federated Google Landmark v2 dataset for simulation.

inaturalist module: Libraries for the federated iNaturalist dataset for simulation.

shakespeare module: Libraries for the Shakespeare dataset for federated learning simulation.

stackoverflow module: Libraries for the Stackoverflow dataset for federated learning simulation.

Classes

class ClientData: Object to hold a federated dataset.

class FilePerUserClientData: A tff.simulation.datasets.ClientData that maps a set of files to a dataset.

class SqlClientData: A tff.simulation.datasets.ClientData backed by an SQL file.

class TestClientData: A tff.simulation.datasets.ClientData intended for test purposes.

class TransformingClientData: Transforms client data, potentially expanding by adding pseudo-clients.

Functions

build_dataset_mixture(...): Build a new dataset that probabilistically returns examples.

build_single_label_dataset(...): Build a new dataset that only yields examples with a particular label.

build_synthethic_iid_datasets(...): Constructs an iterable of IID clients from a tff.simulation.datasets.ClientData.

load_and_parse_sql_client_data(...): Load a ClientData arises by parsing a serialized SqlClientData.

save_to_sql_client_data(...): Serialize a federated dataset into a SQL database compatible with SqlClientData.