Save the date! Google I/O returns May 18-20 Register now

Module: tff.simulation

Libraries for running Federated Learning simulations.


datasets module: Datasets for running Federated Learning simulations.

models module: Models for running Federated Learning simulations.


class CSVMetricsManager: Utility class for saving/loading experiment metrics via a CSV file.

class ClientData: Object to hold a federated dataset.

class FileCheckpointManager: A checkpoint manager backed by a file system.

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

class FromTensorSlicesClientData: ClientData based on

class HDF5ClientData: A tff.simulation.ClientData backed by an HDF5 file.

class MetricsManager: An abstract base class for metrics managers.

class TensorBoardManager: Utility class for saving metrics using tf.summary.

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


build_uniform_client_sampling_fn(...): Builds a function that (pseudo-)randomly samples clients.

compose_dataset_computation_with_computation(...): Builds a new tff.Computation which constructs datasets on clients.

compose_dataset_computation_with_iterative_process(...): Builds a new iterative process which constructs datasets on clients.

run_server(...): Runs a gRPC server hosting a simulation component in this process.

server_context(...): Context manager yielding gRPC server hosting simulation component.