tff.learning.ClientFedAvg

Client TensorFlow logic for Federated Averaging.

Inherits From: ClientDeltaFn

tff.learning.ClientFedAvg(
    model: tff.learning.Model,
    optimizer: tf.keras.optimizers.Optimizer,
    client_weighting: Union[ClientWeighting, ClientWeightFnType] = tff.learning.ClientWeighting.NUM_EXAMPLES,
    use_experimental_simulation_loop: bool = False
)

Args
model	A tff.learning.Model instance.
optimizer	A tf.keras.Optimizer instance.
client_weighting	A value of tff.learning.ClientWeighting that specifies a built-in weighting method, or a callable that takes the output of model.report_local_outputs and returns a tensor that provides the weight in the federated average of model deltas.
use_experimental_simulation_loop	Controls the reduce loop function for input dataset. An experimental reduce loop is used for simulation.

Attributes
variables	Returns all the variables of this object. Note this only includes variables that are part of the state of this object, and not the model variables themselves.

Methods

__call__

View source

@tf.function
__call__(
    dataset, initial_weights
)