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_weight_fn: Optional[Callable[[Any], tf.Tensor]] = None,
    use_experimental_simulation_loop: bool = False
)

model A tff.learning.Model instance.
optimizer A tf.keras.Optimizer instance.
client_weight_fn an optional 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. If not provided, the default is the total number of examples processed on device.
use_experimental_simulation_loop Controls the reduce loop function for input dataset. An experimental reduce loop is used for simulation.

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
)