tff.framework.ComposingExecutorFactory

View source on GitHub

Factory class encapsulating executor compositional logic.

Inherits From: ExecutorFactory

tff.framework.ComposingExecutorFactory(
    *,
    max_fanout: int,
    unplaced_ex_factory: tff.framework.UnplacedExecutorFactory,
    flat_stack_fn: Callable[[executor_factory.CardinalitiesType], Sequence[executor_base.Executor]
        ],
    local_computation_factory: local_computation_factory_base.LocalComputationFactory = tensorflow_computation_factory.TensorFlowComputationFactory()
)

This class is responsible for aggregating lists of executors into a compositional hierarchy based on the max_fanout parameter.

Methods

clean_up_executor

View source

clean_up_executor(
    cardinalities: tff.framework.CardinalitiesType
)

Holds no executors internally, so passes on cleanup.

create_executor

View source

create_executor(
    cardinalities: tff.framework.CardinalitiesType
) -> tff.framework.Executor

Creates an executor hierarchy of maximum width self._max_fanout.

First creates a flat list of executors to aggregate, using the flat_stack_fn passed in at initialization. Then composes this list into a hierarchy of width at most self._max_fanout.

Args
cardinalities	A mapping from placements to integers specifying the cardinalities at each placement

Returns
An executor_base.Executor satisfying the conditions above.