tff.framework.DataDescriptor

Represents fully-specified data-yielding computations.

Inherits From: CardinalityFreeDataDescriptor, Ingestable, TypedObject, CardinalityCarrying

tff.framework.DataDescriptor(
    comp: Optional[tff.Computation],
    arg: object,
    arg_type: tff.types.Type,
    cardinality: Optional[int] = None
)

Similar to CardinalityFreeDataDescriptor, but additionally accepts a cardinality argument, allowing callers to explcitly specify the number of clients this data descriptor is intended to represent.

comp The computation that materializes the data, of some type (T -> U) where T is the type of the argument arg and U is the type of the materialized data that's being produced. This can be None, in which case it's assumed to be an identity function (and T in that case must be identical to U).
arg The argument to be passed as input to comp if comp is not None, or to be treated as the computed result. Must be recognized by the TFF runtime as a payload of type T.
arg_type The type of the argument (T references above). An instance of tff.Type.
cardinality If of federated type, placed at clients, this int specifies the number of clients represented by this DataDescriptor.

ValueError if the arguments don't satisfy the constraints listed above.

cardinality Returns the cardinality information associated with this object.
type_signature Returns the TFF type of this object (an instance of tff.Type).

Methods

ingest

View source

ingest(
    executor
)

Causes this object to ingest itself into the given executor.

Args
executor An instance of executor_base.Executor to ingest into.

Returns
An instance of executor_value_base.ExecutorValue returned by the executor.