Missed TensorFlow World? Check out the recap. Learn more

tff.FederatedType

View source on GitHub

Class FederatedType

An implementation of tff.Type representing federated types in TFF.

Inherits From: Type

Used in the tutorials:

__init__

View source

__init__(
    member,
    placement,
    all_equal=None
)

Constructs a new federated type instance.

Args:

  • member: An instance of tff.Type (or something convertible to it) that represents the type of the member components of each value of this federated type.
  • placement: The specification of placement that the member components of this federated type are hosted on. Must be either a placement literal such as tff.SERVER or tff.CLIENTS to refer to a globally defined placement, or a placement label to refer to a placement defined in other parts of a type signature. Specifying placement labels is not implemented yet.
  • all_equal: A bool value that indicates whether all members of the federated type are equal (True), or are allowed to differ (False). If all_equal is None, the value is selected as the default for the placement, e.g., True for tff.SERVER and False for tff.CLIENTS.

Properties

all_equal

member

placement

Methods

__eq__

View source

__eq__(other)

Determines whether two type definitions are identical.

Note that this notion of equality is stronger than equivalence. Two types with equivalent definitions may not be identical, e.g., if they represent templates with differently named type variables in their definitions.

Args:

  • other: The other type to compare against.

Returns:

True iff type definitions are syntatically identical (as defined above), or False otherwise.

Raises:

  • NotImplementedError: If not implemented in the derived class.

__ne__

View source

__ne__(other)

Return self!=value.

compact_representation

View source

compact_representation()

Returns the compact string representation of this type.

formatted_representation

View source

formatted_representation()

Returns the formatted string representation of this type.