tff.structure.Struct

Represents a struct-like structure with named and/or unnamed fields.

tff.structure.Struct(
    elements: Iterable[tuple[Optional[str], _T]]
)

Structs are similar to collections.namedtuple in that their elements can be accessed by name or by index. However, Structs provide a performance improvement over collections.namedtuple by using a single class to represent values with many different possible structures, rather than creating a brand new class for every new instance.

Structs are commonly used inside Tensorflow Federated as a standard intermediate representation of other structure types, including lists, tuples, dicts, namedtuples, and attr.s classes.

Example:

x = Struct([('foo', 10), (None, 20), ('bar', 30)])

len(x) == 3
x[0] == 10
x[1] == 20
x[2] == 30
list(iter(x)) == [10, 20, 30]
dir(x) == ['bar', 'foo']
x.foo == 10
x['bar'] == 30

Note that field names are optional, allowing Struct to be used like an ordinary positional tuple.

elements An iterable of element specifications, each being a pair consisting of the element name (either str, or None), and the element value. The order is significant.

TypeError if the elements are not a list, or if any of the items on the list is not a pair with a string at the first position.

Methods

named

View source

@classmethod
named(
    **kwargs
) -> 'Struct'

Constructs a new Struct with all named elements.

unnamed

View source

@classmethod
unnamed(
    *args
) -> 'Struct'

Constructs a new Struct with all unnamed elements.

__eq__

View source

__eq__(
    other: object
) -> bool

Return self==value.

__getitem__

View source

__getitem__(
    key: Union[int, str, slice]
) -> _T

__iter__

View source

__iter__() -> Iterator[_T]

__len__

View source

__len__() -> int

__ne__

View source

__ne__(
    other: object
) -> bool

Return self!=value.