tff.program.DatasetDataSourceIterator

View source on GitHub

A tff.program.FederatedDataSourceIterator backed by tf.data.Datasets.

Inherits From: FederatedDataSourceIterator

tff.program.DatasetDataSourceIterator(
    datasets: Sequence[tf.data.Dataset],
    federated_type: tff.types.FederatedType
)

A tff.program.FederatedDataSourceIterator backed by a sequence of tf.data.Dataset's, onetf.data.Dataset' per client. It selects datasources uniformly at random, with replacement over successive calls of select() but without replacement within a single call of select().

Args
datasets	A sequence of tf.data.Dataset's to use to yield the data from this data source. </td> </tr><tr> <td>federated_type<a id="federated_type"></a> </td> <td> The type of the data returned by callingselect` on an iterator.

Raises
ValueError	If datasets is empty or if each tf.data.Dataset in datasets does not have the same type specification.

Attributes
federated_type	The type of the data returned by calling select.

Methods

from_bytes

View source

@classmethod
from_bytes(
    buffer: bytes
) -> 'DatasetDataSourceIterator'

Deserializes the object from bytes.

select

View source

select(
    num_clients: Optional[int] = None
) -> object

Returns a new selection of data from this iterator.

Args
num_clients	A number of clients to use when selecting data. Must be a positive integer and less than the number of datasets.

Raises
ValueError	If num_clients is not a positive integer or if num_clients is not less than the number of datasets.

to_bytes

View source

to_bytes() -> bytes

Serializes the object to bytes.

__eq__

View source

__eq__(
    other: object
) -> bool

Return self==value.