View source on GitHub
|
A strategy for resolving federated types and intrinsics.
Inherits From: FederatingStrategy
tff.framework.FederatedResolvingStrategy(
executor: tff.framework.FederatingExecutor,
target_executors: Dict[str, tff.framework.Executor]
)
This strategy implements the federating_executor.FederatingStrategy
interface and provides logic for resolving federated types and federated
intrinsics on an underlying collection of target executor(s) associated with
individual placements.
This strategy only supports the placements:
Note that this strategy does not have a built-in concept of intermediate aggregation, partitioning placements, clustering clients, etc.
Args | |
|---|---|
executor
|
A federating_executor.FederatingExecutor to use to handle
unplaced types, computations, and processing.
|
target_executors
|
A dict mapping placements to a collection of target
executors associated with individual placements. The keys in this
dictionary are placement literals. The values can be either single
executors (if there only is a single participant associated with that
placement, e.g. tff.SERVER) or lists of executors.
|
Raises | |
|---|---|
TypeError
|
If target_executors is not a dict, where each key is a
placement_literals.PlacementLiteral and each value is either an
executor_base.Executor or a list of executor_base.Executors.
|
ValueError
|
If target_executors contains a
placement_literals.PlacementLiteral key that is not a kind supported
by the FederatedResolvingStrategy.
|
Methods
close
close()
Release resources associated with this strategy, if any.
compute_federated_aggregate
compute_federated_aggregate(
arg
)
Returns an embedded call for a federated aggregate.
compute_federated_apply
compute_federated_apply(
arg
)
Returns an embedded call for a federated apply.
compute_federated_broadcast
compute_federated_broadcast(
arg
)
Returns an embedded call for a federated broadcast.
compute_federated_collect
compute_federated_collect(
arg
)
Returns an embedded call for a federated collect.
compute_federated_eval_at_clients
compute_federated_eval_at_clients(
arg
)
Returns an embedded call for a federated eval at clients.
compute_federated_eval_at_server
compute_federated_eval_at_server(
arg
)
Returns an embedded call for a federated eval at server.
compute_federated_intrinsic
compute_federated_intrinsic(
uri, arg=None
)
Returns an embedded call for a federated intrinsic.
| Args | |
|---|---|
uri
|
The URI of an intrinsic to embed. |
arg
|
An optional embedded argument of the call, or None if no argument
is supplied.
|
compute_federated_map
compute_federated_map(
arg
)
Returns an embedded call for a federated map.
compute_federated_map_all_equal
compute_federated_map_all_equal(
arg
)
Returns an embedded call for a federated map all equal.
compute_federated_mean
compute_federated_mean(
arg
)
Returns an embedded call for a federated mean.
compute_federated_reduce
compute_federated_reduce(
arg
)
Returns an embedded call for a federated reduce.
compute_federated_secure_sum
compute_federated_secure_sum(
arg
)
Returns an embedded call for a federated secure sum.
compute_federated_sum
compute_federated_sum(
arg
)
Returns an embedded call for a federated sum.
compute_federated_value
compute_federated_value(
value, type_signature
)
Returns an embedded value for a federated type.
| Args | |
|---|---|
value
|
An object to embed in the executor. |
type_signature
|
A tff.Type, the type of value.
|
compute_federated_value_at_clients
compute_federated_value_at_clients(
arg
)
Returns an embedded call for a federated value at clients.
compute_federated_value_at_server
compute_federated_value_at_server(
arg
)
Returns an embedded call for a federated value at server.
compute_federated_weighted_mean
compute_federated_weighted_mean(
arg
)
Returns an embedded call for a federated weighted mean.
compute_federated_zip_at_clients
compute_federated_zip_at_clients(
arg
)
Returns an embedded call for a federated zip at clients.
compute_federated_zip_at_server
compute_federated_zip_at_server(
arg
)
Returns an embedded call for a federated zip at server.
factory
@classmethodfactory( target_executors: Dict[str,tff.framework.Executor] )
ingest_value
ingest_value(
value: Any,
type_signature: tff.Type
) -> tff.framework.ExecutorValue