fedn.network.controller package

The controller package is responsible for orchestrating the federated learning process. It’s acts as a gRPC client and sends round config tasks to the fedn.network.combiner.Combiner.

Submodules

fedn.network.controller.control module

exception fedn.network.controller.control.CombinersNotDoneException(message)

Bases: Exception

Exception class for when model is None

class fedn.network.controller.control.Control(statestore)

Bases: ControlBase

Controller, implementing the overall global training, validation and inference logic.

Parameters:

statestore (class: fedn.network.statestorebase.StateStorageBase) – A StateStorage instance.

infer_instruct(config)

Main entrypoint for executing the inference compute plan.

: param config: configuration for the inference round

inference_round(config)

Execute an inference round.

: param config: configuration for the inference round

reduce(combiners)

Combine updated models from Combiner nodes into one global model.

: param combiners: dict of combiner names(key) and model IDs(value) to reduce : type combiners: dict

round(session_config, round_id)

Execute one global round.

: param session_config: The session config. : type session_config: dict : param round_id: The round id. : type round_id: str

session(config)

Execute a new training session. A session consists of one

or several global rounds. All rounds in the same session have the same round_config.

Parameters:

config (dict) – The session config.

start_session(session_id: str, rounds: int)

exception fedn.network.controller.control.MisconfiguredStorageBackend(message)

Bases: Exception

Exception class for when storage backend is misconfigured.

Parameters:

message (str) – The exception message.

exception fedn.network.controller.control.NoModelException(message)

Bases: Exception

Exception class for when model is None

Parameters:

message (str) – The exception message.

exception fedn.network.controller.control.UnsupportedStorageBackend(message)

Bases: Exception

Exception class for when storage backend is not supported. Passes

fedn.network.controller.controlbase module

class fedn.network.controller.controlbase.ControlBase(statestore)

Bases: ABC

Base class and interface for a global controller.

Override this class to implement a global training strategy (control).

Parameters:

statestore (fedn.network.statestore.statestorebase.StateStoreBase) – The statestore object.

commit(model_id, model=None, session_id=None)

Commit a model to the global model trail. The model commited becomes the lastest consensus model.

Parameters:

• model_id (str (uuid)) – Unique identifier for the model to commit.

• model (BytesIO) – The model object to commit

• session_id (str) – Unique identifier for the session

create_round(round_data)

Initialize a new round in backend db.

create_session(config, status='Initialized')

Initialize a new session in backend db.

evaluate_round_participation_policy(compute_plan, nr_active_clients)

Evaluate policy for combiner round-participation. A combiner participates if it is responsive and reports enough active clients to participate in the round.

evaluate_round_start_policy(combiners)

Check if the policy to start a round is met.

Parameters:

combiners (list) – A list of combiners

Returns:

True if the round policy is mer, otherwise False

Return type:

bool

evaluate_round_validity_policy(round)

Check if the round is valid.

At the end of the round, before committing a model to the global model trail, we check if the round validity policy has been met. This can involve e.g. asserting that a certain number of combiners have reported in an updated model, or that criteria on model performance have been met.

Parameters:

round – The round object

Rtype round:

dict

Returns:

True if the policy is met, otherwise False

Return type:

bool

get_combiner(name)

get_compute_package(compute_package='')

Parameters:

compute_package

Returns:

get_compute_package_name()

Returns:

get_events()

Returns:

get_helper()

Get a helper instance from global config.

Returns:

Helper instance.

Return type:

fedn.utils.plugins.helperbase.HelperBase

get_latest_round()

get_latest_round_id()

get_model_info()

Returns:

get_participating_combiners(combiner_round_config)

Assemble a list of combiners able to participate in a round as descibed by combiner_round_config.

get_state()

Get the current state of the controller.

Returns:

The current state.

Return type:

fedn.network.state.ReducerState

idle()

Check if the controller is idle.

Returns:

True if idle, False otherwise.

Return type:

bool

abstract reduce(combiners)

request_model_updates(combiners)

Ask Combiner server to produce a model update.

Parameters:

combiners (tuple (combiner, comboner_round_config)) – A list of combiners

abstract round(config, round_number)

abstract session(config)

set_compute_package(filename, path)

Persist the configuration for the compute package.

set_round_config(round_id, round_config)

Upate round in backend db.

Parameters:

• round_id (str) – The round unique identifier

• round_config (dict) – The round configuration

set_round_data(round_id, round_data)

Set round data.

Parameters:

• round_id (str) – The round unique identifier

• round_data – The status

set_round_status(round_id, status)

Set the round round stats.

Parameters:

• round_id (str) – The round unique identifier

• status (str) – The status

set_session_status(session_id, status)

Set the round round stats.

Parameters:

• round_id (str) – The round unique identifier

• status (str) – The status

state()

Get the current state of the controller.

Returns:

The state

Rype:

str

exception fedn.network.controller.controlbase.MisconfiguredHelper

Bases: Exception

exception fedn.network.controller.controlbase.MisconfiguredStorageBackend

Bases: Exception

exception fedn.network.controller.controlbase.UnsupportedStorageBackend

Bases: Exception