nvflare.app_common.workflows.model_controller module

class ModelController(persistor_id='persistor', ignore_result_error: bool = False, allow_empty_global_weights: bool = False, task_check_period: float = 0.5)

Bases: Controller, FLComponentWrapper, ABC

FLModel based controller.

Parameters:

• persistor_id (str, optional) – ID of the persistor component. Defaults to “persistor”.

• ignore_result_error (bool, optional) – whether this controller can proceed if client result has errors. Defaults to False.

• allow_empty_global_weights (bool, optional) – whether to allow empty global weights. Some pipelines can have empty global weights at first round, such that clients start training from scratch without any global info. Defaults to False.

• task_check_period (float, optional) – interval for checking status of tasks. Defaults to 0.5.

broadcast_model(task_name: str = 'train', data: FLModel | None = None, targets: List[Client] | List[str] | None = None, timeout: int = 0, wait_time_after_min_received: int = 10, blocking: bool = True, callback: Callable[[FLModel], None] | None = None) → List

Send a task with data to a list of targets.

Parameters:

• task_name (str, optional) – name of the task. Defaults to “train”.

• data (FLModel, optional) – FLModel to be sent to clients. If no data is given, send empty FLModel.

• targets (List[str], optional) – the list of target client names or None (all clients). Defaults to None.

• timeout (int, optional) – time to wait for clients to perform task. Defaults to 0, i.e., never time out.

• wait_time_after_min_received (int, optional) – time to wait after minimum number of clients responses has been received. Defaults to 10.

• blocking (bool, optional) – whether to block to wait for task result. Defaults to True.

• callback (Callable[[FLModel], None], optional) – callback when a result is received, only called when blocking=False. Defaults to None.

Returns:

List[FLModel] if blocking=True else None

control_flow(abort_signal: Signal, fl_ctx: FLContext) → None

This is the control logic for the RUN.

NOTE: this is running in a separate thread, and its life is the duration of the RUN.

Parameters:

• fl_ctx – the FL context

• abort_signal – the abort signal. If triggered, this method stops waiting and returns to the caller.

load_model()

process_result_of_unknown_task(client: Client, task_name: str, client_task_id: str, result: Shareable, fl_ctx: FLContext) → None

Process result when no task is found for it.

This is called when a result submission is received from a client, but no standing task can be found for it (from the task queue)

This could happen when: - the client’s submission is too late - the task is already completed - the Controller lost the task, e.g. the Server is restarted

Parameters:

• client – the client that the result comes from

• task_name – the name of the task

• client_task_id – ID of the task

• result – the result from the client

• fl_ctx – the FL context that comes with the client’s submission

abstract run()

Main run routine called by the Controller’s control_flow to execute the workflow.

Returns: None.

save_model(model)

start_controller(fl_ctx: FLContext) → None

Starts the controller.

This method is called at the beginning of the RUN.

Parameters:

• fl_ctx – the FL context. You can use this context to access services provided by the

• example (framework. For) –

• your (you can get Command Register from it and register) –

• modules. (admin command) –

stop_controller(fl_ctx: FLContext)

Stops the controller.

This method is called right before the RUN is ended.

Parameters:

• fl_ctx – the FL context. You can use this context to access services provided by the

• example (framework. For) –

• your (you can get Command Register from it and unregister) –

• modules. (admin command) –