nvflare.app_common.workflows.scatter_and_gather module

class ScatterAndGather(min_clients: int = 1000, num_rounds: int = 5, start_round: int = 0, wait_time_after_min_received: int = 10, aggregator_id='aggregator', persistor_id='', shareable_generator_id='shareable_generator', train_task_name='train', train_timeout: int = 0, ignore_result_error: bool = False, allow_empty_global_weights: bool = False, task_check_period: float = 0.5, persist_every_n_rounds: int = 1, snapshot_every_n_rounds: int = 1)[source]

Bases: Controller

The controller for ScatterAndGather Workflow.

The ScatterAndGather workflow defines FederatedAveraging on all clients. The model persistor (persistor_id) is used to load the initial global model which is sent to all clients. Each client sends it’s updated weights after local training which is aggregated (aggregator_id). The shareable generator is used to convert the aggregated weights to shareable and shareable back to weight. The model_persistor also saves the model after training.

Parameters:

  • min_clients (int, optional) – The minimum number of clients responses before SAG starts to wait for wait_time_after_min_received. Note that SAG will move forward when all available clients have responded regardless of this value. Defaults to 1000.

  • num_rounds (int, optional) – The total number of training rounds. Defaults to 5.

  • start_round (int, optional) – Start round for training. Defaults to 0.

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

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

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

  • shareable_generator_id (str, optional) – ID of the shareable generator. Defaults to “shareable_generator”.

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

  • train_timeout (int, optional) – Time to wait for clients to do local training.

  • 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.

  • persist_every_n_rounds (int, optional) – persist the global model every n rounds. Defaults to 1. If n is 0 then no persist.

  • snapshot_every_n_rounds (int, optional) – persist the server state every n rounds. Defaults to 1. If n is 0 then no persist.

Raises:

  • TypeError – when any of input arguments does not have correct type

  • ValueError – when any of input arguments is out of range

control_flow(abort_signal: Signal, fl_ctx: FLContext) → None[source]

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.

get_persist_state(fl_ctx: FLContext) → dict[source]

Generate data from state to be persisted.

Parameters:

fl_ctx – FLContext

Returns:

A dict serializable persist data

handle_event(event_type: str, fl_ctx: FLContext)[source]

Handles events.

Parameters:

  • event_type (str) – event type fired by workflow.

  • fl_ctx (FLContext) – FLContext information.

process_result_of_unknown_task(client: Client, task_name, client_task_id, result: Shareable, fl_ctx: FLContext) → None[source]

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

restore(state_data: dict, fl_ctx: FLContext)[source]

Restore the state from persisted data.

Parameters:

  • state_data – serialized persist data

  • fl_ctx – FLContext

start_controller(fl_ctx: FLContext) → None[source]

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)[source]

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) –