fedscale.cloud package

Subpackages

• fedscale.cloud.aggregation package
  • Submodules
  • fedscale.cloud.aggregation.aggregator module
  • fedscale.cloud.aggregation.android_aggregator module
  • fedscale.cloud.aggregation.optimizers module
  • Module contents
• fedscale.cloud.execution package
  • Submodules
  • fedscale.cloud.execution.client_base module
  • fedscale.cloud.execution.data_processor module
  • fedscale.cloud.execution.executor module
  • fedscale.cloud.execution.optimizers module
  • fedscale.cloud.execution.rl_client module
  • fedscale.cloud.execution.tensorflow_client module
  • fedscale.cloud.execution.torch_client module
  • Module contents

Submodules

fedscale.cloud.client_manager module

class fedscale.cloud.client_manager.ClientManager(mode, args, sample_seed=233)

Bases: object

clientOnHost(client_ids, host_id)

clientSampler(client_id)

getAllClients()

getAllClientsLength()

getAllMetrics()

getClient(client_id)

getClientLenOnHost(host_id)

getClientReward(client_id)

getClientSize(client_id)

getClientsInfo()

getCurrentclient_ids(host_id)

getDataInfo()

getFeasibleClients(cur_time)

getSampleRatio(client_id, host_id, even=False)

getUniqueId(host_id, client_id)

get_completion_time(client_id, batch_size, local_steps, upload_size, download_size)

get_median_reward()

get_score(host_id, client_id)

isClientActive(client_id, cur_time)

next_client_id_to_run(host_id)

registerClientScore(client_id, reward)

registerDuration(client_id, batch_size, local_steps, upload_size, download_size)

registerScore(client_id, reward, auxi=1.0, time_stamp=0, duration=1.0, success=True)

registerSpeed(host_id, client_id, speed)

register_client(host_id: int, client_id: int, size: int, speed: Dict[str, float], duration: float = 1) → None

Register client information to the client manager.

Parameters:

• host_id (int) – executor Id.

• client_id (int) – client Id.

• size (int) – number of samples on this client.

• speed (Dict[str, float]) – device speed (e.g., compuutation and communication).

• duration (float) – execution latency.

register_feedback(client_id: int, reward: float, auxi: float = 1.0, time_stamp: float = 0, duration: float = 1.0, success: bool = True) → None

Collect client execution feedbacks of last round.

Parameters:

• client_id (int) – client Id.

• reward (float) – execution utilities (processed feedbacks).

• auxi (float) – unprocessed feedbacks.

• time_stamp (float) – current wall clock time.

• duration (float) – system execution duration.

• success (bool) – whether this client runs successfully.

resampleClients(num_of_clients, cur_time=0)

select_participants(num_of_clients: int, cur_time: float = 0) → List[int]

Select participating clients for current execution task.

Parameters:

• num_of_clients (int) – number of participants to select.

• cur_time (float) – current wall clock time.

Returns:

indices of selected clients.

Return type:

List[int]

fedscale.cloud.commons module

fedscale.cloud.config_parser module

fedscale.cloud.fllibs module

fedscale.cloud.fllibs.import_libs()

fedscale.cloud.fllibs.init_dataset()

fedscale.cloud.fllibs.init_model()

fedscale.cloud.resource_manager module

class fedscale.cloud.resource_manager.ResourceManager(experiment_mode)

Bases: object

Schedule training tasks across GPUs/CPUs

get_next_task(client_id=None)

get_task_length() → int

Number of tasks left in the queue

Returns:

Number of tasks left in the queue

Return type:

int

has_next_task(client_id=None)

register_tasks(clientsToRun)

remove_client_task(client_id)

Module contents