openfl.component.collaborator

class openfl.component.collaborator.Collaborator(collaborator_name, aggregator_uuid, federation_uuid, client, task_runner, task_config, opt_treatment='RESET', device_assignment_policy='CPU_ONLY', delta_updates=False, compression_pipeline=None, db_store_rounds=1, **kwargs)

The Collaborator object class.

Class Attributes:

  • collaborator_name (str) – The common name for the collaborator.

  • aggregator_uuid (str) – The unique id for the client.

  • federation_uuid (str) – The unique id for the federation.

  • client (object) – The client object.

  • task_runner (object) – The task runner object.

  • task_config (dict) – The task configuration.

  • opt_treatment (str) – The optimizer state treatment.

  • device_assignment_policy (str) – The device assignment policy.

  • delta_updates (bool) – If True, only model delta gets sent. If False, whole model gets sent to collaborator.

  • compression_pipeline (object) – The compression pipeline.

  • db_store_rounds (int) – The number of rounds to store in the database.

  • single_col_cert_common_name (str) – The common name for the single column certificate.

Note

* - Plan setting.

do_task(task, round_number)

Perform the specified task.

Parameters:

  • task (list_of_str) – List of tasks.

  • round_number (int) – Actual round number.

get_aggregated_tensor_from_aggregator(tensor_key, require_lossless=False)

Return the decompressed tensor associated with the requested tensor key.

If the key requests a compressed tensor (in the tag), the tensor will be decompressed before returning. If the key specifies an uncompressed tensor (or just omits a compressed tag), the decompression operation will be skipped.

Parameters:

  • tensor_key (namedtuple) – The requested tensor.

  • require_lossless (bool) – Should compression of the tensor be allowed in flight? For the initial model, it may affect convergence to apply lossy compression. And metrics shouldn’t be compressed either.

Returns:

nparray – The decompressed tensor associated with the requested tensor key.

get_data_for_tensorkey(tensor_key)

Resolve the tensor corresponding to the requested tensorkey.

Parameters:

  • tensor_key (namedtuple) – Tensorkey that will be resolved locally or

  • tensors. (remotely. May be the product of other)

Returns:

nparray – The decompressed tensor associated with the requested tensor key.

get_numpy_dict_for_tensorkeys(tensor_keys)

Get tensor dictionary for specified tensorkey set.

Parameters:

tensor_keys (namedtuple) – Tensorkeys that will be resolved locally or remotely. May be the product of other tensors.

get_tasks()

Get tasks from the aggregator.

Returns:

  • tasks (list_of_str) – List of tasks.

  • round_number (int) – Actual round number.

  • sleep_time (int) – Sleep time.

  • time_to_quit (bool) – bool value for quit.

named_tensor_to_nparray(named_tensor)

Convert named tensor to a numpy array.

Parameters:

named_tensor (protobuf) – The tensor to convert to nparray.

Returns:

decompressed_nparray (nparray) – The nparray converted.

nparray_to_named_tensor(tensor_key, nparray)

Construct the NamedTensor Protobuf.

Includes logic to create delta, compress tensors with the TensorCodec, etc.

Parameters:

  • tensor_key (namedtuple) – Tensorkey that will be resolved locally or remotely. May be the product of other tensors.

  • nparray – The decompressed tensor associated with the requested tensor key.

Returns:

named_tensor (protobuf) – The tensor constructed from the nparray.

run()

Run the collaborator.

run_simulation()

Specific function for the simulation.

After the tasks have been performed for a roundquit, and then the collaborator object will be reinitialized after the next round.

send_task_results(tensor_dict, round_number, task_name)

Send task results to the aggregator.

Parameters:

  • tensor_dict (dict) – Tensor dictionary.

  • round_number (int) – Actual round number.

  • task_name (string) – Task name.

set_available_devices(cuda: Tuple[str] = ())

Set available CUDA devices.

Parameters:

cuda (Tuple[str]) – Tuple containing string indices of available CUDA devices, (‘1’, ‘3’).

collaborator

Collaborator module.