client#

Package Contents#

ClientManager

Base class for ClientManager.

ClientActiveManager

Active communication NetworkManager for client in asynchronous FL pattern.

ClientPassiveManager

Passive communication NetworkManager for client in synchronous FL pattern.

ScaleClientPassiveManager

Special client manager for SerialTrainer.

SerialTrainer

Base class. Train multiple clients in sequence with a single process.

SubsetSerialTrainer

Train multiple clients in a single process.

ORDINARY_TRAINER

SERIAL_TRAINER

ORDINARY_TRAINER = 0#
SERIAL_TRAINER = 1#
class ClientManager(network, trainer)#

Bases: fedlab.core.network_manager.NetworkManager

Base class for ClientManager.

ClientManager defines client activation for different communication stages.

Parameters

  • network (DistNetwork) – Network configuration.

  • trainer (ClientTrainer) – Subclass of ClientTrainer. Provides train() and model. Define local client training procedure.

setup(self)#

Initialization stage.

ClientManager reports number of clients simulated by current client process.

class ClientActiveManager(network, trainer, logger=Logger())#

Bases: ClientManager

Active communication NetworkManager for client in asynchronous FL pattern.

Parameters

  • network (DistNetwork) – network configuration.

  • trainer (ClientTrainer) – Subclass of ClientTrainer. Provides train() and model. Define local client training procedure.

  • logger (Logger, optional) – object of Logger.

main_loop(self)#

Actions to perform on receiving new message, including local training

  1. client requests data from server (ACTIVE)

  2. after receiving data, client will train local model

  3. client will synchronize with server actively

synchronize(self)#

Synchronize local model with server

class ClientPassiveManager(network, trainer, logger=Logger())#

Bases: ClientManager

Passive communication NetworkManager for client in synchronous FL pattern.

Parameters

  • network (DistNetwork) – network configuration.

  • trainer (ClientTrainer) – Subclass of ClientTrainer. Provides train() and model. Define local client training procedure.

  • logger (Logger) – object of Logger.

main_loop(self)#

Actions to perform when receiving a new message, including local training.

Main procedure of each client:

  1. client waits for data from server (PASSIVELY).

  2. after receiving data, client start local model training procedure.

  3. client synchronizes with server actively.

synchronize(self)#

Synchronize local model with server

class ScaleClientPassiveManager(network, trainer, logger=Logger())#

Bases: fedlab.core.client.manager.ClientPassiveManager

Special client manager for SerialTrainer.

We modify the communication agreements to create a mapping from client id to process rank. Thus, ScaleClientPassiveManager is able to simulate multiple clients in sequence.

Parameters

  • network (DistNetwork) – Distributed network to use.

  • trainer (ClientTrainer) – Subclass of ClientTrainer, providing train() and model. For more client simulation with single process, you are supposed to use SerialTrainer here.

  • logger (Logger) – object of Logger.

main_loop(self)#

Actions to perform when receiving new message, including local training.

synchronize(self)#

Synchronize local model with server actively

class SerialTrainer(model, client_num, aggregator=None, cuda=False, logger=Logger())#

Bases: fedlab.core.client.trainer.ClientTrainer

Base class. Train multiple clients in sequence with a single process.

Parameters

  • model (torch.nn.Module) – Model used in this federation.

  • client_num (int) – Number of clients in current trainer.

  • aggregator (Aggregators, callable, optional) – Function to perform aggregation on a list of serialized model parameters.

  • cuda (bool) – Use GPUs or not. Default: False.

  • logger (Logger, optional) – object of Logger.

abstract _train_alone(self, model_parameters, train_loader)#

Train local model with model_parameters on train_loader.

Parameters
abstract _get_dataloader(self, client_id)#

Get DataLoader for client_id.

train(self, model_parameters, id_list, aggregate=False)#

Train local model with different dataset according to client id in id_list.

Parameters

  • model_parameters (torch.Tensor) – Serialized model parameters.

  • id_list (list[int]) – Client id in this training serial.

  • aggregate (bool) – Whether to perform partial aggregation on this group of clients’ local models at the end of each local training round.

Note

Normally, aggregation is performed by server, while we provide aggregate option here to perform partial aggregation on current client group. This partial aggregation can reduce the aggregation workload of server.

Returns

Serialized model parameters / list of model parameters.

class SubsetSerialTrainer(model, dataset, data_slices, aggregator=None, logger=Logger(), cuda=False, args=None)#

Bases: SerialTrainer

Train multiple clients in a single process.

Customize _get_dataloader() or _train_alone() for specific algorithm design in clients.

Parameters

  • model (torch.nn.Module) – Model used in this federation.

  • dataset (torch.utils.data.Dataset) – Local dataset for this group of clients.

  • data_slices (list[list]) – subset of indices of dataset.

  • aggregator (Aggregators, callable, optional) – Function to perform aggregation on a list of model parameters.

  • logger (Logger, optional) – object of Logger.

  • cuda (bool) – Use GPUs or not. Default: False.

  • args (dict, optional) – Uncertain variables.

Note

len(data_slices) == client_num, that is, each sub-index of dataset corresponds to a client’s local dataset one-by-one.

_get_dataloader(self, client_id)#

Return a training dataloader used in train() for client with id

Parameters

client_id (int) – client_id of client to generate dataloader

Note

client_id here is not equal to client_id in global FL setting. It is the index of client in current SerialTrainer.

Returns

DataLoader for specific client’s sub-dataset

_train_alone(self, model_parameters, train_loader)#

Single round of local training for one client.

Note

Overwrite this method to customize the PyTorch training pipeline.

Parameters