syft.federated.train_config

Module Contents

class syft.federated.train_config.TrainConfig(model: torch.jit.ScriptModule, loss_fn: torch.jit.ScriptModule, owner: AbstractWorker = None, batch_size: int = 32, epochs: int = 1, optimizer: str = 'SGD', optimizer_args: dict = {'lr': 0.1}, id: Union[int, str] = None, max_nr_batches: int = -1, shuffle: bool = True, loss_fn_id: int = None, model_id: int = None)

TrainConfig abstraction.

A wrapper object that contains all that is needed to run a training loop remotely on a federated learning setup.

__str__(self)

Returns the string representation of a TrainConfig.

_wrap_and_send_obj(self, obj, location)

Wrappers object and send it to location.

send(self, location: BaseWorker)

Gets the pointer to a new remote object.

One of the most commonly used methods in PySyft, this method serializes the object upon which it is called (self), sends the object to a remote worker, creates a pointer to that worker, and then returns that pointer from this function.

Parameters

location – The BaseWorker object which you want to send this object to. Note that this is never actually the BaseWorker but instead a class which instantiates the BaseWorker abstraction.

Returns

A weakref instance.

get(self, location)

get_model(self)

get_loss_fn(self)

static simplify(worker: AbstractWorker, train_config: TrainConfig)

Takes the attributes of a TrainConfig and saves them in a tuple.

Attention: this function does not serialize the model and loss_fn attributes of a TrainConfig instance, these are serialized and sent before. TrainConfig keeps a reference to the sent objects using _model_id and _loss_fn_id which are serialized here.

Parameters

• worker – the worker doing the serialization

• train_config – a TrainConfig object

Returns

a tuple holding the unique attributes of the TrainConfig object

Return type

tuple

static detail(worker: AbstractWorker, train_config_tuple: tuple)

This function reconstructs a TrainConfig object given it’s attributes in the form of a tuple.

Parameters

• worker – the worker doing the deserialization

• train_config_tuple – a tuple holding the attributes of the TrainConfig

Returns

A TrainConfig object

Return type

train_config