lib.sedna.algorithms.aggregation

Submodules

Package Contents

Classes

FedAvg

Federated averaging algorithm

MistNet

Abstract class of aggregator

AggClient

Client that interacts with cloud aggregator

FedAvgV2

Abstract class of aggregator
class lib.sedna.algorithms.aggregation.FedAvg[source]

Bases: BaseAggregation, abc.ABC

Federated averaging algorithm

aggregate(clients: List[AggClient])[source]

Calculate the average weight according to the number of samples

Parameters:

clients (List) – All clients in federated learning job

Returns:

update_weights – final weights use to update model layer

Return type:

Array-like

class lib.sedna.algorithms.aggregation.MistNet(cut_layer, epsilon=100)[source]

Bases: BaseAggregation, abc.ABC

Abstract class of aggregator

aggregate(clients: List[AggClient])[source]

Some algorithms can be aggregated in sequence, but some can be calculated only after all aggregated data is uploaded. therefore, this abstractmethod should consider that all weights are uploaded.

Parameters:

clients (List) – All clients in federated learning job

Returns:

final weights use to update model layer

Return type:

Array-like

class lib.sedna.algorithms.aggregation.AggClient[source]

Client that interacts with cloud aggregator

Parameters:

  • num_samples (int) – number of samples for the current weights

  • weights (List) – weights of the layer as a list of number-like array, such as [[0, 0, 0, 0], [0, 0, 0, 0] … ]

num_samples: int
weights: List
class lib.sedna.algorithms.aggregation.FedAvgV2[source]

Bases: BaseAggregation, abc.ABC

Abstract class of aggregator

aggregate(clients: List[AggClient])[source]

Some algorithms can be aggregated in sequence, but some can be calculated only after all aggregated data is uploaded. therefore, this abstractmethod should consider that all weights are uploaded.

Parameters:

clients (List) – All clients in federated learning job

Returns:

final weights use to update model layer

Return type:

Array-like