openfl.interface.aggregation_functions.geometric_median.GeometricMedian

class openfl.interface.aggregation_functions.geometric_median.GeometricMedian(*args, **kwargs)

Bases: AggregationFunction

Geometric median aggregation.

__init__()

Initialize common AggregationFunction params.

Default: Read only access to TensorDB

Methods

__init__()	Initialize common AggregationFunction params.
call(local_tensors, *_)	Aggregate tensors.

call(local_tensors, *_)

Aggregate tensors.

Parameters:

• local_tensors (list[openfl.utilities.LocalTensor]) – List of local tensors to aggregate.

• db_iterator –

  iterator over history of all tensors. Columns: - ‘tensor_name’: name of the tensor.

  Examples for `torch.nn.Module`s: ‘conv1.weight’, ‘fc2.bias’.

  • ’round’: 0-based number of round corresponding to this

    tensor.

  • ’tags’: tuple of tensor tags. Tags that can appear:

    • ’model’ indicates that the tensor is a model parameter.

    • ’trained’ indicates that tensor is a part of a training

      result. These tensors are passed to the aggregator node after local learning.

    • ’aggregated’ indicates that tensor is a result of

      aggregation. These tensors are sent to collaborators for the next round.

    • ’delta’ indicates that value is a difference between

      rounds for a specific tensor.

    also one of the tags is a collaborator name if it corresponds to a result of a local task.

  • ’nparray’: value of the tensor.

• tensor_name – name of the tensor

• fl_round – round number

• tags – tuple of tags for this tensor

Returns:

aggregated tensor

Return type:

geometric_median (np.ndarray)