fl4health.strategies.aggregate_utils module

aggregate_losses(results, weighted=True)

Aggregate evaluation results obtained from multiple clients.

Parameters:

• results (list[tuple[int, float]]) – A list of sample counts and loss values (in that order). The sample counts from each client (training or validation samples where appropriate) are used if weighted averaging is requested.

• weighted (bool, optional) – Whether or not the aggregation is a weighted average (by the sample counts provided in the tuple) or a uniform average. Defaults to True.

Returns:

the weighted or unweighted average of the loss values in the results list.

Return type:

float

aggregate_results(results, weighted=True)

Compute weighted or unweighted average.

Parameters:

• results (list[tuple[NDArrays, int]]) – This is a set of NDArrays (list of numpy arrays) and the number of relevant samples from each client (training or validation samples where appropriate). These are to be aggregated together in a weighted or unweighted average. The NDArrays most often represent model states.

• weighted (bool, optional) – Whether or not the aggregation is a weighted average (by the sample counts provided in the tuple) or a uniform average. Defaults to True.

Returns:

Aggregated numpy arrays by the desired averaging.

Return type:

NDArrays