extract participating collaborator dataset size in a federation round in the aggregator function

[dskhanirfan]

Hi
How can I extract each participating collaborator dataset size in a federation round in the aggregator function? Is ShardDescriptor and ShardDataset available in aggregator function? it available in db_iterator?

[itrushkin]

Hi @dskhanirfan! ShardDescriptor and ShardDataset are not available in AggregationFunction, but every LocalTensor that participates in aggregation has its weight equal to relative data size. For example, if federation consists of 100 samples divided equally across 10 collaborators, weights of all local_tensors in aggregation function would be equal to 0.1. And default weighted averaging (WeightedAverage in OpenFL) sums these local tensors with corresponding relative weights. Do you really need absolute values for tensor weights?

[dskhanirfan]

Hi @itrushkin I require the absolute dataset size of all participating collaborators in the aggregator function