Attend the Women in ML Symposium on December 7 Register now

tff.aggregators.add_measurements

Wraps AggregationFactory to report additional measurements.

The function client_measurement_fn should be a Python callable that will be called as client_measurement_fn(value) or client_measurement_fn(value, weight) depending on whether inner_agg_factory is weighted or unweighted. It must be traceable by TFF and expect tff.Value objects placed at CLIENTS as inputs, and return a collections.OrderedDict mapping string names to tensor values placed at SERVER, which will be added to the measurement dict produced by the inner_agg_factory.

Similarly, server_measurement_fn should be a Python callable that will be called as server_measurement_fn(result) where result is the result (on server) of the inner aggregation.

One or both of client_measurement_fn and server_measurement_fn must be specified.

inner_agg_factory The factory to wrap and add measurements.
client_measurement_fn A Python callable that will be called on value (and/or weight) provided to the next function to compute additional measurements of the client values/weights.
server_measurement_fn A Python callable that will be called on the result of aggregation at server to compute additional measurements of the result.

An AggregationFactory that reports additional measurements.