Missed TensorFlow World? Check out the recap. Learn more

Module: tff.backends.mapreduce

View source on GitHub

Utility classes and functions for integration with MapReduce-like backends.

This module contains utility components for interfacing between TFF and backend systems that offer MapReduce-like capabilities, i.e., systems that can perform parallel processing on a set of clients, and then aggregate the results of such processing on the server. Systems of this type do not support the full expressiveness of TFF, but they are common enough in practice to warrant a dedicated set of utility functions, and many examples of TFF computations, including those constructed by tff.learning, can be compiled by TFF into a form that can be deployed on such systems.


class CanonicalForm: Standardized representation of logic deployable to MapReduce-like systems.


get_canonical_form_for_iterative_process(...): Constructs tff.backends.mapreduce.CanonicalForm given iterative process.

get_iterative_process_for_canonical_form(...): Creates tff.utils.IterativeProcess from a canonical form.