TensorFlow 2.0 Beta is available Learn more


Quickly returns a small synthetic dataset, useful for unit tests, etc.


Defined in python/simulation/datasets/emnist/load_data.py.

Each client produced has exactly 10 examples, one of each digit. The images are derived from a fixed set of hard-coded images, and transformed using tff.simulation.datasets.emnist.infinite_emnist to produce the desired number of clients.


  • num_clients: The number of syntehtic clients to generate.


Tuple of (train, test) where the tuple elements are tff.simulation.ClientData objects matching the characteristics (other than size) of those provided by tff.simulation.datasets.emnist.load_data.