tff.simulation.datasets.build_dataset_mixture

Build a new dataset that probabilistically returns examples.

a The first tf.data.Dataset.
b The second tf.data.Dataset.
a_probability The float probability to select the next example from the a dataset.
seed An optional int seed for creating a tf.random.Generator. If set to None, the generator will be created via tf.random.Generator.from_non_deterministic_state. Note that setting a random seed alone is not enough to guarantee complete reproducibility of results. For example, random number generation is not guaranteed to be consistent across TensorFlow versions (see https://www.tensorflow.org/guide/versions#what_is_not_covered).

A tf.data.Dataset that returns examples from dataset a with probability a_probability, and examples form dataset b with probability (1 - a_probability). The dataset will yield the number of examples equal to the smaller of a or b.