Attend the Women in ML Symposium on December 7 Register now


Stay organized with collections Save and categorize content based on your preferences.

Creates a DataHandler, providing standardized access to a Dataset.

See DataHandler for the list and definition of the arguments. See the implementation of, evaluate(), or predict() methods for complete usage examples. As a rule of tumb, get_data_handler() accepts the same inputs as the x argument of


  def step(iterator):
    data = next(iterator)
    # result <= Do something with data
    return result
  tf_step = tf.function(step, reduce_retracing=True)

  # Assume x is a Dataset.
  data_handler = data_adapter.get_data_handler(x=x)
  # Epoch iteration
  for epo_idx, iterator in data_handler.enumerate_epochs():
      # Stop on dataset exhaustion.
      with data_handler.catch_stop_iteration():
        for step in data_handler.steps(): # Step iteration
            step_result = step(iterator)

*args Arguments passed to the DataHandler constructor.
**kwargs Arguments passed to the DataHandler constructor.

A DataHandler object. If the model's cluster coordinate is set (e.g. the model was defined under a parameter-server strategy), returns a _ClusterCoordinatorDataHandler.