TFX의 TensorFlow 2.x

# Build a Keras model.
def _keras_model_builder():
  """Creates a Keras model."""
  ...

  model = tf.keras.Model(inputs=inputs, outputs=output)
  model.compile()

  return model


# Write a typical trainer function
def trainer_fn(trainer_fn_args, schema):
  """Build the estimator, using model_to_estimator."""
  ...

  # Model to estimator
  estimator = tf.keras.estimator.model_to_estimator(
      keras_model=_keras_model_builder(), config=run_config)

  return {
      'estimator': estimator,
      ...
  }

def run_fn(fn_args: TrainerFnArgs):
  """Train the model based on given args.

  Args:
    fn_args: Holds args used to train the model as name/value pairs.
  """
  tf_transform_output = tft.TFTransformOutput(fn_args.transform_output)

  # Train and eval files contains transformed examples.
  # _input_fn read dataset based on transformed schema from tft.
  train_dataset = _input_fn(fn_args.train_files, fn_args.data_accessor,
                            tf_transform_output.transformed_metadata.schema)
  eval_dataset = _input_fn(fn_args.eval_files, fn_args.data_accessor,
                           tf_transform_output.transformed_metadata.schema)

  model = _build_keras_model()

  model.fit(
      train_dataset,
      steps_per_epoch=fn_args.train_steps,
      validation_data=eval_dataset,
      validation_steps=fn_args.eval_steps)

  signatures = {
      'serving_default':
          _get_serve_tf_examples_fn(model,
                                    tf_transform_output).get_concrete_function(
                                        tf.TensorSpec(
                                            shape=[None],
                                            dtype=tf.string,
                                            name='examples')),
  }
  model.save(fn_args.serving_model_dir, save_format='tf', signatures=signatures)

def _get_serve_tf_examples_fn(model, tf_transform_output):
  """Returns a function that parses a serialized tf.Example."""

  # the layer is added as an attribute to the model in order to make sure that
  # the model assets are handled correctly when exporting.
  model.tft_layer = tf_transform_output.transform_features_layer()

  @tf.function
  def serve_tf_examples_fn(serialized_tf_examples):
    """Returns the output to be used in the serving signature."""
    feature_spec = tf_transform_output.raw_feature_spec()
    feature_spec.pop(_LABEL_KEY)
    parsed_features = tf.io.parse_example(serialized_tf_examples, feature_spec)

    transformed_features = model.tft_layer(parsed_features)

    return model(transformed_features)

  return serve_tf_examples_fn

def _get_serve_tf_examples_fn(model, schema):

  @tf.function
  def serve_tf_examples_fn(serialized_tf_examples):
    feature_spec = _get_raw_feature_spec(schema)
    feature_spec.pop(_LABEL_KEY)
    parsed_features = tf.io.parse_example(serialized_tf_examples, feature_spec)
    return model(parsed_features)

  return serve_tf_examples_fn


def run_fn(fn_args: TrainerFnArgs):
  schema = io_utils.parse_pbtxt_file(fn_args.schema_file, schema_pb2.Schema())

  # Train and eval files contains raw examples.
  # _input_fn reads the dataset based on raw data schema.
  train_dataset = _input_fn(fn_args.train_files, fn_args.data_accessor, schema)
  eval_dataset = _input_fn(fn_args.eval_files, fn_args.data_accessor, schema)

  model = _build_keras_model()

  model.fit(
      train_dataset,
      steps_per_epoch=fn_args.train_steps,
      validation_data=eval_dataset,
      validation_steps=fn_args.eval_steps)

  signatures = {
      'serving_default':
          _get_serve_tf_examples_fn(model, schema).get_concrete_function(
              tf.TensorSpec(shape=[None], dtype=tf.string, name='examples')),
  }
  model.save(fn_args.serving_model_dir, save_format='tf', signatures=signatures)

  mirrored_strategy = tf.distribute.MirroredStrategy()
  with mirrored_strategy.scope():
    model = _build_keras_model()

  # Rest of the code can be unchanged.
  model.fit(...)

import logging
logging.getLogger("tensorflow").setLevel(logging.INFO)