The Trainer TFX Pipeline Component

The Trainer TFX pipeline component trains a TensorFlow model.

Trainer takes:

  • tf.Examples used for training and eval.
  • A user provided module file that defines the trainer logic.
  • A data schema create by a SchemaGen pipeline component and optionally altered by the developer.
  • Proto definition of train args and eval args.
  • Optional transform graph produced by upstream Transform component.
  • Optional base models used for scenarios such as warmstart.

Trainer emits: A SavedModel and an EvalSavedModel

Trainer and TensorFlow

Trainer makes extensive use of the Python TensorFlow API for training models.

Configuring a Trainer Component

A Trainer pipeline component is typically very easy to develop and requires little customization, since all of the work is done by the Trainer TFX component. Your TensorFlow modeling code however may be arbitrarily complex.

Typical code looks like this:

from tfx import components

...

trainer = Trainer(
      module_file=module_file,
      transformed_examples=transform.outputs['transformed_examples'],
      schema=infer_schema.outputs['schema'],
      base_models=latest_model_resolver.outputs['latest_model'],
      transform_graph=transform.outputs['transform_graph'],
      train_args=trainer_pb2.TrainArgs(num_steps=10000),
      eval_args=trainer_pb2.EvalArgs(num_steps=5000))

Trainer invokes a training module, which is specified in the module_file parameter. A typical training module looks like this:

# TFX will call this function
def trainer_fn(trainer_fn_args, schema):
  """Build the estimator using the high level API.

  Args:
    trainer_fn_args: Holds args used to train the model as name/value pairs.
    schema: Holds the schema of the training examples.

  Returns:
    A dict of the following:
      - estimator: The estimator that will be used for training and eval.
      - train_spec: Spec for training.
      - eval_spec: Spec for eval.
      - eval_input_receiver_fn: Input function for eval.
  """
  # Number of nodes in the first layer of the DNN
  first_dnn_layer_size = 100
  num_dnn_layers = 4
  dnn_decay_factor = 0.7

  train_batch_size = 40
  eval_batch_size = 40

  tf_transform_output = tft.TFTransformOutput(trainer_fn_args.transform_output)

  train_input_fn = lambda: _input_fn(  # pylint: disable=g-long-lambda
      trainer_fn_args.train_files,
      tf_transform_output,
      batch_size=train_batch_size)

  eval_input_fn = lambda: _input_fn(  # pylint: disable=g-long-lambda
      trainer_fn_args.eval_files,
      tf_transform_output,
      batch_size=eval_batch_size)

  train_spec = tf.estimator.TrainSpec(  # pylint: disable=g-long-lambda
      train_input_fn,
      max_steps=trainer_fn_args.train_steps)

  serving_receiver_fn = lambda: _example_serving_receiver_fn(  # pylint: disable=g-long-lambda
      tf_transform_output, schema)

  exporter = tf.estimator.FinalExporter('chicago-taxi', serving_receiver_fn)
  eval_spec = tf.estimator.EvalSpec(
      eval_input_fn,
      steps=trainer_fn_args.eval_steps,
      exporters=[exporter],
      name='chicago-taxi-eval')

  run_config = tf.estimator.RunConfig(
      save_checkpoints_steps=999, keep_checkpoint_max=1)

  run_config = run_config.replace(model_dir=trainer_fn_args.serving_model_dir)
  warm_start_from = trainer_fn_args.base_models[
      0] if trainer_fn_args.base_models else None

  estimator = _build_estimator(
      # Construct layers sizes with exponetial decay
      hidden_units=[
          max(2, int(first_dnn_layer_size * dnn_decay_factor**i))
          for i in range(num_dnn_layers)
      ],
      config=run_config,
      warm_start_from=warm_start_from)

  # Create an input receiver for TFMA processing
  receiver_fn = lambda: _eval_input_receiver_fn(  # pylint: disable=g-long-lambda
      tf_transform_output, schema)

  return {
      'estimator': estimator,
      'train_spec': train_spec,
      'eval_spec': eval_spec,
      'eval_input_receiver_fn': receiver_fn
  }