tf.contrib.estimator.export_saved_model_for_mode

tf.contrib.estimator.export_saved_model_for_mode(
    estimator,
    export_dir_base,
    input_receiver_fn,
    assets_extra=None,
    as_text=False,
    checkpoint_path=None,
    strip_default_attrs=False,
    mode=model_fn_lib.ModeKeys.PREDICT
)

Defined in tensorflow/contrib/estimator/python/estimator/export.py.

Exports a single train/eval/predict graph as a SavedModel.

For a detailed guide, see Using SavedModel with Estimators.

Sample usage:

classifier = tf.estimator.LinearClassifier(
    feature_columns=[age, language])
classifier.train(input_fn=input_fn, steps=1000)

feature_spec = {
    'age': tf.placeholder(dtype=tf.int64),
    'language': array_ops.placeholder(dtype=tf.string)
}
label_spec = tf.placeholder(dtype=dtypes.int64)

train_rcvr_fn = tf.contrib.estimator.build_raw_supervised_input_receiver_fn(
    feature_spec, label_spec)

export_dir = tf.contrib.estimator.export_saved_model_for_mode(
    classifier,
    export_dir_base='my_model/',
    input_receiver_fn=train_rcvr_fn,
    mode=model_fn_lib.ModeKeys.TRAIN)

# export_dir is a timestamped directory with the SavedModel, which
# can be used for serving, analysis with TFMA, or directly loaded in.
with ops.Graph().as_default() as graph:
  with session.Session(graph=graph) as sess:
    loader.load(sess, [tag_constants.TRAINING], export_dir)
    weights = graph.get_tensor_by_name(''linear/linear_model/age/weights')
    ...

This method is a wrapper for _export_all_saved_models, and wraps a raw input_receiver_fn in a dictionary to pass in to that function. See _export_all_saved_models for full docs.

See tf.contrib.estimator.export_saved_model_for_mode for the currently exposed version of this function.

Args:

  • estimator: an instance of tf.estimator.Estimator
  • export_dir_base: A string containing a directory in which to create timestamped subdirectories containing exported SavedModels.
  • input_receiver_fn: a function that takes no argument and returns the appropriate subclass of InputReceiver.
  • assets_extra: A dict specifying how to populate the assets.extra directory within the exported SavedModel, or None if no extra assets are needed.
  • as_text: whether to write the SavedModel proto in text format.
  • checkpoint_path: The checkpoint path to export. If None (the default), the most recent checkpoint found within the model directory is chosen.
  • strip_default_attrs: Boolean. If True, default-valued attributes will be removed from the NodeDefs. For a detailed guide, see Stripping Default-Valued Attributes.
  • mode: tf.estimator.ModeKeys value indicating with mode will be exported.

Returns:

The string path to the exported directory.

Raises:

  • ValueError: if input_receiver_fn is None, no export_outputs are provided, or no checkpoint can be found.