tf.contrib.predictor.from_saved_model

View source on GitHub

Constructs a Predictor from a SavedModel on disk.

tf.contrib.predictor.from_saved_model(
    export_dir,
    signature_def_key=None,
    signature_def=None,
    input_names=None,
    output_names=None,
    tags=None,
    graph=None,
    config=None
)

Args:

• export_dir: a path to a directory containing a SavedModel.
• signature_def_key: Optional string specifying the signature to use. If None, then DEFAULT_SERVING_SIGNATURE_DEF_KEY is used. Only one of signature_def_key and signature_def
• signature_def: A SignatureDef proto specifying the inputs and outputs for prediction. Only one of signature_def_key and signature_def should be specified. input_names: A dictionary mapping strings to Tensors in the SavedModel that represent the input. The keys can be any string of the user's choosing. output_names: A dictionary mapping strings to Tensors in the SavedModel that represent the output. The keys can be any string of the user's choosing.
• tags: Optional. Tags that will be used to retrieve the correct SignatureDef. Defaults to DEFAULT_TAGS.
• graph: Optional. The Tensorflow graph in which prediction should be done.
• config: ConfigProto proto used to configure the session.

Returns:

An initialized Predictor.

Raises:

• ValueError: More than one of signature_def_key and signature_def is specified.