A return type for a serving_input_receiver_fn.
This is for use with models that expect a single
SparseTensor as an input feature, as opposed to a dict of features.
ServingInputReceiver always returns a feature dict, even if it contains only one entry, and so can be used only with models that accept such a dict. For models that accept only a single raw feature, the
serving_input_receiver_fn provided to
Estimator.export_savedmodel() should return this
TensorServingInputReceiver instead. See: https://github.com/tensorflow/tensorflow/issues/11674
Note that the receiver_tensors and receiver_tensor_alternatives arguments will be automatically converted to the dict representation in either case, because the SavedModel format requires each input
Tensor to have a name (provided by the dict key).
The expected return values are: features: A single
SparseTensor, representing the feature to be passed to the model. receiver_tensors: a
Tensor, or a dict of string to
Tensor, specifying input nodes where this receiver expects to be fed by default. Typically, this is a single placeholder expecting serialized
tf.Example protos. receiver_tensors_alternatives: a dict of string to additional groups of receiver tensors, each of which may be a
Tensor or a dict of string to
Tensor. These named receiver tensor alternatives generate additional serving signatures, which may be used to feed inputs at different points within the input receiver subgraph. A typical usage is to allow feeding raw feature
Tensors downstream of the tf.parse_example() op. Defaults to None.
Alias for field number 0
Alias for field number 1
Alias for field number 2
@staticmethod __new__( cls, features, receiver_tensors, receiver_tensors_alternatives=None )
Create new instance of TensorServingInputReceiver(features, receiver_tensors, receiver_tensors_alternatives)
© 2018 The TensorFlow Authors. All rights reserved.
Licensed under the Creative Commons Attribution License 3.0.
Code samples licensed under the Apache 2.0 License.