def _v1_multi_metagraph_saved_model(self):
export_graph = ops.Graph()
with export_graph.as_default():
start = array_ops.placeholder(
shape=[None], dtype=dtypes.float32, name="start")
v = resource_variable_ops.ResourceVariable(21.)
first_output = array_ops.identity(start * v, name="first_output")
second_output = array_ops.identity(v, name="second_output")
with session_lib.Session() as session:
session.run(v.initializer)
path = os.path.join(self.get_temp_dir(), "saved_model", str(ops.uid()))
builder = builder_impl.SavedModelBuilder(path)
builder.add_meta_graph_and_variables(
session, tags=["first"],
signature_def_map={
"first_key": signature_def_utils.build_signature_def(
{"first_start": utils_impl.build_tensor_info(start)},
{"first_output": utils_impl.build_tensor_info(
first_output)})})
builder.add_meta_graph(
tags=["second"],
signature_def_map={
"second_key": signature_def_utils.build_signature_def(
{"second_start": utils_impl.build_tensor_info(start)},
{"second_output": utils_impl.build_tensor_info(
second_output)})})
builder.save()
return path
def _v1_multi_input_saved_model(self):
export_graph = ops.Graph()
with export_graph.as_default():
input1 = array_ops.placeholder(shape=[None],
dtype=dtypes.float32,
name="input1")
input2 = array_ops.placeholder(shape=[None],
dtype=dtypes.float32,
name="input2")
v = resource_variable_ops.ResourceVariable(21.)
output = array_ops.identity(input1 * v + input2, name="output")
with session_lib.Session() as session:
session.run(v.initializer)
path = os.path.join(self.get_temp_dir(), "saved_model",
str(ops.uid()))
builder = builder_impl.SavedModelBuilder(path)
builder.add_meta_graph_and_variables(
session,
tags=[tag_constants.SERVING],
signature_def_map={
"serving_default":
signature_def_utils.build_signature_def(
{
"input1": utils_impl.build_tensor_info(input1),
"input2": utils_impl.build_tensor_info(input2)
},
{"output": utils_impl.build_tensor_info(output)})
})
builder.save()
return path
def _v1_multi_metagraph_saved_model(self):
export_graph = ops.Graph()
with export_graph.as_default():
start = array_ops.placeholder(
shape=[None], dtype=dtypes.float32, name="start")
v = resource_variable_ops.ResourceVariable(21.)
first_output = array_ops.identity(start * v, name="first_output")
second_output = array_ops.identity(v, name="second_output")
with session_lib.Session() as session:
session.run(v.initializer)
path = os.path.join(self.get_temp_dir(), "saved_model", str(ops.uid()))
builder = builder_impl.SavedModelBuilder(path)
builder.add_meta_graph_and_variables(
session, tags=["first"],
signature_def_map={
"first_key": signature_def_utils.build_signature_def(
{"first_start": utils_impl.build_tensor_info(start)},
{"first_output": utils_impl.build_tensor_info(
first_output)})})
builder.add_meta_graph(
tags=["second"],
signature_def_map={
"second_key": signature_def_utils.build_signature_def(
{"second_start": utils_impl.build_tensor_info(start)},
{"second_output": utils_impl.build_tensor_info(
second_output)})})
builder.save()
return path
def predict_signature_def(inputs, outputs):
"""Creates prediction signature from given inputs and outputs.
This function produces signatures intended for use with the TensorFlow Serving
Predict API (tensorflow_serving/apis/prediction_service.proto). This API
imposes no constraints on the input and output types.
Args:
inputs: dict of string to `Tensor`.
outputs: dict of string to `Tensor`.
Returns:
A prediction-flavored signature_def.
Raises:
ValueError: If inputs or outputs is `None`.
"""
if inputs is None or not inputs:
raise ValueError('Prediction inputs cannot be None or empty.')
if outputs is None or not outputs:
raise ValueError('Prediction outputs cannot be None or empty.')
signature_inputs = {key: utils.build_tensor_info(tensor)
for key, tensor in inputs.items()}
signature_outputs = {key: utils.build_tensor_info(tensor)
for key, tensor in outputs.items()}
signature_def = build_signature_def(
signature_inputs, signature_outputs,
signature_constants.PREDICT_METHOD_NAME)
return signature_def
def predict_signature_def(inputs, outputs):
"""Creates prediction signature from given inputs and outputs.
This function produces signatures intended for use with the TensorFlow Serving
Predict API (tensorflow_serving/apis/prediction_service.proto). This API
imposes no constraints on the input and output types.
Args:
inputs: dict of string to `Tensor`.
outputs: dict of string to `Tensor`.
Returns:
A prediction-flavored signature_def.
Raises:
ValueError: If inputs or outputs is `None`.
"""
if inputs is None or not inputs:
raise ValueError('Prediction inputs cannot be None or empty.')
if outputs is None or not outputs:
raise ValueError('Prediction outputs cannot be None or empty.')
signature_inputs = {key: utils.build_tensor_info(tensor)
for key, tensor in inputs.items()}
signature_outputs = {key: utils.build_tensor_info(tensor)
for key, tensor in outputs.items()}
signature_def = build_signature_def(
signature_inputs, signature_outputs,
signature_constants.PREDICT_METHOD_NAME)
return signature_def
def classification_signature_def(examples, classes, scores):
"""Creates classification signature from given examples and predictions.
This function produces signatures intended for use with the TensorFlow Serving
Classify API (tensorflow_serving/apis/prediction_service.proto), and so
constrains the input and output types to those allowed by TensorFlow Serving.
Args:
examples: A string `Tensor`, expected to accept serialized tf.Examples.
classes: A string `Tensor`. Note that the ClassificationResponse message
requires that class labels are strings, not integers or anything else.
scores: a float `Tensor`.
Returns:
A classification-flavored signature_def.
Raises:
ValueError: If examples is `None`.
"""
if examples is None:
raise ValueError('Classification `examples` cannot be None.')
if not isinstance(examples, ops.Tensor):
raise ValueError('Classification `examples` must be a string Tensor. '
f'Found `examples` of type {type(examples)}.')
if classes is None and scores is None:
raise ValueError(
'Classification `classes` and `scores` cannot both be '
'None.')
input_tensor_info = utils.build_tensor_info(examples)
if input_tensor_info.dtype != types_pb2.DT_STRING:
raise ValueError(
'Classification input tensors must be of type string. '
f'Found tensors of type {input_tensor_info.dtype}')
signature_inputs = {signature_constants.CLASSIFY_INPUTS: input_tensor_info}
signature_outputs = {}
if classes is not None:
classes_tensor_info = utils.build_tensor_info(classes)
if classes_tensor_info.dtype != types_pb2.DT_STRING:
raise ValueError(
'Classification classes must be of type string Tensor. '
f'Found tensors of type {classes_tensor_info.dtype}.`')
signature_outputs[signature_constants.CLASSIFY_OUTPUT_CLASSES] = (
classes_tensor_info)
if scores is not None:
scores_tensor_info = utils.build_tensor_info(scores)
if scores_tensor_info.dtype != types_pb2.DT_FLOAT:
raise ValueError('Classification scores must be a float Tensor.')
signature_outputs[signature_constants.CLASSIFY_OUTPUT_SCORES] = (
scores_tensor_info)
signature_def = build_signature_def(
signature_inputs, signature_outputs,
signature_constants.CLASSIFY_METHOD_NAME)
return signature_def
def classification_signature_def(examples, classes, scores):
"""Creates classification signature from given examples and predictions.
This function produces signatures intended for use with the TensorFlow Serving
Classify API (tensorflow_serving/apis/prediction_service.proto), and so
constrains the input and output types to those allowed by TensorFlow Serving.
Args:
examples: A string `Tensor`, expected to accept serialized tf.Examples.
classes: A string `Tensor`. Note that the ClassificationResponse message
requires that class labels are strings, not integers or anything else.
scores: a float `Tensor`.
Returns:
A classification-flavored signature_def.
Raises:
ValueError: If examples is `None`.
"""
if examples is None:
raise ValueError('Classification examples cannot be None.')
if not isinstance(examples, ops.Tensor):
raise ValueError('Classification examples must be a string Tensor.')
if classes is None and scores is None:
raise ValueError('Classification classes and scores cannot both be None.')
input_tensor_info = utils.build_tensor_info(examples)
if input_tensor_info.dtype != types_pb2.DT_STRING:
raise ValueError('Classification examples must be a string Tensor.')
signature_inputs = {signature_constants.CLASSIFY_INPUTS: input_tensor_info}
signature_outputs = {}
if classes is not None:
classes_tensor_info = utils.build_tensor_info(classes)
if classes_tensor_info.dtype != types_pb2.DT_STRING:
raise ValueError('Classification classes must be a string Tensor.')
signature_outputs[signature_constants.CLASSIFY_OUTPUT_CLASSES] = (
classes_tensor_info)
if scores is not None:
scores_tensor_info = utils.build_tensor_info(scores)
if scores_tensor_info.dtype != types_pb2.DT_FLOAT:
raise ValueError('Classification scores must be a float Tensor.')
signature_outputs[signature_constants.CLASSIFY_OUTPUT_SCORES] = (
scores_tensor_info)
signature_def = build_signature_def(
signature_inputs, signature_outputs,
signature_constants.CLASSIFY_METHOD_NAME)
return signature_def
def _supervised_signature_def(method_name,
inputs,
loss=None,
predictions=None,
metrics=None):
"""Creates a signature for training and eval data.
This function produces signatures that describe the inputs and outputs
of a supervised process, such as training or evaluation, that
results in loss, metrics, and the like. Note that this function only requires
inputs to be not None.
Args:
method_name: Method name of the SignatureDef as a string.
inputs: dict of string to `Tensor`.
loss: dict of string to `Tensor` representing computed loss.
predictions: dict of string to `Tensor` representing the output predictions.
metrics: dict of string to `Tensor` representing metric ops.
Returns:
A train- or eval-flavored signature_def.
Raises:
ValueError: If inputs or outputs is `None`.
"""
if inputs is None or not inputs:
raise ValueError(
'{} inputs cannot be None or empty.'.format(method_name))
signature_inputs = {
key: utils.build_tensor_info(tensor)
for key, tensor in inputs.items()
}
signature_outputs = {}
for output_set in (loss, predictions, metrics):
if output_set is not None:
sig_out = {
key: utils.build_tensor_info(tensor)
for key, tensor in output_set.items()
}
signature_outputs.update(sig_out)
signature_def = build_signature_def(signature_inputs, signature_outputs,
method_name)
return signature_def
def regression_signature_def(examples, predictions):
"""Creates regression signature from given examples and predictions.
This function produces signatures intended for use with the TensorFlow Serving
Regress API (tensorflow_serving/apis/prediction_service.proto), and so
constrains the input and output types to those allowed by TensorFlow Serving.
Args:
examples: A string `Tensor`, expected to accept serialized tf.Examples.
predictions: A float `Tensor`.
Returns:
A regression-flavored signature_def.
Raises:
ValueError: If examples is `None`.
"""
if examples is None:
raise ValueError('Regression `examples` cannot be None.')
if not isinstance(examples, ops.Tensor):
raise ValueError(
'Expected regression `examples` to be of type Tensor. '
f'Found `examples` of type {type(examples)}.')
if predictions is None:
raise ValueError('Regression `predictions` cannot be None.')
input_tensor_info = utils.build_tensor_info(examples)
if input_tensor_info.dtype != types_pb2.DT_STRING:
raise ValueError('Regression input tensors must be of type string. '
f'Found tensors with type {input_tensor_info.dtype}.')
signature_inputs = {signature_constants.REGRESS_INPUTS: input_tensor_info}
output_tensor_info = utils.build_tensor_info(predictions)
if output_tensor_info.dtype != types_pb2.DT_FLOAT:
raise ValueError(
'Regression output tensors must be of type float. '
f'Found tensors with type {output_tensor_info.dtype}.')
signature_outputs = {
signature_constants.REGRESS_OUTPUTS: output_tensor_info
}
signature_def = build_signature_def(
signature_inputs, signature_outputs,
signature_constants.REGRESS_METHOD_NAME)
return signature_def
def _v1_output_shape_saved_model(self):
export_graph = ops.Graph()
with export_graph.as_default():
start = array_ops.placeholder(
shape=[None], dtype=dtypes.float32, name="start")
output = array_ops.identity(start, name="output")
output.set_shape([1]) # Ok to use [1] because shape is only informational
with session_lib.Session() as session:
path = os.path.join(self.get_temp_dir(), "saved_model", str(ops.uid()))
builder = builder_impl.SavedModelBuilder(path)
builder.add_meta_graph_and_variables(
session,
tags=[tag_constants.SERVING],
signature_def_map={
"serving_default":
signature_def_utils.build_signature_def(
{"start": utils_impl.build_tensor_info(start)},
{"output": utils_impl.build_tensor_info(output)})
})
builder.save()
return path
def _v1_output_shape_saved_model(self):
export_graph = ops.Graph()
with export_graph.as_default():
start = array_ops.placeholder(
shape=[None], dtype=dtypes.float32, name="start")
output = array_ops.identity(start, name="output")
output.set_shape([1]) # Ok to use [1] because shape is only informational
with session_lib.Session() as session:
path = os.path.join(self.get_temp_dir(), "saved_model", str(ops.uid()))
builder = builder_impl.SavedModelBuilder(path)
builder.add_meta_graph_and_variables(
session,
tags=[tag_constants.SERVING],
signature_def_map={
"serving_default":
signature_def_utils.build_signature_def(
{"start": utils_impl.build_tensor_info(start)},
{"output": utils_impl.build_tensor_info(output)})
})
builder.save()
return path
def _supervised_signature_def(
method_name, inputs, loss=None, predictions=None,
metrics=None):
"""Creates a signature for training and eval data.
This function produces signatures that describe the inputs and outputs
of a supervised process, such as training or evaluation, that
results in loss, metrics, and the like. Note that this function only requires
inputs to be not None.
Args:
method_name: Method name of the SignatureDef as a string.
inputs: dict of string to `Tensor`.
loss: dict of string to `Tensor` representing computed loss.
predictions: dict of string to `Tensor` representing the output predictions.
metrics: dict of string to `Tensor` representing metric ops.
Returns:
A train- or eval-flavored signature_def.
Raises:
ValueError: If inputs or outputs is `None`.
"""
if inputs is None or not inputs:
raise ValueError('{} inputs cannot be None or empty.'.format(method_name))
signature_inputs = {key: utils.build_tensor_info(tensor)
for key, tensor in inputs.items()}
signature_outputs = {}
for output_set in (loss, predictions, metrics):
if output_set is not None:
sig_out = {key: utils.build_tensor_info(tensor)
for key, tensor in output_set.items()}
signature_outputs.update(sig_out)
signature_def = build_signature_def(
signature_inputs, signature_outputs, method_name)
return signature_def
def regression_signature_def(examples, predictions):
"""Creates regression signature from given examples and predictions.
This function produces signatures intended for use with the TensorFlow Serving
Regress API (tensorflow_serving/apis/prediction_service.proto), and so
constrains the input and output types to those allowed by TensorFlow Serving.
Args:
examples: A string `Tensor`, expected to accept serialized tf.Examples.
predictions: A float `Tensor`.
Returns:
A regression-flavored signature_def.
Raises:
ValueError: If examples is `None`.
"""
if examples is None:
raise ValueError('Regression examples cannot be None.')
if not isinstance(examples, ops.Tensor):
raise ValueError('Regression examples must be a string Tensor.')
if predictions is None:
raise ValueError('Regression predictions cannot be None.')
input_tensor_info = utils.build_tensor_info(examples)
if input_tensor_info.dtype != types_pb2.DT_STRING:
raise ValueError('Regression examples must be a string Tensor.')
signature_inputs = {signature_constants.REGRESS_INPUTS: input_tensor_info}
output_tensor_info = utils.build_tensor_info(predictions)
if output_tensor_info.dtype != types_pb2.DT_FLOAT:
raise ValueError('Regression output must be a float Tensor.')
signature_outputs = {signature_constants.REGRESS_OUTPUTS: output_tensor_info}
signature_def = build_signature_def(
signature_inputs, signature_outputs,
signature_constants.REGRESS_METHOD_NAME)
return signature_def
def _unfed_placeholder_signature(self):
export_graph = ops.Graph()
with export_graph.as_default():
x = array_ops.placeholder(name="x", shape=[], dtype=dtypes.float32)
output = x * random_ops.random_normal([2])
with session_lib.Session() as session:
path = os.path.join(self.get_temp_dir(), "saved_model", str(ops.uid()))
b = builder_impl.SavedModelBuilder(path)
b.add_meta_graph_and_variables(
session,
tags=[tag_constants.SERVING],
signature_def_map={
"key": signature_def_utils.build_signature_def(
{}, dict(value=utils_impl.build_tensor_info(output)))})
b.save()
return path
def _signature_with_no_inputs(self):
export_graph = ops.Graph()
with export_graph.as_default():
array_ops.placeholder(name="x", shape=[], dtype=dtypes.float32)
output = random_ops.random_normal([2])
with session_lib.Session() as session:
path = os.path.join(self.get_temp_dir(), "saved_model", str(ops.uid()))
b = builder_impl.SavedModelBuilder(path)
b.add_meta_graph_and_variables(
session,
tags=[tag_constants.SERVING],
signature_def_map={
"key": signature_def_utils.build_signature_def(
{}, dict(value=utils_impl.build_tensor_info(output)))})
b.save()
return path
def _tensor_dict_to_tensorinfo(tensor_dict):
return {
key: utils_impl.build_tensor_info(value)
for key, value in tensor_dict.items()
}
def _tensor_dict_to_tensorinfo(tensor_dict):
return {key: utils_impl.build_tensor_info(value)
for key, value in tensor_dict.items()}
