def get_export_signature(examples, features, predictions):
"""Create an export signature with named input and output signatures."""
iris_labels = get_vocabulary(metadata_path)
prediction = tf.argmax(predictions, 1)
labels = tf.contrib.lookup.index_to_string(prediction,
mapping=iris_labels,
default_value=UNKNOWN_LABEL)
outputs = {
SCORES_OUTPUT_COLUMN: predictions.name,
KEY_OUTPUT_COLUMN: tf.squeeze(features[KEY_FEATURE_COLUMN]).name,
LABEL_OUTPUT_COLUMN: labels.name
}
inputs = {EXAMPLES_PLACEHOLDER_KEY: examples.name}
tf.add_to_collection(INPUTS_KEY, json.dumps(inputs))
tf.add_to_collection(OUTPUTS_KEY, json.dumps(outputs))
input_signature = manifest_pb2.Signature()
output_signature = manifest_pb2.Signature()
for name, tensor_name in outputs.iteritems():
output_signature.generic_signature.map[
name].tensor_name = tensor_name
for name, tensor_name in inputs.iteritems():
input_signature.generic_signature.map[
name].tensor_name = tensor_name
# Return None for default classification signature.
return None, {
INPUTS_KEY: input_signature,
OUTPUTS_KEY: output_signature
}
def get_export_signature(examples, features, predictions):
"""Create a classification signature function and add output placeholders."""
inputs = {'examples': examples.name}
tf.add_to_collection('inputs', json.dumps(inputs))
prediction = tf.argmax(predictions, 1)
labels = tf.contrib.lookup.index_to_string(
prediction, mapping=['0', '1'], default_value='UNKNOWN_LABEL')
outputs = {'score': predictions.name,
'key': features[KEY_FEATURE_COLUMN].name,
'predicted_click_value': labels.name}
tf.add_to_collection('outputs', json.dumps(outputs))
output_signature = manifest_pb2.Signature()
input_signature = manifest_pb2.Signature()
for name, tensor_name in outputs.iteritems():
output_signature.generic_signature.map[name].tensor_name = tensor_name
for name, tensor_name in inputs.iteritems():
input_signature.generic_signature.map[name].tensor_name = tensor_name
# Return None for default classification signature..
return None, {'inputs': input_signature, 'outputs': output_signature}
def _signature_fn(examples, features, predictions):
"""Create a classification signature function and add to collections."""
# Mark the inputs.
inputs = {"examples": examples.name}
tf.add_to_collection("inputs", json.dumps(inputs))
concat_embeddings = tf.get_collection(CONCAT_EMBEDDINGS_KEY)[0]
outputs = {
"score": predictions.name,
"key": features[FLAGS.id_field].name,
"target": features[FLAGS.target_field + "_string"].name,
"embeddings": concat_embeddings.name
}
tf.add_to_collection("outputs", json.dumps(outputs))
output_signature = manifest_pb2.Signature()
for name, tensor_name in outputs.iteritems():
output_signature.generic_signature.map[name].tensor_name = tensor_name
input_signature = manifest_pb2.Signature()
for name, tensor_name in inputs.iteritems():
input_signature.generic_signature.map[name].tensor_name = tensor_name
# Create a classification signature for serving prediction.
signature = manifest_pb2.Signature()
signature.classification_signature.input.tensor_name = examples.name
signature.classification_signature.scores.tensor_name = predictions.name
# Returns a tuple of None default signature and a dictionary of named
# signatures with inputs and outputs.
return signature, {"inputs": input_signature, "outputs": output_signature}
def _build_signature(examples, features, predictions):
"""Create a generic signature function with input and output signatures."""
iris_labels = VocabGetter(metadata_path).get_vocab().keys()
prediction = tf.argmax(predictions, 1)
labels = tf.contrib.lookup.index_to_string(prediction,
mapping=iris_labels,
default_value=UNKNOWN_LABEL)
target = tf.contrib.lookup.index_to_string(tf.squeeze(features[TARGET_KEY]),
mapping=iris_labels,
default_value=UNKNOWN_LABEL)
outputs = {SCORES_COLUMN: predictions.name,
KEY_COLUMN: tf.squeeze(features[KEY_COLUMN]).name,
TARGET_COLUMN: target.name,
LABEL_COLUMN: labels.name}
inputs = {EXAMPLES_KEY: examples.name}
tf.add_to_collection(OUTPUTS_KEY, json.dumps(outputs))
tf.add_to_collection(INPUTS_KEY, json.dumps(inputs))
input_signature = manifest_pb2.Signature()
output_signature = manifest_pb2.Signature()
for name, tensor_name in outputs.iteritems():
output_signature.generic_signature.map[name].tensor_name = tensor_name
for name, tensor_name in inputs.iteritems():
input_signature.generic_signature.map[name].tensor_name = tensor_name
# Return None for default classification signature.
return None, {INPUTS_KEY: input_signature,
OUTPUTS_KEY: output_signature}
def get_export_signature(examples, features, predictions):
"""Create a classification signature function and add output placeholders."""
inputs = {'examples': examples.name}
tf.add_to_collection('inputs', json.dumps(inputs))
# TensorForest currently outputs a dict of both probabilities and
# class predictions.
if isinstance(predictions, dict) and (eval_metrics.INFERENCE_PROB_NAME
in predictions):
predictions = predictions[eval_metrics.INFERENCE_PROB_NAME]
prediction = tf.argmax(predictions, 1)
labels = tf.contrib.lookup.index_to_string(prediction,
mapping=['0', '1'],
default_value='UNKNOWN_LABEL')
outputs = {
'score': predictions.name,
'key': features[KEY_FEATURE_COLUMN].name,
'predicted_click_value': labels.name
}
tf.add_to_collection('outputs', json.dumps(outputs))
output_signature = manifest_pb2.Signature()
input_signature = manifest_pb2.Signature()
for name, tensor_name in outputs.iteritems():
output_signature.generic_signature.map[name].tensor_name = tensor_name
for name, tensor_name in inputs.iteritems():
input_signature.generic_signature.map[name].tensor_name = tensor_name
# Return None for default classification signature..
return None, {'inputs': input_signature, 'outputs': output_signature}
def generic_signature(name_tensor_map):
"""Creates a generic signature of name to Tensor name.
Args:
name_tensor_map: Map from logical name to Tensor.
Returns:
A Signature message.
"""
signature = manifest_pb2.Signature()
for name, tensor in six.iteritems(name_tensor_map):
signature.generic_signature.map[name].tensor_name = tensor.name
return signature
def regression_signature(input_tensor, output_tensor):
"""Creates a regression signature.
Args:
input_tensor: Tensor specifying the input to a graph.
output_tensor: Tensor specifying the output of a graph.
Returns:
A Signature message.
"""
signature = manifest_pb2.Signature()
signature.regression_signature.input.tensor_name = input_tensor.name
signature.regression_signature.output.tensor_name = output_tensor.name
return signature
def classification_signature(input_tensor,
classes_tensor=None,
scores_tensor=None):
"""Creates a classification signature.
Args:
input_tensor: Tensor specifying the input to a graph.
classes_tensor: Tensor specifying the output classes of a graph.
scores_tensor: Tensor specifying the scores of the output classes.
Returns:
A Signature message.
"""
signature = manifest_pb2.Signature()
signature.classification_signature.input.tensor_name = input_tensor.name
if classes_tensor is not None:
signature.classification_signature.classes.tensor_name = classes_tensor.name
if scores_tensor is not None:
signature.classification_signature.scores.tensor_name = scores_tensor.name
return signature