def test_convert_for_regress_invalid_output_example_spec(self):
prediction_log = text_format.Parse(
"""
regress_log {
request {
input {
example_list {
examples {
features {
feature: {
key: "regress_input"
value: { bytes_list: { value: "feature" } }
}
}
}
}
}
}
response {
result {
regressions {
value: 0.7
}
}
}
}
""", prediction_log_pb2.PredictionLog())
output_example_spec = text_format.Parse(
"""
output_columns_spec {
}
""", bulk_inferrer_pb2.OutputExampleSpec())
with self.assertRaises(ValueError):
utils.convert(prediction_log, output_example_spec)
def test_convert_for_predict_invalid_output_example_spec(self):
example = text_format.Parse(
"""
features {
feature { key: "predict_input" value: { bytes_list: { value: "feature" } } }
}""", tf.train.Example())
prediction_log = text_format.Parse(
"""
predict_log {
request {
inputs {
key: "%s"
value {
dtype: DT_STRING
tensor_shape { dim { size: 1 } }
}
}
}
response {
outputs {
key: "output_float"
value {
dtype: DT_FLOAT
tensor_shape { dim { size: 1 } dim { size: 2 }}
float_val: 0.1
float_val: 0.2
}
}
outputs {
key: "output_bytes"
value {
dtype: DT_STRING
tensor_shape { dim { size: 1 }}
string_val: "prediction"
}
}
}
}
""" % (utils.INPUT_KEY), prediction_log_pb2.PredictionLog())
# The ending quote cannot be recognized correctly when `string_val` field
# is directly set with a serialized string quoted in the text format.
prediction_log.predict_log.request.inputs[
utils.INPUT_KEY].string_val.append(example.SerializeToString())
output_example_spec = text_format.Parse(
"""
output_columns_spec {
}
""", bulk_inferrer_pb2.OutputExampleSpec())
with self.assertRaises(ValueError):
utils.convert(prediction_log, output_example_spec)
def test_convert_for_regress(self):
prediction_log = text_format.Parse(
"""
regress_log {
request {
input {
example_list {
examples {
features {
feature: {
key: "regress_input"
value: { bytes_list: { value: "feature" } }
}
}
}
}
}
}
response {
result {
regressions {
value: 0.7
}
}
}
}
""", prediction_log_pb2.PredictionLog())
output_example_spec = text_format.Parse(
"""
output_columns_spec {
regress_output {
value_column: 'regress_value'
}
}
""", bulk_inferrer_pb2.OutputExampleSpec())
expected_example = text_format.Parse(
"""
features {
feature: {
key: "regress_input"
value: { bytes_list: { value: "feature" } }
}
feature: {
key: "regress_value"
value: { float_list: { value: 0.7 } }
}
}
""", tf.train.Example())
self.assertProtoEquals(
expected_example, utils.convert(prediction_log,
output_example_spec))
def convert_to_dict(prediction_log: prediction_log_pb2.PredictionLog,
output_example_spec: _OutputExampleSpecType) \
-> tf.train.Example:
"""Converts given `prediction_log` to a `tf.train.Example`.
Args:
prediction_log: The input prediction log.
output_example_spec: The spec for how to map prediction results to
columns in example.
Returns:
A `tf.train.Example` converted from the given prediction_log.
Raises:
ValueError: If the inference type or signature name in spec does not
match that in prediction_log.
"""
tf_example = convert(prediction_log=prediction_log,
output_example_spec=output_example_spec)
return tf_example
def test_convert_for_multi_inference(self):
prediction_log = text_format.Parse(
"""
multi_inference_log {
request {
input {
example_list {
examples {
features {
feature: {
key: "input"
value: { bytes_list: { value: "feature" } }
}
}
}
}
}
}
response {
results {
model_spec {
signature_name: 'classification'
}
classification_result {
classifications {
classes {
label: '1'
score: 0.6
}
classes {
label: '0'
score: 0.4
}
}
}
}
results {
model_spec {
signature_name: 'regression'
}
regression_result {
regressions {
value: 0.7
}
}
}
}
}
""", prediction_log_pb2.PredictionLog())
output_example_spec = text_format.Parse(
"""
output_columns_spec {
signature_name: 'classification'
classify_output {
label_column: 'classify_label'
score_column: 'classify_score'
}
}
output_columns_spec {
signature_name: 'regression'
regress_output {
value_column: 'regress_value'
}
}
""", bulk_inferrer_pb2.OutputExampleSpec())
expected_example = text_format.Parse(
"""
features {
feature: {
key: "input"
value: { bytes_list: { value: "feature" } }
}
feature: {
key: "classify_label"
value: { bytes_list: { value: "1" value: "0"} }
}
feature: {
key: "classify_score"
value: { float_list: { value: 0.6 value: 0.4} }
}
feature: {
key: "regress_value"
value: { float_list: { value: 0.7} }
}
}
""", tf.train.Example())
self.assertProtoEquals(
expected_example, utils.convert(prediction_log,
output_example_spec))