def test_make_json_formatted_for_single_chart_classification(self):
"""Test making a classification chart with a single point on it."""
inference_result_proto = classification_pb2.ClassificationResponse()
classification = inference_result_proto.result.classifications.add()
inference_class = classification.classes.add()
inference_class.label = 'class_a'
inference_class.score = 0.7
inference_class = classification.classes.add()
inference_class.label = 'class_b'
inference_class.score = 0.3
original_feature = inference_utils.OriginalFeatureList(
'feature_name', [2.], 'float_list')
mutant_feature = inference_utils.MutantFeatureValue(original_feature,
index=0,
mutant_value=20)
jsonable = inference_utils.make_json_formatted_for_single_chart(
[mutant_feature], inference_result_proto, 0)
self.assertEqual(['class_a', 'class_b'], sorted(jsonable.keys()))
self.assertEqual(1, len(jsonable['class_a']))
self.assertEqual(20, jsonable['class_a'][0]['step'])
self.assertAlmostEqual(0.7, jsonable['class_a'][0]['scalar'])
self.assertEqual(1, len(jsonable['class_b']))
self.assertEqual(20, jsonable['class_b'][0]['step'])
self.assertAlmostEqual(0.3, jsonable['class_b'][0]['scalar'])
def test_pb_request():
request = classification_pb2.ClassificationRequest()
request.model_spec.name = "generic_model"
request.model_spec.signature_name = DEFAULT_SERVING_SIGNATURE_DEF_KEY
example = request.input.example_list.examples.add()
data = [4.9, 2.5, 4.5, 1.7]
example.features.feature[PREDICT_INPUTS].float_list.value.extend(data)
serialized_output = requests.post("http://localhost:8080/invocations",
data=request.SerializeToString(),
headers={
'Content-type':
'application/octet-stream',
'Accept': 'application/octet-stream'
}).content
classification_response = classification_pb2.ClassificationResponse()
classification_response.ParseFromString(serialized_output)
classifications_classes = classification_response.result.classifications[
0].classes
assert len(classifications_classes) == 3
for c in classifications_classes:
assert c.score < 1
assert c.score > 0
def convert_prediction_values(values, serving_bundle, model_spec=None):
"""Converts tensor values into ClassificationResponse or
RegressionResponse.
Args:
values: For classification, a 2D list of numbers. The first dimension is for
each example being predicted. The second dimension are the probabilities
for each class ID in the prediction. For regression, a 1D list of numbers,
with a regression score for each example being predicted.
serving_bundle: A `ServingBundle` object that contains the information about
the serving request that the response was generated by.
model_spec: Optional model spec to put into the response.
Returns:
A ClassificationResponse or RegressionResponse.
"""
if serving_bundle.model_type == "classification":
response = classification_pb2.ClassificationResponse()
for example_index in range(len(values)):
classification = response.result.classifications.add()
for class_index in range(len(values[example_index])):
class_score = classification.classes.add()
class_score.score = values[example_index][class_index]
class_score.label = str(class_index)
else:
response = regression_pb2.RegressionResponse()
for example_index in range(len(values)):
regression = response.result.regressions.add()
regression.value = values[example_index]
if model_spec:
response.model_spec.CopyFrom(model_spec)
return response
def convert_predict_response(pred, serving_bundle):
"""Converts a PredictResponse to ClassificationResponse or RegressionResponse.
Args:
pred: PredictResponse to convert.
serving_bundle: A `ServingBundle` object that contains the information about
the serving request that the response was generated by.
Returns:
A ClassificationResponse or RegressionResponse.
"""
output = pred.outputs[serving_bundle.predict_output_tensor]
raw_output = output.float_val
output_index = 0
if serving_bundle.model_type == 'classification':
response = classification_pb2.ClassificationResponse()
response.model_spec.CopyFrom(pred.model_spec)
for _ in range(output.tensor_shape.dim[0].size):
classification = response.result.classifications.add()
for class_index in range(output.tensor_shape.dim[1].size):
class_score = classification.classes.add()
class_score.score = raw_output[output_index]
output_index += 1
class_score.label = str(class_index)
else:
response = regression_pb2.RegressionResponse()
response.model_spec.CopyFrom(pred.model_spec)
for _ in range(output.tensor_shape.dim[0].size):
regression = response.result.regressions.add()
regression.value = raw_output[output_index]
output_index += 1
return response
def test_wrap_inference_results_classification(self):
"""Test wrapping a classification result."""
inference_result_proto = classification_pb2.ClassificationResponse()
classification = inference_result_proto.result.classifications.add()
inference_class = classification.classes.add()
inference_class.label = 'class_b'
inference_class.score = 0.3
inference_class = classification.classes.add()
inference_class.label = 'class_a'
inference_class.score = 0.7
wrapped = inference_utils.wrap_inference_results(inference_result_proto)
self.assertEqual(1, len(wrapped.classification_result.classifications))
self.assertEqual(
2, len(wrapped.classification_result.classifications[0].classes))
def Classify(self, request, context):
try:
example = request.input.example_list.examples[0]
start_time_millis = self.getCurrentTimeinMillis()
time_1 = datetime.now()
if self._Validations(request, context) is True:
time_2 = datetime.now()
example = request.input.example_list.examples[0]
idx = 0
final_input = []
final_labels = []
for k in example.features.feature.keys():
v = example.features.feature[k]
int_list_value = list(v.int64_list.value)
float_list_value = list(v.float_list.value)
byte_list_value = list(v.bytes_list.value)
if len(int_list_value) > 0:
final_input.append(int_list_value)
final_labels.append(k)
elif len(float_list_value) > 0:
final_input.append(float_list_value)
final_labels.append(k)
elif len(byte_list_value) > 0:
final_input.append(byte_list_value)
final_labels.append(k)
else:
LOG.info("Input param empty, ignoring it")
idx = idx + 1
classification_request = pd.DataFrame.from_records(
list(zip(*final_input)), columns=final_labels)
try:
classification_outputs = self.model.wrapper_classification_func(
classification_request)
LOG.info(
"the output from the model's classification function %s",
classification_outputs)
response = tensorflow__serving_dot_apis_dot_classification__pb2.ClassificationResponse(
)
except grpc.RpcError as grpc_error:
end_time_millis = self.getCurrentTimeinMillis()
LOG.error("Error while doing Classification")
requestDurationInMillis = str(end_time_millis -
start_time_millis)
self.logRequests(self.model.model_name,
requestDurationInMillis, str(0))
LOG.error("grpc error : %s", str(grpc_error))
s = getattr(grpc_error, 'message', str(grpc_error))
raise grpc.RpcError(grpc_error)
return None
classifications = []
for idx in range(0, classification_request.shape[0]):
classes = []
classification = tensorflow__serving_dot_apis_dot_classification__pb2.Classifications(
)
for k, v in classification_outputs.items():
class_data = tensorflow__serving_dot_apis_dot_classification__pb2.Class(
)
class_data.label = "Class-" + k
class_data.score = v[idx]
classes.append(class_data)
classification.classes.extend(classes)
classifications.append(classification)
response.result.classifications.extend(classifications)
return response
else:
return None
except Exception as ex:
s = getattr(ex, 'message', str(ex))
raise Exception(s)
return None
