def predict(self, request_data, request_timeout=10):
self.logger.info('Sending request to tfserving model')
self.logger.info('Host: {}'.format(self.host))
self.logger.info('Model name: {}'.format(self.model_name))
self.logger.info('Model version: {}'.format(self.model_version))
# Create gRPC client and request
t = time.time()
channel = grpc.insecure_channel(self.host)
self.logger.debug(
'Establishing insecure channel took: {}'.format(time.time() - t))
t = time.time()
stub = PredictionServiceStub(channel)
self.logger.debug('Creating stub took: {}'.format(time.time() - t))
t = time.time()
request = PredictRequest()
self.logger.debug(
'Creating request object took: {}'.format(time.time() - t))
request.model_spec.name = self.model_name
if self.model_version > 0:
request.model_spec.version.value = self.model_version
# The 'map<string, TensorProto> inputs = 2;' in predict.proto corresponds to
# the parameter inputs that we send to the signature definition in our model.
# Here we fill in the inputs. These protos are copied from tensorflow serving
# apis:
# https://github.com/tensorflow/serving/tree/r1.7/tensorflow_serving/apis
t = time.time()
for d in request_data:
tensor_proto = make_tensor_proto(d['data'], d['in_tensor_dtype'])
request.inputs[d['in_tensor_name']].CopyFrom(tensor_proto)
self.logger.debug('Making tensor protos took: {}'.format(time.time() - t))
try:
# In the bottom of predict.proto, PredictResponse is defined. This message
# has one field: 'map<string, TensorProto> outputs = 1;'. The key in this
# map is our ‘add’ from our model.
t = time.time()
predict_response = stub.Predict(request, timeout=request_timeout)
self.logger.debug(
'Actual request took: {} seconds'.format(time.time() - t))
predict_response_dict = predict_response_to_dict(predict_response)
keys = [k for k in predict_response_dict]
self.logger.info('Got predict_response with keys: {}'.format(keys))
return predict_response_dict
except RpcError as e:
self.logger.error(e)
self.logger.error('Prediction failed!')
return {}
def predict(self, request_data, request_timeout=10, max_message_size=4):
self.logger.info('Sending request to tfserving model')
self.logger.info('Host: {}'.format(self.host))
self.logger.info('Model name: {}'.format(self.model_name))
self.logger.info('Model version: {}'.format(self.model_version))
# Create gRPC client and request
t = time.time()
# Changed to allow big images
options = [('grpc.max_message_length', max_message_size * 1024 * 1024),
('grpc.max_send_message_length', max_message_size * 1024 * 1024),
('grpc.max_receive_message_length', max_message_size * 1024 * 1024)
]
channel = grpc.insecure_channel(self.host, options=options)
self.logger.debug('Establishing insecure channel took: {}'.format(time.time() - t))
t = time.time()
stub = PredictionServiceStub(channel)
self.logger.debug('Creating stub took: {}'.format(time.time() - t))
t = time.time()
request = PredictRequest()
self.logger.debug('Creating request object took: {}'.format(time.time() - t))
request.model_spec.name = self.model_name
if self.model_version > 0:
request.model_spec.version.value = self.model_version
t = time.time()
for d in request_data:
tensor_proto = make_tensor_proto(d['data'], d['in_tensor_dtype'])
request.inputs[d['in_tensor_name']].CopyFrom(tensor_proto)
self.logger.debug('Making tensor protos took: {}'.format(time.time() - t))
try:
t = time.time()
predict_response = stub.Predict(request, timeout=request_timeout)
self.logger.debug('Actual request took: {} seconds'.format(time.time() - t))
predict_response_dict = predict_response_to_dict(predict_response)
keys = [k for k in predict_response_dict]
self.logger.info('Got predict_response with keys: {}'.format(keys))
return predict_response_dict
except RpcError as e:
self.logger.error(e)
self.logger.error('Prediction failed!')
return {}
def predict(self, request_data, request_timeout=10):
self.logger.info('Sending request to tfserving model')
self.logger.info('Host: {}'.format(self.host))
self.logger.info('Model name: {}'.format(self.model_name))
self.logger.info('Model version: {}'.format(self.model_version))
t = time.time()
request = PredictRequest()
self.logger.debug(
'Creating request object took: {}'.format(time.time() - t))
request.model_spec.name = self.model_name
if self.model_version > 0:
request.model_spec.version.value = self.model_version
t = time.time()
for d in request_data:
tensor_proto = make_tensor_proto(d['data'], d['in_tensor_dtype'])
request.inputs[d['in_tensor_name']].CopyFrom(tensor_proto)
self.logger.debug('Making tensor protos took: {}'.format(time.time() -
t))
try:
t = time.time()
predict_response = self.stub.Predict(request,
timeout=request_timeout)
self.logger.debug(
'Actual request took: {} seconds'.format(time.time() - t))
predict_response_dict = predict_response_to_dict(predict_response)
keys = [k for k in predict_response_dict]
self.logger.info('Got predict_response with keys: {}'.format(keys))
return predict_response_dict
except RpcError as e:
self.logger.error(e)
self.logger.error('Prediction failed!')
return {'error': str(e)}
return {}
def predict(self,
request_data: List[Dict[str, Any]],
request_timeout: int = 10):
""" Get a model prediction on request data
Args:
request_data: List of input graph nodes in the model, containing the following required fields:
'data': The data to get predictions on
'in_tensor_dtype': The datatype of the input, legal types are the keys amongst "dtype_to_number", defined in util.py
'in_tensor_name': The name of the models input graph node, often 'inputs'
request_timeout: timeout in seconds
Returns:
Empty dict on error, otherwise a dict with keys set as output graph node names and values set to their predicted value
Example:
>>> prod_client = ProdClient(host='localhost:9000', model_name='mnist', model_version=1)
>>> prod_client.predict(
... request_data=[{
... 'data': np.asarray([image]),
... 'in_tensor_dtype': 'DT_UINT8',
... 'in_tensor_name': 'inputs',
... }]
... )
"""
self.logger.info('Sending request to tfserving model')
self.logger.info('Host: {}'.format(self.host))
self.logger.info('Model name: {}'.format(self.model_name))
self.logger.info('Model version: {}'.format(self.model_version))
# Create gRPC client and request
t = time.time()
with grpc.insecure_channel(self.host, options=self.options) as channel:
self.logger.debug(
'Establishing insecure channel took: {}'.format(time.time() -
t))
t = time.time()
stub = PredictionServiceStub(channel)
self.logger.debug('Creating stub took: {}'.format(time.time() - t))
t = time.time()
request = PredictRequest()
self.logger.debug(
'Creating request object took: {}'.format(time.time() - t))
request.model_spec.name = self.model_name
if self.model_version > 0:
request.model_spec.version.value = self.model_version
t = time.time()
for d in request_data:
tensor_proto = make_tensor_proto(d['data'],
d['in_tensor_dtype'])
request.inputs[d['in_tensor_name']].CopyFrom(tensor_proto)
self.logger.debug(
'Making tensor protos took: {}'.format(time.time() - t))
try:
t = time.time()
predict_response = stub.Predict(request,
timeout=request_timeout)
self.logger.debug(
'Actual request took: {} seconds'.format(time.time() - t))
predict_response_dict = predict_response_to_dict(
predict_response)
keys = [k for k in predict_response_dict]
self.logger.info(
'Got predict_response with keys: {}'.format(keys))
return predict_response_dict
except RpcError as e:
self.logger.error(e)
self.logger.error('Prediction failed!')
return {}
def predict(self, request_data, request_timeout=10):
# SELF: is the class variable
# REQUEST_DATA: is the input to the model
# REQUEST_TIMEOUT: Time in seconds after which request will end
# Start with logger informations
self.logger.info('Sending request to tfserving model')
self.logger.info('Host: {}'.format(self.host))
self.logger.info('Model name: {}'.format(self.model_name))
self.logger.info('Model version: {}'.format(self.model_version))
# Create gRPC client and request
t = time.time()
channel = grpc.insecure_channel(self.host)
self.logger.debug(
'Establishing insecure channel took: {}'.format(time.time() - t))
# We will check the time taken for each operation
t = time.time()
stub = PredictionServiceStub(channel)
self.logger.debug('Creating stub took: {}'.format(time.time() - t))
t = time.time()
request = PredictRequest()
self.logger.debug(
'Creating request object took: {}'.format(time.time() - t))
# Here we will define the model name
request.model_spec.name = self.model_name
# Specific version of the model
if self.model_version > 0:
request.model_spec.version.value = self.model_version
t = time.time()
# Put inputs in the request
for d in request_data:
tensor_proto = make_tensor_proto(d['data'], d['in_tensor_dtype'])
request.inputs[\
d['in_tensor_name']].CopyFrom(tensor_proto)
self.logger.debug(\
'Making tensor protos took: {}'.format(time.time() - t))
# Now we are ready for prediction following line will help us in that
try:
t = time.time()
predict_response = stub.Predict(request, timeout=request_timeout)
self.logger.debug(\
'Actual request took: {} seconds'.format(time.time() - t))
predict_response_dict = \
predict_response_to_dict(predict_response)
keys = [k for k in predict_response_dict]
self.logger.info('\
Got predict_response with keys: {}'.format(keys))
return predict_response_dict
except RpcError as e:
self.logger.error(e)
self.logger.error('Prediction failed!')
return {}