def preprocess(self, request, context):
"""Takes request data and de-serializes the data into an object for prediction.
When an InvokeEndpoint operation is made against an Endpoint running SageMaker model server,
the model server receives two pieces of information:
- The request Content-Type, for example "application/json"
- The request data, which is at most 5 MB (5 * 1024 * 1024 bytes) in size.
The input_fn is responsible to take the request data and pre-process it before prediction.
Args:
input_data (obj): The request data.
content_type (str): The context for the request
Returns:
(obj): The data-frame ready for prediction.
"""
logging.info('Inside the Pre-processing Function')
request_processor = context.request_processor[0]
request_property = request_processor.get_request_properties()
content_type = utils.retrieve_content_type_header(request_property)
logging.info(content_type)
bytes_data = request[0].get('body')
s = str(bytes_data, 'utf-8')
logging.info(s)
data = StringIO(s)
df = pd.io.json.read_json(data, orient='columns')
logging.info(df)
return df
def transform(self, data, context):
"""Take a request with input data, deserialize it, make a prediction, and return a
serialized response.
Args:
data (obj): the request data.
context (obj): metadata on the incoming request data.
Returns:
list[obj]: The serialized prediction result wrapped in a list if
inference is successful. Otherwise returns an error message
with the context set appropriately.
"""
try:
properties = context.system_properties
model_dir = properties.get("model_dir")
self.validate_and_initialize(model_dir=model_dir)
input_data = data[0].get("body")
request_processor = context.request_processor[0]
request_property = request_processor.get_request_properties()
content_type = utils.retrieve_content_type_header(request_property)
accept = request_property.get("Accept") or request_property.get(
"accept")
if not accept or accept == content_types.ANY:
accept = self._environment.default_accept
if content_type in content_types.UTF8_TYPES:
input_data = input_data.decode("utf-8")
result = self._transform_fn(self._model, input_data, content_type,
accept)
response = result
response_content_type = accept
if isinstance(result, tuple):
# handles tuple for backwards compatibility
response = result[0]
response_content_type = result[1]
context.set_response_content_type(0, response_content_type)
return [response]
except Exception as e: # pylint: disable=broad-except
trace = traceback.format_exc()
if isinstance(e, BaseInferenceToolkitError):
return self.handle_error(context, e, trace)
else:
return self.handle_error(
context,
GenericInferenceToolkitError(
http_client.INTERNAL_SERVER_ERROR, str(e)),
trace,
)
def transform(self, data, context):
"""Take a request with input data, deserialize it, make a prediction, and return a
serialized response.
Args:
data (obj): the request data.
context (obj): metadata on the incoming request data.
Returns:
list[obj]: the serialized prediction result wrapped in a list.
"""
if not self._initialized:
try:
sys_properties = context._system_properties
model_dir = sys_properties.get('model_dir')
self.validate_and_initialize(model_dir)
except Exception as e:
if isinstance(e, BaseInferenceToolkitError):
logging.error("Error loading model: {}".format(e))
return self.handle_error(context, e.status_code, e.message)
else:
raise e
self._initialized = True
try:
input_data = data[0].get('body')
request_processor = context.request_processor[0]
request_property = request_processor.get_request_properties()
content_type = utils.retrieve_content_type_header(request_property)
accept = request_property.get('Accept') or request_property.get('accept')
if not accept or accept == content_types.ANY:
accept = self._environment.default_accept
result = self._transform_fn(self._model, input_data, content_type, accept)
response = result
response_content_type = accept
if isinstance(result, tuple):
# handles tuple for backwards compatibility
response = result[0]
response_content_type = result[1]
context.set_response_content_type(0, response_content_type)
return [response]
except Exception as e:
if isinstance(e, BaseInferenceToolkitError):
logging.error(e)
return self.handle_error(context, e.status_code, e.message)
else:
raise e
def transform(self, data, context):
"""Take a request with input data, deserialize it, make a prediction, and return a
serialized response.
Args:
data (obj): the request data.
context (obj): metadata on the incoming request data.
Returns:
list[obj]: the serialized prediction result wrapped in a list.
"""
self.validate_and_initialize()
input_data = data[0].get('body')
request_processor = context.request_processor[0]
request_property = request_processor.get_request_properties()
content_type = utils.retrieve_content_type_header(request_property)
accept = request_property.get('Accept') or request_property.get(
'accept')
if not accept or accept == content_types.ANY:
accept = self._environment.default_accept
if content_type in content_types.UTF8_TYPES:
input_data = input_data.decode('utf-8')
result = self._transform_fn(self._model, input_data, content_type,
accept)
response = result
response_content_type = accept
if isinstance(result, tuple):
# handles tuple for backwards compatibility
response = result[0]
response_content_type = result[1]
context.set_response_content_type(0, response_content_type)
return [response]
def test_content_type_header(content_type_key):
request_property = {content_type_key: CONTENT_TYPE}
result = retrieve_content_type_header(request_property)
assert result == CONTENT_TYPE
def transform(self, data, context):
"""Take a request with input data, deserialize it, make a prediction, and return a
serialized response.
NOTE: This is almost a copy of the original Transformer method, except it does not decode the utf-8 data.
This is done for backwards compatibility.
See line removed here:
https://github.com/aws/sagemaker-inference-toolkit/blob/master/src/sagemaker_inference/transformer.py#L123
Args:
data (obj): the request data.
context (obj): metadata on the incoming request data.
Returns:
list[obj]: the serialized prediction result wrapped in a list.
"""
if not self._initialized:
try:
sys_properties = context._system_properties
model_dir = sys_properties.get('model_dir')
self.validate_and_initialize(model_dir)
except Exception as e:
if isinstance(e, BaseInferenceToolkitError):
logging.error("Error loading model: {}".format(e))
return self.handle_error(context, e.status_code, e.message)
else:
raise e
self._initialized = True
try:
input_data = data[0].get('body')
request_processor = context.request_processor[0]
request_property = request_processor.get_request_properties()
content_type = utils.retrieve_content_type_header(request_property)
accept = request_property.get('Accept') or request_property.get(
'accept')
if not accept or accept == content_types.ANY:
accept = self._environment.default_accept
result = self._transform_fn(self._model, input_data, content_type,
accept)
response = result
response_content_type = accept
if isinstance(result, tuple):
# handles tuple for backwards compatibility
response = result[0]
response_content_type = result[1]
context.set_response_content_type(0, response_content_type)
return [response]
except Exception as e:
if isinstance(e, BaseInferenceToolkitError):
logging.error(e)
return self.handle_error(context, e.status_code, e.message)
else:
return self.handle_error(context, http.HTTPStatus.BAD_REQUEST,
e.message)
