def default_input_fn(self, input_data, content_type):
"""Take request data and deserialize it 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's content type, for example "application/json"
- The request data
The ``input_fn`` is responsible for preprocessing request data before prediction.
Args:
input_data (obj): the request data
content_type (str): the request's content type
Returns:
mxnet.io.NDArrayIter: data ready for prediction.
Raises:
sagemaker_containers.beta.framework.errors.UnsupportedFormatError: if an unsupported
accept type is used.
"""
if content_type not in self.VALID_CONTENT_TYPES:
raise errors.UnsupportedFormatError(content_type)
np_array = encoders.decode(input_data, content_type)
ndarray = mx.nd.array(np_array).as_in_context(_context())
# We require model to only have one input
[data_shape] = self._model.data_shapes
# Reshape flattened CSV as specified by the model
if content_type == content_types.CSV:
_, target_shape = data_shape
ndarray = ndarray.reshape(target_shape)
# Batch size is first dimension of model input
model_batch_size = data_shape[1][0]
pad_rows = max(0, model_batch_size - ndarray.shape[0])
# If ndarray has fewer rows than model_batch_size, then pad it with zeros.
if pad_rows:
num_pad_values = pad_rows
for dimension in ndarray.shape[1:]:
num_pad_values *= dimension
padding_shape = tuple([pad_rows] + list(ndarray.shape[1:]))
padding = mx.ndarray.zeros(shape=padding_shape)
ndarray = mx.ndarray.concat(ndarray, padding, dim=0)
model_input = mx.io.NDArrayIter(ndarray, batch_size=model_batch_size,
last_batch_handle='pad')
if pad_rows:
# Update the getpad method on the model_input data iterator to return the amount of
# padding. MXNet will ignore the last getpad() rows during Module predict.
def _getpad():
return pad_rows
model_input.getpad = _getpad
return model_input
def default_input_fn(self, input_data, content_type):
"""Take request data and deserialize it 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's content type, for example "application/json"
- The request data
The ``input_fn`` is responsible for preprocessing request data before prediction.
Args:
input_data (obj): the request data
content_type (str): the request's content type
Returns:
mxnet.nd.array: an MXNet NDArray
Raises:
sagemaker_containers.beta.framework.errors.UnsupportedFormatError: if an unsupported
content type is used.
"""
if content_type in self.VALID_CONTENT_TYPES:
np_array = encoders.decode(input_data, content_type)
return mx.nd.array(np_array).as_in_context(_context())
else:
raise errors.UnsupportedFormatError(content_type)
def default_output_fn(self, prediction, accept):
"""Serialize the prediction into a response.
Args:
prediction (mxnet.nd.array): an MXNet NDArray that is the result of a prediction
accept (str): the accept content type expected by the client
Returns:
sagemaker_containers.beta.framework.worker.Response: a Flask response object
Raises:
sagemaker_containers.beta.framework.errors.UnsupportedFormatError: if an unsupported
accept type is used.
"""
if accept in self.VALID_CONTENT_TYPES:
return worker.Response(encoders.encode(prediction.asnumpy().tolist(), accept), accept)
else:
raise errors.UnsupportedFormatError(accept)
