def test_split_oriented_json_to_numpy_array():
# test that datatype for "zip" column is not converted to "int64"
jstr = '{"columns":["zip","cost","count"],"index":[0,1,2],' \
'"data":[["95120",10.45,-8],["95128",23.0,-1],["95128",12.1,1000]]}'
df = pyfunc_scoring_server.parse_split_oriented_json_input_to_numpy(jstr)
assert set(df.columns) == {'zip', 'cost', 'count'}
assert set(str(dt) for dt in df.dtypes) == {'object', 'float64', 'int64'}
async def invocations(self, request: Request) -> Response:
"""
This custom handler is meant to mimic the behaviour of the existing
scoring server in MLflow.
For details about its implementation, please consult the original
implementation in the MLflow repository:
https://github.com/mlflow/mlflow/blob/master/mlflow/pyfunc/scoring_server/__init__.py
"""
content_type = request.headers.get("content-type", None)
raw_data = await request.body()
as_str = raw_data.decode("utf-8")
if content_type == CONTENT_TYPE_CSV:
csv_input = StringIO(as_str)
data = parse_csv_input(csv_input=csv_input)
elif content_type == CONTENT_TYPE_JSON:
data = infer_and_parse_json_input(as_str, self._input_schema)
elif content_type == CONTENT_TYPE_JSON_SPLIT_ORIENTED:
data = parse_json_input(
json_input=StringIO(as_str),
orient="split",
schema=self._input_schema,
)
elif content_type == CONTENT_TYPE_JSON_RECORDS_ORIENTED:
data = parse_json_input(
json_input=StringIO(as_str),
orient="records",
schema=self._input_schema,
)
elif content_type == CONTENT_TYPE_JSON_SPLIT_NUMPY:
data = parse_split_oriented_json_input_to_numpy(as_str)
else:
content_type_error_message = (
"This predictor only supports the following content types, "
f"{CONTENT_TYPES}. Got '{content_type}'.")
raise InferenceError(content_type_error_message)
try:
raw_predictions = self._model.predict(data)
except MlflowException as e:
raise InferenceError(e.message)
except Exception:
error_message = (
"Encountered an unexpected error while evaluating the model. Verify"
" that the serialized input Dataframe is compatible with the model for"
" inference.")
raise InferenceError(error_message)
result = StringIO()
predictions_to_json(raw_predictions, result)
return Response(content=result.getvalue(),
media_type="application/json")
def test_parse_json_input_split_oriented_to_numpy_array():
size = 200
data = OrderedDict([("col_m", [random_int(0, 1000) for _ in range(size)]),
("col_z", [random_str(4) for _ in range(size)]),
("col_a", [random_int() for _ in range(size)])])
p0 = pd.DataFrame.from_dict(data)
np_array = np.array([[a, b, c] for a, b, c in
zip(data['col_m'], data['col_z'], data['col_a'])],
dtype=object)
p1 = pd.DataFrame(np_array).infer_objects()
p2 = pyfunc_scoring_server.parse_split_oriented_json_input_to_numpy(
p0.to_json(orient="split"))
np.testing.assert_array_equal(p1, p2)
