def test_get_jsonnable_obj():
from mlflow.pyfunc.scoring_server import _get_jsonable_obj
py_ary = [["a", "b", "c"], ["e", "f", "g"]]
np_ary = _get_jsonable_obj(np.array(py_ary))
assert json.dumps(py_ary, cls=NumpyEncoder) == json.dumps(np_ary, cls=NumpyEncoder)
np_ary = _get_jsonable_obj(np.array(py_ary, dtype=type(str)))
assert json.dumps(py_ary, cls=NumpyEncoder) == json.dumps(np_ary, cls=NumpyEncoder)
def predict(self, deployment_name, df):
"""
Predict on the specified deployment using the provided dataframe.
Compute predictions on the pandas DataFrame ``df`` using the specified deployment.
Note that the input/output types of this method matches that of `mlflow pyfunc predict`
(we accept a pandas DataFrame as input and return either a pandas DataFrame,
pandas Series, or numpy array as output).
:param deployment_name: Name of deployment to predict against
:param df: Pandas DataFrame to use for inference
:return: A pandas DataFrame, pandas Series, or numpy array
"""
try:
service = Webservice(self.workspace, deployment_name)
except Exception as e:
raise MlflowException(
'Failure retrieving deployment to predict against') from e
# Take in DF, parse to json using split orient
input_data = _get_jsonable_obj(df, pandas_orient='split')
if not service.scoring_uri:
raise MlflowException(
'Error attempting to call webservice, scoring_uri unavailable. '
'This could be due to a failed deployment, or the service is not ready yet.\n'
'Current State: {}\n'
'Errors: {}'.format(service.state, service.error))
# Pass split orient json to webservice
# Take records orient json from webservice
resp = ClientBase._execute_func(service._webservice_session.post,
service.scoring_uri,
data=json.dumps(
{'input_data': input_data}))
if resp.status_code == 401:
if service.auth_enabled:
service_keys = service.get_keys()
service._session.headers.update(
{'Authorization': 'Bearer ' + service_keys[0]})
elif service.token_auth_enabled:
service_token, refresh_token_time = service.get_access_token()
service._refresh_token_time = refresh_token_time
service._session.headers.update(
{'Authorization': 'Bearer ' + service_token})
resp = ClientBase._execute_func(service._webservice_session.post,
service.scoring_uri,
data=input_data)
if resp.status_code == 200:
# Parse records orient json to df
return parse_json_input(json.dumps(resp.json()), orient='records')
else:
raise MlflowException('Failure during prediction:\n'
'Response Code: {}\n'
'Headers: {}\n'
'Content: {}'.format(resp.status_code,
resp.headers,
resp.content))
def invoke(self, data):
"""
Invoke inference on input data. The input data must be pandas dataframe or json instance.
"""
content_type = scoring_server.CONTENT_TYPE_JSON
post_data = json.dumps(
scoring_server._get_jsonable_obj(data, pandas_orient="split"))
response = requests.post(
url=self.url_prefix + "/invocations",
data=post_data,
headers={"Content-Type": content_type},
)
if response.status_code != 200:
raise Exception(
f"Invocation failed (error code {response.status_code}, response: {response.text})"
)
return scoring_server.infer_and_parse_json_input(response.text)
def predictions_to_json(raw_predictions, output):
predictions = _get_jsonable_obj(raw_predictions, pandas_orient="records")
json.dump(predictions, output, cls=NumpyEncoder)
