1  import pandas as pd
 2  import requests
 3  from sktime.datasets import load_longley
 4  from sktime.forecasting.model_selection import temporal_train_test_split
 5  
 6  y, X = load_longley()
 7  y_train, y_test, X_train, X_test = temporal_train_test_split(y, X)
 8  
 9  # Define local host and endpoint url
10  host = "127.0.0.1"
11  url = f"http://{host}:5000/invocations"
12  
13  # Model scoring via REST API requires transforming the configuration DataFrame
14  # into JSON format. As numpy ndarray type is not JSON serializable we need to
15  # convert the exogenous regressor into a list. The wrapper instance will convert
16  # the list back to ndarray type as required by sktime predict methods. For more
17  # details read the MLflow deployment API reference.
18  # (https://mlflow.org/docs/latest/models.html#deploy-mlflow-models)
19  X_test_list = X_test.to_numpy().tolist()
20  predict_conf = pd.DataFrame([
21      {
22          "fh": [1, 2, 3, 4],
23          "predict_method": "predict_interval",
24          "coverage": [0.9, 0.95],
25          "X": X_test_list,
26      }
27  ])
28  
29  # Create dictionary with pandas DataFrame in the split orientation
30  json_data = {"dataframe_split": predict_conf.to_dict(orient="split")}
31  
32  # Score model
33  response = requests.post(url, json=json_data)
34  print(f"\nPyfunc 'predict_interval':\n${response.json()}")