Creating an automated ML endpoint service, calling with an error

Question

Follow the "Tutorial: Training a Classification Model in Azure Machine Learning Studio with No-Code AutoML" to create the endpoint service, sample Python code call, and error message.

{"message": "An unexpected error occurred in scoring script. Check the logs for more info."}，

Endpoint log error message

{

"error": {

    "code": "UserError",

    "message": "Expected column(s) 0 not found in fitted data.",

    "target": "X",

    "inner_error": {

        "code": "BadArgument",

        "inner_error": {

            "code": "MissingColumnsInData"

        }

    },

    "reference_code": "17049f70-3bbe-4060-a63f-f06590e784e5"

}

}

The input data used.

data = {

"Inputs": {

    #"columns": ["age", "job", "marital", "education", "default", "housing", "loan", "contact", "month", "duration", "campaign",

     #           "pdays", "previous", "poutcome", "emp.var.rate", "cons.price.idx", "cons.conf.idx", "euribor3m", "nr.employed"],

    "columns": [0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16,17,18],

    "index": [0, 1],

    "data": [

        [57, "technician", "married", "high.school", "no", "no", "yes", "cellular", "may", 371, 1, 999, 1, "failure", -1.8, 92.893, -46.2, 1.299, 5099.1],

        [30, "blue-collar", "single", "basic.9y", "no", "yes", "no", "cellular", "jul", 221, 1, 999, 0, "nonexistent", 1.4, 93.994, -36.4, 4.857, 5191]

    ]

}

}

Please guide me how to solve it.

Answer 1

@Xu, Pengcheng (CN - AB) I think the request data might not be correctly formatted in this case. I do not have this setup to test but the same example with the same dataset is available to test in azureml-examples repo, Please check this notebook where the section "Test the deployment" uses the following format for request instead of the one mentioned in your request.

test_data = pd.read_csv("./data/test-mltable-folder/bank_marketing_test_data.csv")

test_data = test_data.drop("y", axis=1)

test_data_json = test_data.to_json(orient="records", indent=4)
data = (
    '{ \
          "input_data": {"data": '
    + test_data_json
    + "}}"
)

request_file_name = "sample-request-bankmarketing.json"

with open(request_file_name, "w") as request_file:
    request_file.write(data)

ml_client.online_endpoints.invoke(
    endpoint_name=online_endpoint_name,
    deployment_name="bankmarketing-deploy",
    request_file=request_file_name,
)

If you can use the sample file and print the request data, you can find the correct format that you can use with your deployment. Thanks!!

Answer 2

I was getting the same error message for a classification model trained with iris data.

Adding column names to data solved the issue.

COLUMN_NAMES = ["sepal_length", "sepal_width", "petal_length", "petal_width"]

@app.route('/predict', methods=['POST'])
def predict():

    global model

    try:
        raw_data = request.data
        data = json.loads(raw_data)["data"]

        data_df = pd.DataFrame(data, columns=COLUMN_NAMES)    # Add column names here !


        result = model.predict(data_df)

        return jsonify({"predictions": result.tolist()})

    except Exception as e:
        logging.error(f"Error processing request: {str(e)}")
        return jsonify({"error": str(e)}), 400

Answer 3

I was getting the same error.

Adding column names for the dataset solved my case.

# Define column names that match the training data
COLUMN_NAMES = ["sepal_length", "sepal_width", "petal_length", "petal_width"]
@app.route('/predict', methods=['POST'])
def predict():
    global model
    if model is None:
        return jsonify({"error": "Model is not initialized."}), 500
    try:
        raw_data = request.data
        data = json.loads(raw_data)["data"]
        data_df = pd.DataFrame(data, columns=COLUMN_NAMES)     # Add Columns Here 
        result = model.predict(data_df)
        return jsonify({"predictions": result.tolist()})
    except Exception as e:
        return jsonify({"error": str(e)}), 400

Answer 4

First thing I think about is to verify the column names and their order in the input data match exactly with what was used during training because what I understood from the error message is that the service is expecting column names instead of numerical indices.

To explain better, the input data provided for scoring should have the same structure (column names and data types) as the data used for training the model.

So based on the example provided, you should replace the numerical indices with the corresponding column names.

data = {

    "Inputs": {

        "columns": [

            "age", "job", "marital", "education", "default", "housing", "loan", "contact", "month", "duration", "campaign",

            "pdays", "previous", "poutcome", "emp.var.rate", "cons.price.idx", "cons.conf.idx", "euribor3m", "nr.employed"

        ],

        "index": [0, 1],

        "data": [

            [57, "technician", "married", "high.school", "no", "no", "yes", "cellular", "may", 371, 1, 999, 1, "failure", -1.8, 92.893, -46.2, 1.299, 5099.1],

            [30, "blue-collar", "single", "basic.9y", "no", "yes", "no", "cellular", "jul", 221, 1, 999, 0, "nonexistent", 1.4, 93.994, -36.4, 4.857, 5191]

        ]

    }

}