Object of type ndarray is not JSON serializable: API endpoint-CodePudding

I have a Python script which takes a list as input schema and return a record as output schema like this-

# Initialize the deployment environment
def init():
    global regressor
    model_path = os.path.join(os.getenv('AZUREML_MODEL_DIR'), 'rul.pickle')
    regressor = joblib.load(model_path)
    

standard_sample_input = StandardPythonParameterType([19.0,168.88745536965698,448.15396972858315,97.63243953079672,39.92323264491041,65.49201623238903,
  101.29277662751305,23.70447278514443,1.7526720815415802e-12,169.15575133063476,415.03985455721136,98.13586382473882])
                                        
sample_input = StandardPythonParameterType({'input1': standard_sample_input})                                        

sample_output = StandardPythonParameterType([16.62585])
outputs = StandardPythonParameterType({'Results':sample_output})                                        

@input_schema('Inputs', sample_input)
@output_schema(outputs)

# Run the model and return the scored result.    
def run(Inputs):
    response = ''
    try:
        #input_df = np.array(json.loads(input_df)['input_df'])
        #prediction = regressor.predict(input_df)    
        col = ['machineID', 'Vibration_ArmsRM12', 'Vibration_DKWRM12', 'Vibration_PeakRM12', 'Vibration_VrmsRM12', 'Performance_CurrentRM12', 'Performance_PowerRM12', 'Performace_SpeedRM12', 'Stat_EnergyRM12', 'Vibration_ArmsRM24', 'Vibration_DKWRM24', 'Vibration_PeakRM24', 'Vibration_VrmsRM24', 'Performance_CurrentRM24', 'Performance_PowerRM24']
        Inputs  = json.dumps(Inputs)
        Inputs = ast.literal_eval(Inputs)
        Inputs = Inputs.values()
        Inputs = pd.DataFrame(Inputs, columns=col)
        prediction = regressor.predict(Inputs)
        
        print(prediction)
    except Exception as e:
        print("Error: {0}",str(e))
        return (str(e))

    print(prediction)
    return prediction

Now I am deploying this script and getting a REST Endpoint. But when I am calling this endpoint with the input schema type, I am getting an error that TypeError: Object of type ndarray is not JSON serializable.

What I fail to understand is that there's no presence of numpy or numpy array anywhere in my code. It is just a list I am passing. I am just passing a JSON as input and converting that JSON into dictionary, picking up only the values and converting it into a dataframe and passing it to the model to predict.

This is the input I am passing to test the endpoint.

{"Inputs":{"input1":[18.0, 170.54412436868537, 457.122701576177, 99.03802220789595, 40.94298920765353, 0.0, 0.0, 4.585221091701897e-13, 434.66205951841266, 168.7749263236577, 454.938029602499, 99.54961810956478, 40.798219749135974]}}

CodePudding user response：

The values returned from the call to regressor.predict(Inputs) & assigned to the prediction variable are likely to be a numpy.ndarray.

When you return this, your endpoint will attempt to serialize the array, leading to the reported error.

The array can be converted to a list (which is serialisable) by calling .tolist(). ie.

return prediction.tolist()