Service End point Failure in Azure ML through SDK

Question

I have been running a simple Linear regression model to predict prices and publishing its endpoint on the azure portal in free subscription trial.

I am facing problem in score.py program where creating the defination of init() and run()

The error that I am getting is 'list' object has no attribute 'predict' when trying to predict new observations on the published endpoint.

Please find the code of Deploy configuration

from azureml.core import Workspace, Experiment
print("Accessing the workspace from job....")
ws=Workspace.from_config()


# Get the input dataset
print("Accessing the Adult Income dataset...")
input_ds = ws.datasets.get('pricing')


# -------------------------------------------------
# Create custom environment
# -------------------------------------------------
from azureml.core import Environment
from azureml.core.environment import CondaDependencies

# Create the environment
myenv = Environment(name="MyEnvironment")

# Create the dependencies object
myenv_dep = CondaDependencies.create(conda_packages=['scikit-learn', 'pip','pandas'],
                                     pip_packages=['azureml-defaults', 'azureml-interpret'])

myenv.python.conda_dependencies = myenv_dep

# Register the environment
print("Registering the environment...")
myenv.register(ws)

# Creat an Azure Kubernets Service provisioning Configuration

from azureml.core.compute import AksCompute, ComputeTarget

cluster_name='aks-cluster-12'

if cluster_name not in ws.compute_targets:
    print(cluster_name,"does not exist.Creating a new one")
    print('Creating provisiong config for Aks cluster')
    

    aks_config=AksCompute.provisioning_configuration(location='centralindia',
                                                 vm_size='Standard_DS11_v2',
                                                 agent_count=1,
                                                 cluster_purpose='DevTest')
    print("Creating the AKS cluster")



    production_cluster=ComputeTarget.create(ws,cluster_name,aks_config)
    
    production_cluster.wait_for_completion(show_output=True)
else:
    print(cluster_name,"exists. using it..")
    production_cluster=ws.compute_targets[cluster_name]

# Creat the inference Configuration

from azureml.core.model import InferenceConfig

inference_config= InferenceConfig(environment=myenv ,entry_script='scoringscriptnew.py',
                                 source_directory=".")

from azureml.core.webservice import AksWebservice 

deploy_config=AksWebservice.deploy_configuration(cpu_cores=1,
                                                 memory_gb=0.5)

# Deploy 

from azureml.core.model import Model
model=ws.models['regression01']

service= Model.deploy(workspace=ws,name='regression',models=[model],
                      inference_config=inference_config,
                      deployment_config=deploy_config,
                      deployment_target=production_cluster)

service.wait_for_deployment(show_output=True)

And code of score.py

import json
import joblib
from azureml.core.model import Model
import pandas as pd

# Called when the service is loaded

def init():
    global predictor
    
    # Get the path to the registered model file and load it
    model_path = Model.get_model_path('regression01')
    predictor = joblib.load(model_path)


# Called when a request is received
def run(raw_data):
    # Get the input data as a dictionary
    data_dict = json.loads(raw_data)['data']
    
    # Convert dictionary to pandas dataframe
    data = pd.DataFrame.from_dict(data_dict)
    
    # Transform the data
    # data = one_hot.transform(data)
    

    # difference of train and deploy

    # Get a prediction from the model
    predictions = predictor.predict(data)

    # Return the predictions
    return predictions

Can someone help me understand if the init() function is not performing and if not what could be the reason?

TadepalliSairam · Answer

The method which was mentioned in score.py is correct and the problem was with the way the model was called. In the model there is no value mentioned predict. Predict was missing and not mentioned in the model calling. Replace the code mentioned below with the existing code block.

output = model[0].predict(df)  this is the sample syntax

In the above syntax we need to mention the model as the list, and we need to assign that as the value.

Replace the existing model code with the code below.

from azureml.core.model import Model
df = pd.DataFrame('regression01', columns=features_name)
output = model[0].predict(df)

replace the above code in the place of below code.

model=ws.models['regression01']

Service End point Failure in Azure ML through SDK

1 Answers