Update real-time webservice with the retrained endpoint or published pipeline

Question

I have already tried with the documentation how-to-deploy-update-web-service. I need a detailed explanation on how to update a web service with the experiment run or pipeline endpoint. There is no clear explanation in the link. I trained the model in the designer from scratch. I need to automate the whole flow of retraining the model and updating the web service. Any help is appreciated.

Please answer the following questions:

1. How do I give the model path for a trained model from the retrained experiment or published pipeline?
2. What is the tags field that I need to mention ?
3. what is the deploy environment I need to specify?
4. What to specify in the place of score.py?

ws = Workspace.get(name="xxx", subscription_id='xxx, resource_group='xxx')
new_model = Model.register(model_path="azureml\ce6116ec-cd48-4115-a567-e0eeed49f5b5\Trained_model",
model_name="Trained_model",
tags=tags)
deploy_env = Environment.get(workspace=ws, name="AzureML-PyTorch-1.4-GPU", version="3")
inference_config = InferenceConfig(entry_script="score.py",
environment=deploy_env)

Answer 1

Hi, you cannot use the SDK to update a web service published from the Azure Machine Learning designer. For designer, you need to go through the deploy process to deploy the endpoint. You can test the real-time endpoint directly in the portal to verify the output of your endpoint. To consume your model, you can use various clients for calling the web service. For a more automated workflow, please check out published pipelines for retraining models.