You have finished with training your ML model: it is optimal, hyperparameters are tuned, everything is fine.
You are about to deploy it as a real-time inferencing service to an Azure Kubernetes cluster.
In order to have your service in production, there is a list of activities you have to execute.
Connect to your workspace Register the model Prepare inference configuration Prepare entry script Choose a compute target Define a deployment configuration Deploy the model Which of the above activities are recommended but not required?
Click on the arrows to vote for the correct answer
A. B. C. D.Answer: A.
Option A is CORRECT because in case you proceed without registering a model, you need to manually specify a source directory in your InferenceConfig and ensure that model is in that directory.
Model registration can make model management easier.
Therefore, registering a model is recommended but not required.
Option B is incorrect because although registering the model is only recommended, an entry script must be prepared because it receives the data submitted to a deployed web service and passes it to the model, it is the interface between the client and your service.
Option C is incorrect because both of these activities are required for a successful model deployment.
Inference configuration describes how to set up the web-service, and a compute target to host the service must be chosen, too.
Option D is incorrect because an entry script is a mandatory component of the model deployment.An entry script must be prepared because it receives the data submitted to a deployed web service and passes it to the model, it is the interface between the client and your service.
Reference:
In order to deploy an ML model as a real-time inferencing service to an Azure Kubernetes cluster, there are several activities that need to be executed. These activities are as follows:
Connect to your workspace: This activity involves connecting to the Azure Machine Learning workspace where the trained model is stored. This step is necessary to access the model and perform further actions.
Register the model: This activity involves registering the trained model in the Azure Machine Learning workspace. This step is necessary to track the versioning of the model, as well as to use it in the deployment process.
Prepare inference configuration: This activity involves creating an inference configuration file that specifies the environment and dependencies required for running the model. This step is necessary to ensure that the model runs correctly in the deployed environment.
Prepare entry script: This activity involves creating an entry script that specifies how the model should be loaded and used for making predictions. This step is necessary to ensure that the model runs correctly in the deployed environment.
Choose a compute target: This activity involves choosing a compute target, which is the environment where the model will be deployed and served as an inferencing service. This step is necessary to ensure that the deployed model has the required resources to run smoothly.
Define a deployment configuration: This activity involves defining a deployment configuration file that specifies the settings for deploying the model to the chosen compute target. This step is necessary to ensure that the deployment process runs smoothly and the model is deployed correctly.
Deploy the model: This activity involves deploying the model as a real-time inferencing service to the Azure Kubernetes cluster. This step is necessary to make the model available for use in production environments.
Out of these activities, some are recommended but not required. Let's take a closer look at the answer options:
A. Register the model: Registering the model is highly recommended as it provides version control and tracking capabilities, but it is not strictly required for deployment.
B. Prepare entry script; Register the model: Both of these activities are highly recommended, but not strictly required for deployment.
C. Prepare inference configuration; Choose a compute target: Preparing the inference configuration is highly recommended, as it ensures that the model runs correctly in the deployed environment. Choosing a compute target is also highly recommended, as it ensures that the deployed model has the required resources to run smoothly. Both of these activities are not strictly required for deployment, but they are highly recommended.
D. Prepare entry script: This activity is highly recommended, but not strictly required for deployment.
In summary, all of the listed activities are recommended for deployment of an ML model as a real-time inferencing service to an Azure Kubernetes cluster. However, some activities are more important than others, and some may be optional depending on the specific deployment scenario.