Databricks Machine Learning Associate Question 623
Single answerYou are deploying a machine learning model using Databricks Model Serving. The model is designed to predict customer churn based on input features such as customer activity and subscription details. During deployment, you want to ensure that the model can handle real-time predictions and that any updates to the model version can be managed with minimal downtime. Which approach should you take?
- A
Deploy the model as a batch job using Databricks Jobs and schedule regular execution.
- B
Serve the model using Databricks Model Serving with an endpoint and enable version management.
- C
Export the model to a third-party tool and deploy it there for real-time serving.
- D
Save the model as a file in the Databricks File Store and load it on demand for predictions.
Show answer and explanation
Correct answer: B
Explanation
Databricks Model Serving is the most suitable option for real-time predictions in Databricks, as it provides REST API endpoints, supports real-time model inference, and includes features for managing multiple model versions with minimal downtime. Other options either fail to meet the real-time requirement or introduce unnecessary complexity.
- A. Incorrect.
Batch jobs are not suitable for real-time predictions as they operate on pre-scheduled intervals, making them inappropriate for scenarios requiring immediate responses.
- B. Correct.
Databricks Model Serving is specifically designed for real-time predictions and supports endpoint creation along with version management, ensuring minimal downtime during model updates.
- C. Incorrect.
While exporting the model to a third-party tool can support real-time serving, it adds unnecessary complexity and is not a best practice when using Databricks' native features.
- D. Incorrect.
Saving the model as a file and loading it on demand is inefficient for real-time predictions due to latency and lack of scalability.