Databricks Machine Learning Associate Question 654
Select 3You are tasked with deploying a machine learning model for real-time inference on Databricks. The incoming data needs to be split between multiple model endpoints based on certain conditions, such as user type or region. Which of the following approaches can be used to implement this data splitting in Databricks?
- A
Use a UDF (User-Defined Function) in a streaming notebook to route data to specific model endpoints.
- B
Leverage Databricks Feature Store to automatically split and route data between endpoints.
- C
Utilize a Delta Live Table pipeline to pre-process the data and direct it to different endpoints based on conditions.
- D
Implement a REST API that receives incoming data, applies routing logic, and forwards the data to the appropriate model endpoint.
- E
Configure AutoML in Databricks to handle real-time data routing between endpoints.
Show answer and explanation
Correct answers: A, C, D
Explanation
In real-time inference scenarios, splitting data between endpoints often requires custom logic based on the specific use case. Using a UDF in a streaming notebook, a Delta Live Table pipeline, or a REST API are valid approaches to handle this functionality in Databricks. However, tools like the Feature Store and AutoML are designed for other purposes, such as feature management and model training, and do not support routing of data between endpoints.
- A. Correct.
This is a valid approach as UDFs can be used to apply custom logic in a streaming environment to determine the routing of data to specific endpoints.
- B. Incorrect.
The Databricks Feature Store is used for managing features for machine learning models and does not natively provide functionality for routing data between endpoints.
- C. Correct.
Delta Live Tables can be used to create data pipelines that include pre-processing and conditional logic, making it possible to route data to different endpoints.
- D. Correct.
A REST API is a practical solution for handling real-time data routing logic before forwarding data to the appropriate endpoint.
- E. Incorrect.
AutoML in Databricks is used for automating the training and optimization of models, but it does not handle real-time data routing or endpoint management.