Databricks Machine Learning Associate Question 655
Select 3You are designing a real-time inference system using Databricks and need to split incoming data between two machine learning endpoints based on specific conditions. Which of the following approaches can help you achieve this split effectively?
- A
Use a feature-based routing mechanism to direct data to the appropriate endpoint based on specific feature values.
- B
Deploy a single endpoint and train a multi-task model capable of handling multiple scenarios.
- C
Leverage a streaming job to preprocess incoming data and route it to the appropriate endpoint based on predefined logic.
- D
Split the data randomly between the endpoints to ensure balanced loads.
- E
Configure a REST API gateway to handle the routing logic before passing data to the endpoints.
Show answer and explanation
Correct answers: A, C, E
Explanation
In real-time inference systems, data splitting between endpoints is typically based on specific conditions or features to ensure that the correct model processes the data. Feature-based routing and streaming jobs in Databricks allow dynamic and logical routing of data in real time. Additionally, REST API gateways provide an external mechanism for routing logic before the data reaches the endpoints. Random splitting is not suitable for condition-based routing, and multi-task models do not involve splitting between endpoints.
- A. Correct.
This is correct because feature-based routing is a common and efficient way to direct data to specific endpoints when certain feature values dictate the appropriate model to use.
- B. Incorrect.
This is incorrect because while a multi-task model could handle multiple scenarios, it does not involve splitting data between separate endpoints. This solution combines tasks rather than routing to different endpoints.
- C. Correct.
This is correct because a streaming job in Databricks can be used to apply logic for routing data in real time, enabling dynamic preprocessing and endpoint selection.
- D. Incorrect.
This is incorrect because random splitting does not consider the specific conditions that determine which endpoint should process the data. It can lead to suboptimal inference results.
- E. Correct.
This is correct because a REST API gateway can be configured to handle routing logic, which is a common pattern for real-time inference systems.