Databricks Machine Learning Associate Question 642
Select 2A machine learning engineer is tasked with deploying a real-time fraud detection model using Delta Live Tables (DLT). The engineer needs to implement streaming inference on incoming transaction data to identify potentially fraudulent transactions. Which configuration steps should the engineer take to ensure streaming inference is performed correctly with Delta Live Tables?
- A
Define the Delta Live Table as a streaming table by specifying the 'STREAMING' keyword in the table configuration.
- B
Use a pre-trained machine learning model and apply it within a Python function that is called on each batch of streaming data.
- C
Configure the Delta Live Table to use a 'trigger once' mode to process streaming data in real time.
- D
Set the input source of the Delta Live Table to a streaming source, such as Kafka or a Delta table with streaming enabled.
- E
Enable the 'AUTO OPTIMIZE' feature in the Delta Live Table to automatically optimize the streaming inference process.
Show answer and explanation
Correct answers: B, D
Explanation
Streaming inference with Delta Live Tables involves setting up a streaming source (e.g., Kafka or a Delta table with streaming enabled) and applying a pre-trained machine learning model to incoming data. This configuration ensures that real-time data is processed continuously and predictions are made on the fly. Specifying the correct input source and leveraging Python functions for model inference are essential for achieving this functionality.
- A. Incorrect.
Delta Live Tables does not use a 'STREAMING' keyword to define streaming tables. Instead, streaming behavior is determined by the nature of the input source.
- B. Correct.
To perform streaming inference, a pre-trained machine learning model can be applied to each batch of streaming data using a Python function. This is a common approach for integrating ML models into Delta Live Table pipelines.
- C. Incorrect.
The 'trigger once' mode is not suitable for continuous streaming inference as it processes the data only once. Delta Live Tables should be configured to continuously process data for streaming scenarios.
- D. Correct.
Streaming inference requires the input source to be a streaming source, such as Kafka or a Delta table configured for streaming. This ensures that data is processed in real time.
- E. Incorrect.
The 'AUTO OPTIMIZE' feature in Delta Live Tables is used for optimizing data layout and performance but does not directly impact the streaming inference process.