Databricks Machine Learning Professional Question 197
Select 3You are tasked with building a real-time recommendation system on Databricks that provides personalized product suggestions as users interact with your e-commerce platform. The system must process user actions (e.g., clicks and purchases) in real-time and update recommendations instantly. Which combination of tools and techniques would be most appropriate for implementing this solution in Databricks?
- A
Use Apache Kafka to stream user actions into Databricks and process the data using Structured Streaming.
- B
Train a batch machine learning model daily using historical user data and update recommendations once per day.
- C
Leverage Databricks Feature Store to store and retrieve real-time computed features for the recommendation model.
- D
Deploy the trained recommendation model as a REST API using Databricks Model Serving for real-time inference.
- E
Use Delta Live Tables (DLT) to preprocess streaming data and apply machine learning predictions on the fly.
Show answer and explanation
Correct answers: A, C, D
Explanation
To build a real-time recommendation system, you need tools and techniques that support real-time data ingestion, feature computation, and low-latency model inference. Apache Kafka and Structured Streaming handle real-time data processing, the Databricks Feature Store ensures efficient feature management, and Databricks Model Serving provides real-time predictions. These components together form a robust pipeline for real-time recommendations.
- A. Correct.
Using Apache Kafka to stream user actions into Databricks and processing the data using Structured Streaming is a common approach for building real-time systems. It allows you to handle incoming data in real-time and process it efficiently.
- B. Incorrect.
Training a batch machine learning model daily and updating recommendations once per day is not suitable for a real-time use case, as it introduces significant delays in updating recommendations.
- C. Correct.
Databricks Feature Store is specifically designed for storing and retrieving features in real-time, making it a critical component for real-time recommendation systems.
- D. Correct.
Deploying the trained recommendation model as a REST API using Databricks Model Serving enables low-latency predictions, which is essential for real-time use cases.
- E. Incorrect.
While Delta Live Tables (DLT) is useful for streamlining ETL pipelines, it is not specifically designed for real-time machine learning inference or feature engineering in real-time systems.