Databricks Machine Learning Associate Question 449
Select 2A team is building a recommendation system for an e-commerce platform where product suggestions need to be updated in real-time as users interact with the platform. They also need historical data to retrain the model periodically. What are the key differences between online and offline feature tables that the team should consider when designing their feature store?
- A
Online feature tables are optimized for low-latency lookups to support real-time inference, while offline feature tables are optimized for batch processing and model training.
- B
Offline feature tables are typically stored in high-performance databases designed for real-time lookups, whereas online feature tables are stored in data lakes for large-scale querying.
- C
Online feature tables usually store the most recent feature values, while offline feature tables store historical feature values to support model training.
- D
Offline feature tables require stringent low-latency constraints to serve features in real-time, whereas online feature tables do not prioritize latency.
- E
Online feature tables are commonly used to serve features for batch inference, while offline feature tables are used for real-time inference.
Show answer and explanation
Correct answers: A, C
Explanation
The primary difference between online and offline feature tables lies in their purpose and optimization. Online feature tables are designed to support real-time inference by providing low-latency access to the most recent feature values. In contrast, offline feature tables are optimized for batch processing and store historical data to support model training and large-scale analysis. Understanding these differences helps teams design efficient feature stores tailored to their machine learning workflows.
- A. Correct.
Correct. Online feature tables are designed for low-latency access to ensure real-time inference, whereas offline feature tables are optimized for batch processing and are typically used for model training.
- B. Incorrect.
Incorrect. This is the opposite of the typical setup. Online feature tables are stored in high-performance databases for real-time lookups, while offline feature tables are often stored in data lakes to facilitate batch querying and training.
- C. Correct.
Correct. Online feature tables prioritize storing the most recent feature values for real-time use, while offline feature tables store historical values to enable retraining and analysis.
- D. Incorrect.
Incorrect. Offline feature tables are optimized for batch processing and do not have low-latency requirements. Online feature tables are the ones that prioritize low-latency lookups for real-time use cases.
- E. Incorrect.
Incorrect. Online feature tables are used for real-time inference, while offline feature tables are used for batch inference and training purposes.