Online vs offline feature stores in MLOps - Practice Questions

Challenge - 5 Problems

🧠 Conceptual

intermediate

Primary purpose of an online feature store

What is the main purpose of an online feature store in a machine learning system?

A To store historical feature data for batch training jobs

B To provide real-time feature data for low-latency model predictions

C To archive raw data for long-term storage

D To manage model versioning and deployment

💡 Hint

Think about where features are needed instantly during prediction.

Online feature stores serve features with very low latency to support real-time model inference.

🧠 Conceptual

intermediate

Key difference between online and offline feature stores

Which statement best describes the key difference between online and offline feature stores?

A Online stores are used only for data backup; offline stores are for real-time serving

B Offline stores require low latency; online stores tolerate high latency

C Offline stores provide features for training; online stores provide features for serving

D Online stores handle batch data; offline stores handle streaming data

💡 Hint

Consider the use case of training versus prediction.

Offline feature stores provide historical data for model training, while online stores provide fresh features for live predictions.

💻 Command Output

advanced

Latency difference in feature retrieval

Given a feature store setup, which option best describes the expected latency difference when retrieving features from online vs offline stores?

A Online store retrieval latency is in milliseconds; offline store retrieval latency is in minutes or hours

B Offline store retrieval latency is in milliseconds; online store retrieval latency is in minutes or hours

C Both online and offline stores have similar latency in seconds

D Online store retrieval latency is in hours; offline store retrieval latency is in milliseconds

💡 Hint

Think about real-time vs batch processing speeds.

Online feature stores are designed for very low latency (milliseconds) to support real-time predictions, while offline stores handle large batch data with higher latency.

✅ Best Practice

advanced

Best practice for feature consistency

What is a best practice to ensure feature consistency between online and offline feature stores?

A Use the same feature computation logic and data sources for both stores

B Update online store features weekly to match offline store

C Only use offline store features for both training and serving

D Compute features independently in online and offline stores to optimize speed

💡 Hint

Consistency means features should be the same regardless of source.

Using the same logic and data sources ensures that features used during training and serving match, preventing prediction errors.

❓ Troubleshoot

expert

Troubleshooting stale features in online store

If a model serving system returns poor predictions due to stale features, which is the most likely cause related to the online feature store?

A The feature computation logic differs between training and serving

B The offline feature store has missing historical data

C The model was trained on outdated offline features

D The online feature store is not updating features frequently enough

💡 Hint

Focus on the freshness of features during serving.

Stale features in the online store mean the data served to the model is outdated, causing poor predictions.