Section 1
Data Prep & Feature Engineering
Preview
Preview: Data Preparation and Feature Engineering
- 4 of 24 lesson topics
- 1 overview segment
- 3 core concepts
- 2 exam tips
Lesson Topics
- Feature Store Online Store
- Feature Store Offline Store
- Reusable Feature Definitions
- Training-Serving Feature Consistency
Overview
Data preparation and feature engineering are critical steps in building machine learning models. The exam tests how SageMaker Feature Store streamlines these processes, ensuring consistency and efficiency across training and inference.
Core Concepts
- **Online Store:** Provides managed low-latency feature lookup for real-time inference. Feature groups can use supported online storage types, including the standard managed tier and supported in-memory options; throughput, latency, quotas, and cost depend on storage type, throughput configuration, record size, API usage, and workload.
- **Offline Store:** Used for durable storage of historical feature data for training, batch analysis, and point-in-time queries. Offline data is stored in the configured S3 location using the supported Feature Store layout and table format, and can be analyzed through Athena or other supported processing paths.
- **Reusable Feature Groups:** Features and metadata are organized into feature groups that can be discovered, shared, governed, and reused according to permissions and supported cross-account or organizational configurations. Reuse still requires compatible schemas, definitions, processing logic, freshness, and governance.
Exam Tips
- Prioritize understanding the core purpose of each Feature Store component (online vs. offline) before considering specific configurations.
- Always consider training-serving skew and whether the scenario includes shared processing logic, compatible feature groups, freshness, and point-in-time-correct retrieval.